University of Southern California Dissertations and Theses

Number marking and definiteness in Bangla

(USC Thesis Other)

Number marking and definiteness in Bangla

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content i NUMBER MARKING AND DEFINITENESS IN BANGLA by Priyanka Biswas A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (LINGUISTICS) December 2016 Copyright 2016 Priyanka Biswas ii ACKNOWLEDGEMENTS No earthly parameter could measure how much I owe to Roumi Pancheva, my Doktormutter, for being an extraordinary advisor. At the risk of sounding cliché, I must say that she has been a true friend, philosopher and guide. Roumi, both in great depth and breadth, has shaped my thought process. I am grateful for those numerous meetings that transformed my ramblings (about work and life) into transparent concepts. She taught me to go beyond the limits of my comfort zone, and always encouraged me to connect the questions to the big picture while seeking the truth that lies within. I take this opportunity to express my limitless gratitude for the most outstanding human being I have ever known in my life, and I will cherish this opportunity throughout my life. I am extremely grateful to my wonderful committee members for their support and advice. Working with them has been my utmost privilege, and I have learnt many importatnt traits that will be my assets for life. Andrew Simpson amazed me with his abilities to construct focused, structured syntactic hypothesis from a quagmire of data. Elsi Kaiser taught me how being optimistic improves everything. I have always come out of a meeting with her feeling hopeful and loaded with newer directions to explore. Audrey Li has been enthusiastic about my work and provided many constructive ideas. Thanks to Gabriel Uzquiano-Cruz for elaborating on the big questions. Among other faculty members of the USC Linguistics department, I must thank Barry Schein for many interesting ideas that came up during the meetings. TA work with Khalil Iskarous and Hajime Hoji has been particularly intriguing. I learned the nitty-gritties of effective teaching while working with them. Thanks are due to many past and present faculty members - Maria Luisa Zubizarreta, Rachel Walker, Louis Goldstein, late Jean Roger Vergnaud, and Hagit Borer for their support in many ways. Their presence made my graduate school experience truly worthwhile and intellectually stimulating. My mentors in Delhi University, Tanmoy Bhattacharya and Tista Bagchi, have been pivotal at the beginning of this journey. Tanmoy encouraged me to go beyond what I had ever dreamt of. I thank Probal Dasgupta for intermittently updating me with interesting data and new areas of research. I am grateful to Veneeta Dayal for her input on many parts of my work. I am thankful to Rajesh Bhatt, Ashwini Deo, Hilda Koopman, Tim Stowell, Dominique Sportiche, and Anoop Mahajan, whom I met in many conferences and shared the love of linguistics. I am iii grateful to Joyce Perez, Brandon Washington and Guillermo Ruiz for their administrative support. I thank USC graduate school for the prestigious Morkovin Research Fellowship. I feel truly honored and grateful to have a number of great friends who form a big part of my life. Christina Hagedorn has been my ‘little Buddha’ to show me lights. Ellen O’Connor has been a role model to me by bringing perfection to a new level. Girls’ days out with them have been rejuvenating from time to time. Barbara Tomaszewicz has been a confidante and an epitome of hard work. Iris Ouyang has never ceased to amaze me with her depth of knowledge and diligence. Arunima Choudhury goes back a long time and our friendship has strengthened over the years. Binh Ngo, with her ever-cheerful persona has grown close in recent times. My friendship with Mythili Menon and Syed Saurov has evolved over the years in India and here. I thank Sarah Ouwayda, Ulli Steindl and Thomas Borer their congeniality. Victor Barrés has been the go-to friend for any help in computational linguistics in the recent days, and a colleague in the adventure of parenting. I thank George Toney for being a caring big brother and a wonderful trainer, and for accommodating me to the Sheepdogs. Ishani and Anumitra have been dear friends for a long time, and so have been Sonali and Arpita. I thank them for always being there at the horizon. Thanks to Arunima-Shubhendu, Sohini-Shubhomoy, Stephanie-Diego for good times in LA. Thanks to the ever-welcoming baristas at National-Sepulveda Starbucks who kept my dissertation going over never-ending supply of coffee. Apologies to those whom I have forgotten to acknowledge, it has been a long journey and definitely many people have inspired me over the years. My parents, Prabir Kumar and Anita Biswas, have provided me with unconditional support these many years of being a student. Their pride in me kept my chin up. My mother, particularly, for her unconditional support throughout my life. Their contribution to my success is undeniable. My parents-in-law and the extended family trusted me and provided the much needed support of different kinds. A parallel world that has been my life-source throughout these years, which has kept me going and improving, supported me through failure and success, provided me with the space and strength, and fueled me with delicious food, belongs to Suman. And now, this world is even more beautiful with Joyee, who brings a new meaning to love and life. iv TABLE OF CONTENTS ACKNOWLEDGEMENTS…….………………………..……………………………………….ii LIST OF TABLES…….……………………………………………………………………… viii LIST OF FIGURES…….……………………………………………………………………..….ix Chapter 1: Introduction ............................................................................................... ……………1 1.1 Numeral classifiers vs. noun classifiers ................................................................................ 2 1.2 Bangla ................................................................................................................................... 3 1.2.1 Is Bangla a noun classifier language? ............................................................................ 8 1.2.2 Is Bangla a numeral classifier language? ....................................................................... 9 1.3 The question: Number-marking in Bangla ......................................................................... 15 1.4 Overview of the chapters .................................................................................................... 16 Chapter 2: Bangla bare nominals .................................................................................................. 18 2.1 Introduction ......................................................................................................................... 18 2.2 Properties of Bangla bare nouns ......................................................................................... 18 2.2.1 Number neutrality ......................................................................................................... 18 2.2.2 Arguments to kind and generic predicates ................................................................... 19 2.2.3 Indefiniteness ................................................................................................................ 20 2.2.3.1 Indefinites vs. Pseudo-incorporation ......................................................................... 22 2.2.4 Definiteness .................................................................................................................. 26 2.2.5 Summary of observations ............................................................................................. 29 2.3 How Bangla differs from other classifier languages ........................................................... 29 2.3.1 Mandarin Chinese ......................................................................................................... 29 2.3.2 Yi .................................................................................................................................. 31 2.3.3 Thai ............................................................................................................................... 32 2.4 Bare nouns: Basic assumptions ........................................................................................... 34 2.4.1 Bare nouns as kind terms .............................................................................................. 34 2.4.2 Bare nouns as properties ............................................................................................... 40 2.5 Proposal: Syntactic individuation and the bare noun .......................................................... 42 2.5.1 Individuation ................................................................................................................ 42 2.5.2 Bare nouns .................................................................................................................... 43 2.5.3 Bare nouns in classifier languages ............................................................................... 45 2.6 Summary of the chapter ...................................................................................................... 45 Chapter 3: Definiteness in Bangla ................................................................................................ 47 Introduction ......................................................................................................................... 47 3.1 Bangla definite descriptions ................................................................................................ 48 3.2 3.2.1 Bangla definite forms ................................................................................................... 48 3.2.2 Section summary .......................................................................................................... 52 Licensing contexts for Bangla definite forms ..................................................................... 52 3.3 Bangla definite forms in licensing contexts ........................................................................ 55 3.4 3.4.1 Context A: Discourse-anaphoric definite reference ..................................................... 55 3.4.2 Context B: Effects of visibility and pointing to a referent ........................................... 59 3.4.3 Context C: Bridging cross-reference ............................................................................ 61 3.4.4 Context D: Non-anaphoric salient uniqueness ............................................................. 64 3.4.4.1 Context D.1 - Global uniques .................................................................................. 64 3.4.4.2 Context D.2: Situational uniques ............................................................................. 66 v 3.4.4.3 Context D.3: Inferenced uniques ............................................................................. 68 3.4.5 Summary of observations ............................................................................................. 70 Existing theories of Bangla definiteness ............................................................................. 71 3.5 3.5.1 Specificity vs. Definiteness .......................................................................................... 71 3.5.2 Familiarity, anaphoricity & uniqueness ....................................................................... 75 3.5.3 Maximality and uniqueness presupposition ................................................................. 76 3.5.4 Section summary .......................................................................................................... 79 Towards the proposal: Domain restriction and definiteness ............................................... 80 3.6 3.6.1 Domain restricted quantifiers with –Ta and –gulo ....................................................... 81 3.6.2 Domain restriction and DPs ......................................................................................... 83 3.6.3 Two DPs in Bangla ....................................................................................................... 84 3.6.3.1 Anaphoric DPs ......................................................................................................... 85 3.6.3.2 Unique DPs .............................................................................................................. 86 3.6.3.3 Bangla free choice items and –gulo ......................................................................... 87 3.6.4 Section Summary ......................................................................................................... 88 Conclusions ......................................................................................................................... 88 3.7 Chapter 4: -Ta ............................................................................................................................... 90 Distributions of –Ta ............................................................................................................ 90 4.1 4.1.1 –Ta with numerals ........................................................................................................ 90 4.1.2 –Ta with nouns ............................................................................................................. 91 4.1.3 –Ta with quantifiers ...................................................................................................... 93 New observations: Interpretations in the absence of –Ta ................................................... 94 4.2 4.2.1 Numerals without –Ta .................................................................................................. 94 4.2.1.1 Approximation ......................................................................................................... 94 4.2.1.2 Attributive modifiers ................................................................................................ 97 4.2.2 Measure NPs without –Ta: Weak noun phrases ........................................................... 98 4.2.3 Quantifiers with -Ta: Exact measure ............................................................................ 99 4.2.4 Optionality of classifiers cross-linguistically ............................................................. 102 4.2.5 Section summary ........................................................................................................ 103 Previous work on classifiers and number marking ........................................................... 104 4.3 4.3.1 Classifiers function between numeral and atomic entities (Jiang 2012) .................... 104 4.3.1.1 Limitations ............................................................................................................. 106 4.3.2 Classifiers as number markers (Nomoto 2013) .......................................................... 107 4.3.2.1 Two types of numerals cross-linguistically ........................................................... 108 4.3.2.2 Limitations ............................................................................................................. 110 4.3.3 Numerals as non-restrictive modifiers (Ionin & Matushansky 2006) ........................ 111 4.3.4 Section summary ........................................................................................................ 112 Associating precise counting with partition: Towards the proposal ................................. 112 4.4 4.4.1 -Ta is required for precise counting ........................................................................... 113 4.4.2 Proposal: Introducing partition ................................................................................... 114 4.4.3 Continuation: Introducing units in numeral constructions ......................................... 117 4.4.3.1 Numeral constructions with and without -Ta ......................................................... 118 4.4.3.2 Optionality of –Ta with large round numbers ........................................................ 120 4.4.3.3 Measure constructions ........................................................................................... 121 4.4.4 Implications: Explaining the observations ................................................................. 123 4.4.4.1 -Ta-less numeral constructions as attributive adjectives ....................................... 123 vi 4.4.4.2 -Ta with quantifiers ................................................................................................ 124 4.4.4.3 Quantifiers and obligatory mass interpretation ...................................................... 129 4.4.4.4 Bare classifier and definiteness .............................................................................. 132 4.4.4.5 Approximation with numeral doubling .................................................................. 132 4.4.4.6 The exact measure in correlative demonstratives .................................................. 133 Summary ........................................................................................................................... 133 4.5 Chapter 5: Plurality in Bangla ..................................................................................................... 134 Introduction ....................................................................................................................... 135 5.1 Difference between the two plurals .................................................................................. 135 5.2 –gulo ................................................................................................................................. 137 5.3 5.3.1 –gulo and plurality ...................................................................................................... 137 5.3.2 –gulo and definiteness ................................................................................................ 139 5.3.3 Novel observations ..................................................................................................... 142 5.3.3.1 Compatibility with sub-kind predicates ................................................................. 142 5.3.3.2 Distributive interpretations and access to atomic elements ................................... 143 5.3.4 Implications of accessibility of atomic elements in NP–gulo ........................................ 145 5.3.4.1 Scalar count-mass distinction .................................................................................. 146 The proposal: –gulo introduces a partition containing pluralities .................................... 148 5.4 5.4.1 Section summary ........................................................................................................ 150 -ra ...................................................................................................................................... 150 5.5 5.5.1 Associative plural, in Bangla and cross-linguistically ............................................... 150 5.5.2 Differences between the plurals ................................................................................. 152 Proposal............................................................................................................................. 153 5.6 5.6.1 Section Summary ....................................................................................................... 158 Consequences of two plural mechanisms ......................................................................... 158 5.7 Summary of the chapter .................................................................................................... 159 5.8 Chapter 6: Processing difference in quantifiers involving -gulo ................................................ 160 Introduction ....................................................................................................................... 160 6.1 Onek-(gulo): Various contributing factors ........................................................................ 162 6.2 6.2.1 Onek is Cardinal, Onek-gulo is proportional ............................................................. 162 6.2.2 Onek is high in absolute number, Onek-gulo is contextually restricted .................... 163 6.2.3 Size and quantifiability of objects .............................................................................. 164 6.2.4 Contributions of -gulo ................................................................................................ 165 6.3 Two modes of (cardinality) measurement: processing implications ................................ 165 6.4Previous work on quantifier understanding ....................................................................... 167 6.5 Current study ..................................................................................................................... 169 6.6 Methods............................................................................................................................. 170 6.6.1 Participants ................................................................................................................. 170 6.6.2 Materials and design ................................................................................................... 170 6.6.3 Data analysis ............................................................................................................... 171 6.6.4 Results ........................................................................................................................ 171 6.6.5 Discussion .................................................................................................................. 175 Conclusion ........................................................................................................................ 176 6.7 Chapter 7: Conclusions and Implications ................................................................................... 177 7.1 Summary of the dissertation ............................................................................................. 177 vii 7.2 Other closely-related languages in India ........................................................................... 179 .2.1 Oriya (Odia) ............................................................................................................... 179 7 7 Assamese (Asamiya) .................................................................................................. 181 .2.2 Conclusion ........................................................................................................................ 183 7.3 References ……………………………………………………………………………………..183 viii LIST OF FIGURES Figure 1: Granularity of scales .................................................................................................... 120 Figure 2 - Visual representation of one block in the experiment ................................................ 171 Figure 3: Percentage of acceptance ............................................................................................. 172 Figure 4: Reaction time for each quantifier ................................................................................ 173 ix LIST OF TABLES Table 1 Properties of Bangla in comparison to numeral classifier languages .............................. 14 Table 2: Properties of Bangla bare nouns ..................................................................................... 29 Table 3 Summary of bare nouns and their properties across classifier languages ........................ 33 Table 4: Prediction of the NMP and Bangla ................................................................................. 39 Table 5: Bangla definite descriptions in discourse-anaphoric contexts ........................................ 59 Table 6: Bangla definite descriptions in discourse-anaphoric and visible/pointing contexts ....... 61 Table 7: Bangla definite descriptions in discourse-anaphoric, visible/pointing and bridging contexts ................................................................................................................................. 63 Table 8: Bangla definite descriptions in anaphoric and global unique contexts ........................... 65 Table 9: Bangla definite descriptions in anaphoric and unique contexts ...................................... 67 Table 10: Bangla definite descriptions in anaphoric and non-anaphoric contexts ....................... 70 Table 11: Quantifiers with classifiers and count-mass selection .................................................. 83 Table 12: Quantifiers with classifiers and count-mass selection ................................................ 125 Table 13 Combining dimensions and measurements .................................................................. 134 Table 14: Comparison of quantifier strength with and without -gulo ........................................ 141 Table 15: Comparison of associative plurals in Mandarin Chinese, Japanese and Bangla ........ 152 Table 16 - Comparison NP-gulo with NP-ra .............................................................................. 152 Table 17: Individuation across languages and in Bangla ............................................................ 159 Table 18 - Conditions for the experiment and number of items per condition ........................... 170 Table 19: Percentage of acceptance for each quantifier against conditions .............................. 172 Table 20: Reaction time for each quantifier against conditions ................................................ 172 Table 21: Significance of distribution of different ratios across cardinality and ratio ............... 174 1 Chapter 1: Introduction A ‘proto-typical’ classifier language is believed to lack morphological number marking. Mandarin Chinese and Japanese have been often described as lacking plural number marking (Greenberg 1972, Allan 1977, Chierchia 1998). Although there have been some reports of plural affix-like elements in these languages, these markers are different than regular plural markers, for example, English –s. Japanese suffix –tati in gakusei-tati [student-TATI] ‘students’, for example, is considered to be an associative plural due to its prominent interpretation as ‘students and others’. In addition, a regular plural interpretation ‘students’ is also available (Iljic 1994, Kurafuji 2004, Li 1999, Nakanishi & Tomioka 2004). Moreover, as typically found in classifier languages, plural interpretation (number neutral, to be precise) is also available in the bare noun gakusei ‘student(s)’. Classifier languages, thus, have essentially no role in theorizing number marking (e.g. Chierchia 1998b, Corbett 2000). The claim that classifier languages lack number marking is somewhat puzzling because classifiers and number markers do not perform mutually incompatible functions: classifiers portion out units from noun denotation, while number markers quantify over those units. Thus, contrary to what has been suggested by cross-linguistic research, classifiers and number markers could, in principle, be compatible. This long-standing claim has been recently challenged in recent literature (Borer 2005, Nomoto 2013 among others). While Borer (2005) claims that number markers are underlying classifiers, Nomoto (2013) proposes that classifiers are number markers. In this work, I attest, on the basis of the evidence put forward by Bangla, that classifiers and number markers indeed perform similar functions. I discuss the functions of Bangla particles that are often categorized as classifiers, and demonstrate that they portion out units, and leading to answer a broad question concerning quantification that involves measuring the portioned units against appropriate scales (cardinal and non-cardinal). This proposal has implications for counting and quantifier processing. Additionally, it accounts for some long-standing puzzles associated with measure expressions. This chapter begins with a brief typological summary of classifiers and classifier languages (1.1). I simultaneously introduce readers to Bangla classifier particles, and briefly discuss the semantic properties and restrictions of such particles (1.2). I choose two particles that 2 are prevalent in the language and compare their properties with that of the proto-typical noun classifiers (1.2.1) and numeral classifiers (1.2.2). Next I provide a brief background of the questions asked in this dissertation, followed by a summary of the chapters (1.3). 1.1. Numeral classifiers vs. noun classifiers Before categorizing Bangla as a classifier language, let us elaborate on what classifiers are, and how we categorize classifier languages. In functional typology, classifiers are generally defined as noun categorization devices (Aikhenvald 2000). Most prevalent ones across the languages are the numeral classifiers. These classifiers must co-occur with numerals and quantifiers in a sentence. The nominal or noun classifiers do not require a numeral. These classifiers categorize the noun on the basis of its core semantic characteristics such as animacy, sex, humanness, shape etc. Noun classifiers often have independent semantic content and can be used in discourse- anaphoric contexts. Distinct noun classifiers with the same noun result in different words -- a property that is not found with numeral classifiers. Consider the following examples from a noun classifier language, Minangkabau (an Austronesian language spoken in Indonesia), where different noun classifiers result in different meaning (Aikhenvald 2000: 84). (1) a. batang limau CL TREE lemon ‘lemon-tree’ b. buah limau CL FRUIT lemon ‘lemon-fruit’ In the languages of Southeast Asia, noun classifiers are often considered a subtype of numeral classifiers (Aikhenvald 2000). Two properties, obligatory co-occurrence with numerals and unavailability of anaphoric reference, distinguish the numeral classifiers from the noun classifiers. Noun classifiers may co-occur with a numeral, but the presence is not obligatory, as it is in the numeral classifiers. Only noun classifiers may be used anaphorically, a property that 3 numeral classifiers lack. Aikhenvald additionally argues that the classifiers are different in their syntactic position in the NP: while the immediate scope of the noun classifier is the noun; the scope of the numeral classifier is the NumP or QP. As expected of its adjacency with the head noun, a noun classifier characterizes the noun, and is generally affixed to the noun. Quantification and enumeration are the primary functions of a numeral classifier. Consider the following example from Korean. The classifier facilitates counting while providing semantic characterization (e.g. long, slender etc.) to the noun (Aikhenvald 2000: 107). (2) sey calwu-uy yenphil-lul sa-ss-ta. three CL- LONG.SLENDER.ATT pencil-ACC bought ‘(I) bought three pencils.’ Depending on the functions of the set of classifiers, a language is categorized as a noun classifier language or a numeral classifier language. However, cross-linguistically, such distinction is often opaque as languages may contain both noun and numeral classifiers (except Australian languages, possesses only noun classifiers). East Asian languages (such as Mandarin Chinese, Japanese, Vietnamese, and many others) are often claimed to be numeral classifier languages, particularly due to high frequency of such particles with numerals. Aikhenvald (2000: 105) categorizes Bangla as a numeral classifier language and lists five classifiers. Additionally, Bangla has two particles -gulo and –ra, generally categorized as plural classifiers (Dayal 2014 among others). Below I present a brief overview of Bangla particles, and evaluate their properties against typologically attested properties of nominal and numeral classifiers. 1.2. Bangla Bangla is a South Asian language from the Eastern Indo Aryan (Indic) language family, spoken in the Indian subcontinent. As the national language of Bangladesh and one of 23 official languages in India, it has around 189 million native speakers combining speakers in Bangladesh and in the Indian states of West Bengal, Tripura and Andaman (according to 2011 census). The Bangla represented in this dissertation is the standard colloquial version spoken in/near Kolkata, India. As noted, Aikhenvald (2000) lists five classifiers in Bangla: -Ta (countable non-humans), - 4 Ti (diminutive of -Ta), -jon (human), -khana (solid object with flat or rectangular shape), and – khani (diminutive of -khana). However, their properties exceed this categorization. For example, in addition to countable non-human nouns (e.g. table), -Ta also co-occurs with non-honorific human nouns (e.g., boy in (3a)). Dasgupta (2013) claims that –Ta has pejorative connotations, and it does not generally co-occur with human referents that are accorded high levels of respect. In such cases, the diminutive counterpart -Ti is used. This also indicates that –Ti does not always refer to small countable objects (3c), as claimed. The human classifier –jOn is preferred in highly respected human noun referents (3d). –khana and –khani generally combine with flat surface objects like book, notebooks etc., as in (3e & f). (3) a. du-To Tebil ar du-To chele two-CL table and two-CL boy ‘two tables and two boys’ b. du-Ti Tebil two-CL DIM table ‘two (small) tables’ c. du-{#Ta/-Ti} mohila two-CL/-CL DIM woman ‘two women’ d. du-jon mohila two-CL HUM woman ‘two women’ e. du-khana khata two-CL FLAT notebook ‘two notebooks’ f. du-khani SondeS 5 two-CL DIM.FLAT sweet ‘two (pieces of) desserts (a square or diamond shaped dairy-based dessert)’ Most of these particles also appear with nouns in the absence of numerals, much as like the noun classifiers. This word order (i.e. ‘bare classifier’ 1 ) is not available for the human classifier –jon (4c). The diminutive classifier –khani in this word order does not appear with small flat objects, and is preferred with human body parts (e.g., face, feet, hand etc.), used in an endearing connotation (4e & f). (4) a. Tebil-Ta table-CL ‘the table’ b. mohila-{#Ta/-Ti} woman-CL/-CL DIM ‘the woman’ c. *mohila-jon woman-CL HUM d. khata-khana notebook-CL FLAT ‘the notebook’ e. *SondeS-khani sweet-CL DIM.FLAT f. mukh-khani face-CL DIM.FLAT ‘the (endearing) face’ 1 Adopting Simpson’s (2005 and subsequent publications) terminology. 6 In addition, Bangla has two more particles (i.e. the plural –gulo and -ra) that do not co-occur with numerals, and appear only in the bare classifier form, contributing to plural number marking. Similar to –Ta, -gulo is also susceptible to a pejorative connotation, and does not co- occur with human noun referents with high levels of respect. Consider the following examples showing incompatibility of –Ta and –gulo with +honorific nouns such as gentleman (5b). In such cases, the diminutive versions –Ti and –guli are felicitous (5c). Accordingly, these diminutive versions are infelicitous with non-honorific words (5d). (5) a. lok-Ta/-gulo man-TA/-GULO ‘the man’ b. #bhOdrolok-Ta/-gulo gentleman-TA/-GULO ‘the gentleman’ c. mohila-Ti/guli woman hon -TI/-GULI ‘the woman’ d. #magi-Ti/-guli woman non-hon -TI/-GULI ‘the woman’ In addition to these classifiers and their diminutive counterparts ending in [-i], -ra is compatible with animate nouns only (also with inanimate nouns in anthropomorphic world where books can walk, talk etc.). (6) a. {chele/goru}-ra boy/cow-RA 7 ‘boys’ ‘cows’ b. #dOrja-ra door-RA Another restricted particle, –Tuku, predominantly appears with mass nouns to denote diminutive amount. Unlike any other particles discussed so far, it also combines with demonstratives to produce a diminutive demonstrative. (7) a. dudh-Tuku milk-TUKU ‘the little amount of milk’ b. eiTuku hati this-TUKU elephant ‘such a small elephant (in comparison to the standard)’ The distribution of these particles suggests that they have characteristic properties of both numeral and noun classifiers. Among these particles, -Ta, –gulo and -ra are predominant. - khana, -khani and -Tuku rarely occur in common speech. In this dissertation I focus on the properties of –Ta, -gulo and –ra. In the next section I take turn in evaluating the properties of noun and numeral classifier languages to evaluate if Bangla is better categorized as a noun classifier language or a numeral classifier language. I arrive at the generalization that it represents a close affinity to numeral classifier properties. However, as we will see in section 1.2.2, such categorization is not appropriate either as there are interesting patterns in the data that calls for an investigation. 8 1.2.1. Is Bangla a noun classifier language? A characteristic feature of nominal classifiers is that such languages allow the co-occurrence of several noun classifiers within one noun phrase. Reduplication of functional morphemes is generally unavailable in Bangla, as in (8). 2 (8) a. *tin-Ta-Ta bacca three-TA-TA child b. *Onek-gulo-gulo jOl many-GULO-GULO water Furthermore, noun classifiers are often used anaphorically. In the following example from Yidiny (a noun classifier language), use of the classifiers corresponding to nouns (e.g., ‘John’ and ‘snake’) anaphorically relates the noun referents. (9) Context: John saw a snake. xil naj no7 (Craig 1992: 284) saw cl.man cl.animal ‘He(man) saw it(animal).’ -Ta and -gulo are appropriate in anaphoric contexts, however, unlike the example in (9), they do not have independent lexical existence. (10) a. EkTa chele eSechilo. chele-Ta lOmba. one-TA boy came. boy-TA tall ‘A boy came. The boy was tall.’ b. car-Te chele eSechilo. chele-gulo lOmba. 2 Only the diminutive morpheme –Tuku can be reduplicated when it appears in combination with demonstratives and count nouns. 9 four-TA boy came. boy-GULO tall ‘Four boys came. The boys were tall.’ Although, Bangla –Ta and –gulo show anaphoric reference in the absence of numerals, -Ta does not have any anaphoric dependency when numerals are present, as evident from the first parts of the previous examples. Thus, anaphoricity is not inherent in -Ta. Additionally, these particles have no independent lexical existence, and they are bound morphemes. Thus, it is unlikely that Bangla could be categorized as a noun classifier language. 1.2.2. Is Bangla a numeral classifier language? Although we have mentioned the typological definition of classifier languages, let us elaborate on what we mean by a numeral classifier syntactically. A theory-neutral definition often treats it as a ‘measure word’ that must be used before a numeral can combine with any noun. A numeral classifier language thus differs from a number-marking language (e.g., languages of Germanic and Romance language families) where only mass nouns require a measure word in order to be counted (‘a glass of milk’), while count nouns can directly combine with numerals (‘three glasses’). A count noun in a classifier language always requires a measure word, either by means of a classifier or a measure expression (e.g., glasses(s), box(es)). Bangla nouns also require a measure word in the nouns phrases. Below I evaluate Bangla in terms of the properties reported for languages that are generally categorized as numeral classifier languages. I follow Jiang (2012) for the key properties. These properties might differ from its typological descriptions. (11) a. tin-*(Te) kukur three -TA dog ‘three dogs’ b. tin *(gelas) jOl three glass water ‘three glasses of water’ 10 The first key property of a classifier language is that it allows bare arguments (Chierchia 1998a: 354). That is, nouns may appear in a bare form without the presence of numerals, classifiers, determiners, or plural elements in argument positions. Bare nouns in Bangla are allowed in argument positions, however, as we discuss the details in chapter 2, these bare arguments have various interpretations. For example, they can be interpreted as kind terms, definite descriptions, indefinite nouns, pseudo-incorporated nouns, and even as compounds. Such multiplicity of interpretation alludes to the possibility of different (null) functional projections. Below I illustrate the copious readings of Bangla bare arguments. (12) a. kukur probhubOkto prani. (Kind) dog faithful animal ‘The dog is a faithful animal.’ b. kukur rat-e Dake. (Characterizing) dog night-LOC barks ‘Dog(s) bark at night.’ c. ami kukur dekhi-ni. (Indefinite) I dog saw-NEG ‘I didn’t see a/any dog.’ d. kukur biskuT kheyeche. (Definite by non-anaphoric uniqueness) dog biscuit ate ‘The dog ate biscuit(s).’ e. ami cabi bEg-e rekhechilam. (Definite by non-unique familiarity) I key bag-LOC put-PST ‘I kept the key in the bag.’ f. ami boi poRi. (Pseudo-Incorporation) I book read 11 ‘I read some book or other.’ g. nei kaj to khoi bhaj. (Compound, semantically bleached) NEG work PRT puffed-rice fry ‘Idle work’ The second key property is that the bare nouns are generally number-neutral. They do not specify singularity and plurality of the referent. Bangla bare nouns, if not interpreted as definite, indicate number neutrality (e.g., kukur ‘dog(s)’) in (12 a-c, e-f). The third property concerns plurality. Although generally considered lacking plural markers, all classifier languages have ways of marking plurality. However, plural markers never appear with container, measure, and group classifiers (e.g., the counterpart of two cup(*s) of milk). This property distinguishes classifier languages from number marking languages with canonical number markers such as Romance and Germanic languages, where number features are marked not only on count nouns but also on container/measure/group/partitive nouns. For example, plural morphology is marked on count nouns as well as container/measure/group /partitive nouns if the numeral is more than ‘one’ (13c-d). In contrast, Bangla plural ‘classifier’ – gulo (Dayal 2012, 2014, Biswas 2014) never appears with numerals or measure words (13a-b). (13) a. du-*(To) kap two-TA cup ‘two cups’ b. du-(*Ta) kap-(*Ta) ca two-TA cup-TA tea ‘two cups of tea’ c. two cup*(s) d. two cup*(s) of tea 12 However, Bangla -gulo (that is akin to the regular plural English -s) has an entirely different set of properties than that of the –ra (akin to Mandarin men). They have different implications for number marking and definiteness. While –gulo is always interpreted as a definite semantic plural, -ra is interpreted as an associative plural and a human plural. (14) a. kukur-gulo dog-GULO ‘the dogs’ (definite plural) b. chele-ra boy-RA ‘boys and others’ (associative plural) ‘boys’ (indefinite plural) Some classifier languages allow ‘bare classifier’ structure (cf. Simpson 2005) where the noun co- occurs with the classifier without the numeral. Such phrases in Cantonese are interpreted as definite. Bare classifier phrases in Bangla too represent definiteness. (15) a. kap-Ta cup-TA ‘the cup’ b. kap-gulo cup-GULO ‘the cups’ Another important property of classifier languages concerns whether (and how) definiteness is marked. Classifier languages lack overt functional D projections (Chierchia 1998b, but see Simpson 2005 and subsequent work). However, many documented cases where definiteness is marked by either functional categories (i.e. suffixes or article-like markers) or by dedicated word orders suggest that there is DP projection in these languages. Bangla has least five ways to mark 13 definiteness, as shown below. To account for such varied representation of definiteness, first and foremost, we need the presence of functional projections involving D 0 . Furthermore, the definite representation must be more nuanced in terms of how and why so many different ways surface in terms of definiteness (will be discussed in chapter 3). (16) a. kap cup ‘(the) cup’ b. kap-Ta/gulo cup-TA/-GULO ‘the cup’ ‘the cups’ c. kap du-To cup two-TA ‘the two cups’ d. ei kap du-To this cup two-TA ‘these two cups’ e. ei du-To kap this two-TA cup ‘these two cups’ In Sum, Bangla fits to the traditional characterization of a (numeral) classifier language. The following table summarizes the properties discussed here (adapted from Jiang 2012). 14 Properties of a (numeral) classifier language Bangla Bare nouns as arguments ✓ Bare nouns are number neutral ✓ No classifier on container /measure /group /partitive nouns ✓ Markers of plurality have variable interpretations ✓ Allows bare classifiers without numerals ✓ Definiteness marking either by a determiner or by word order ✓ Obligatory measure word on every count noun ? No recurrence of classifiers ✓ Table 1 Properties of Bangla in comparison to numeral classifier languages As reported earlier, Aikhenvald (2000) claims that Bangla is a numeral classifier language. While Indic languages are not generally classifier languages, it is possible that the numeral classifier properties of this small eastern Indo-Aryan group of languages (including Odia, Assamese, Mythili and Bangla) are highly influenced by their Sino-Tibetan and Austro-Asiatic neighbors (e.g., Bodo, Khashi etc.), which have well-developed numeral classifier systems. Bangla –Ta in different syntactic positions partially satisfies both criteria of numeral (Num-Ta N) and noun classifiers (N-Ta). A possible direction would be to postulate that Bangla has two sets of classifiers (such as Minangkabau). However, it is unlikely that the same classifier (i.e. –Ta) is represented in both systems. These Bangla particles have a lot of idiosyncratic features to combine them under one umbrella term. It is possible, as suggested in Simpson (2005) for other east Asian languages, that Bangla classifiers are going through a grammaticalization process where they are losing their classifier properties. It is indeed supported by the number of classifiers found and regularly used in the speech of older generation and newer generation. For example, older speakers use –khana for towel, -gacha for rope (p.c. a 95 year-old native speaker), whereas, younger speakers do not know such classifiers exist (p.c. a 14 year-old native speaker). Thus, I abandon the idea of classifying Bangla into one or other; rather I focus on understanding the properties of these particles. Although Bangla fits the characterization of a numeral classifier language, we will see that a detailed investigation is required to explain the number marking and definiteness properties. As alluded in the overview, definiteness and plural marking is integrated in one form whereas there 15 are various ways of representing definiteness in this language. Along the lines, Bangla also shows two markers of plurality, again varied in terms of their definiteness construal. This dissertation aims to provide a comprehensive account of the Bangla particles that are traditionally defined as classifiers. 1.3. The question: Number-marking in Bangla Going back to our broad question that relates to the co-occurrence restriction of classifiers and plural markers, we see that the classifier –gulo, that is considered to be a plural classifier, does not co-occur with numerals. This is analogous to the cross-linguistic observations supporting the hypothesis. Chierchia (1998b) proposes that all nouns in classifier languages are mass nouns. Mass nouns have cumulative reference. Thus, if a predicate is true of two individuals (a, b), it is also true of their sum (a+b). Plural markers are redundant on mass nouns, and are predicted to be absent from classifier languages. Arguing for a unified account of classifier and number-marking languages, Borer (2005) proposes that bare nouns in all languages denote mass-like properties. Number markers (e.g. English –s) are underlying classifiers, and they individuate the mass-like denotation of the nouns. Since both plural markers and classifiers individuate, they are in complementary distribution. This account allows the co-existence of classifiers and plural markers in a language, but rules out the co-occurrence of the two in the same noun phrase. Cross-linguistic research validates Borer’s claim and Chierchia’s predictions encounter several counter-examples. Recent work suggests, despite the lack of overt morpho-syntactic representation of number marking, a productive underlying syntactic process for number marking may be present, which is represented in non-canonical ways, for example, by reduplication of classifiers in Mandarin Chinese or Thai (Jenks 2011, Nomoto 2013, and Zhang 2013). Lexical properties of nouns (i.e., whether the noun denotation is individuated, and whether individuation is represented in the lexicon) are assumed to distinguish classifier languages from non-classifier languages. Chierchia (1998b) proposes a cross-linguistically applicable semantic parameter (the Nominal Mapping Parameter) that connects the presence of individuation in the noun denotation to the availability of bare arguments in languages, suggesting nouns in classifier languages denote non-individuated (mass) entities. Subsequent work, both on theoretical and experimental domains, refutes such a claim, and suggests that 16 individuation is dependent on the presence of syntax, and not lexically derived (Barner & Snedeker 2005, Bale & Barner 2009, Inagaki & Barner 2009, Li, Dunham & Carey 2009, Doetjes 2012, and subsequent work). Insights from such work introduce new ideas concerning the function of classifiers. For example, Watanabe (2006) proposes that classifiers do not individuate, but facilitate counting by numerals (see also Sudo 2016 for same ideas in Japanese). In this work, I side with the latter group in exploring alternative approaches to the classifier, number marking and denotation of nouns in classifier languages. 1.4. Overview of the chapters In chapter 2, I discuss bare nominal arguments in Bangla. I argue against the neo-Carlsonian assumptions (Chierchia 1998b, Dayal 2012, 2014, Jiang 2012), and show that assuming bare nouns as properties have significant advantages for explaining the observations. I argue for additional null functional projections for different interpretations of the bare arguments in Bangla. This account is in accordance with cross-linguistic properties of the bare nouns and provides optimal features for the lexicon cross-linguistically. I also show that the number neutrality of bare nouns in classifier languages have a significant role towards categorizing them the way they are. Chapter 3 presents a comprehensive account of different types of definiteness constructions in Bangla. I show that Bangla has five morpho-syntactic mechanisms to represent definiteness. I explore these constructions against carefully created contexts, unifying five constructions under one standard syntactic mechanism that depends heavily on definiteness construal by uniqueness and familiarity/anaphoricity. This chapter provides strong support for the presence of functional projections for definiteness, and I argue that there are two DP projections, associated with the type of definiteness construal. Some part of the data in this chapter has been presented in Biswas (2014a) and Simpson & Biswas (2016). Chapter 4 discusses –Ta. The properties of –Ta are very puzzling in Bangla, and lead to interesting unresolved puzzles regarding several important topics including the count-mass distinction, number marking, quantificational properties and definiteness. I put forth a novel proposal that –Ta contributes to portioning out singular portions from the denotation of the noun. The implications of this proposal extend to how nouns combine with numerals and other 17 counting quantifiers alike. This novel proposal, along with the proposals for definiteness, accounts for the presupposition interpretations in quantifiers and anaphoric definite reference in a specific word order. This proposal accounts for new observations regarding measure expressions concerning the presence and absence of –Ta, and explains empirical findings concerning the availability of number marking in classifier languages cross-linguistically. Part of this chapter is influenced by a joint work (O’Connor & Biswas, submitted) Chapter 5 investigates the plurals in Bangla, -gulo and -ra. –gulo, with its [+plural] feature acts as the plural counterpart of –Ta, whereas -ra is categorially different from –gulo and acts as an associative plural. The proposal for –gulo dissociates definiteness from the meaning of -gulo in the obligatory anaphoric contexts, and presuppositional strong quantifier interpretations. –ra, the associative plural, is adjoined at the DP level and it modifies denotation to a group plural, explaining its compatibility with pronouns. This chapter also shows that the count-mass distinction in Bangla is not a binary phenomenon. It is rather a continuum of individuated elements represented along the scale of individuation (following Grimm 2012) that explains the characteristic features of the plural makers in Bangla. Part of this chapter has been presented in Biswas (2013, 2014b). In chapter 6, I report an experimental investigation concerning the role of –gulo in quantifier constructions. I show that the definiteness associated with –gulo turns cardinal quantifiers into proportional quantifiers, restricting the set to the current context. Chapter 7 concludes this dissertation by discussing the implications and cross-linguistic extensions of the proposals in this dissertation. 18 Chapter 2: Bangla bare nominals 2.1 Introduction The scope of this chapter is to explore the distribution of Bangla bare nouns, focusing on their interpretations in different positions and contexts, and how these interpretations can be accounted for syntactically and semantically. Bangla bare nouns appear in a range of positions, and accordingly have a range of interpretations. Bare nouns can be arguments to kind predicates. They can also be interpreted as non-specific indefinites, pseudo-incorporated nouns and as definite in non-anaphoric unique contexts. Bare nouns are also preferred when the referent is human and generally accorded high respect in their role. In this chapter, I first present an overview of the positions and available interpretations of Bangla bare nouns (section 2.2). Then I compare Bangla bare nouns with that of other classifier languages, leading to propose that bare nouns can be either an NP or a DP (section 2.3). I start my proposal by assuming that bare nouns are properties (section 2.4). I argue that there are two syntactic positions that account for the various interpretations of the bare noun phrases in Bangla (section 2.5). 2.2 Properties of Bangla bare nouns 2.2.1 Number neutrality Bangla bare nouns are number neutral (i.e. unspecified for singular or plural). Unlike English nouns, Bangla nouns do not generally require accompanying functional elements (e.g. article, definite determiners etc.) in the argument positions. In the object position, if not accompanied by a case marker, they are underspecified for number. Although a bare noun is constrained in the subject position (i.e. bare human nouns are preferred in the subject position than the inanimate ones), it can be number neutral in either subject or object position. Examples (17a) and (17b) show that the bare nouns are truly number neutral. (17c) shows that in a context where singular number is impossible, only a plural interpretation arises with the bare noun. This suggests that the denotation of the bare noun includes both atomic and non-atomic entities in its domain. Akin to the bare nouns in classifier languages, Bangla bare nouns have narrow scope in the presence of a negation operator, as illustrated in (17b). 19 (17) a. phOl fruit ‘fruit(s)’ b. ami phOl khaini. I fruit ate-NEG ‘I didn’t eat fruit(s).’ c. Saradin gach theke phOl poReche. whole-day tree from fruit dropped ‘Fruits dropped from the tree the whole day.’ 2.2.2 Arguments to kind and generic predicates Bare nouns can be arguments to kind predicates. Due to this property, the bare nouns are often assumed to denote kind terms (18a) (cf. Chierchia 1998, later Dayal 2012, 2014). Bare nouns are also commonly found as arguments to generic/characterizing predicates (18b). This intuition is clear in a context that facilitates a contrastive focus reading, as in (18c). (18) a. kukur probhubOkto prani. (Kind) dog faithful animal ‘The dog is a faithful animal.’ b. kukur rat-e Dake. (Characterizing) dog night-LOC barks ‘Dogs bark at night.’ c. beRal dudh bhalobaSe, kukur nOe. CAT milk loves, DOG NEG ‘The cat loves milk, the dog doesn’t.’ 20 2.2.3 Indefiniteness Bare nouns in object position can be interpreted as three different syntactic objects: weak indefinites, pseudo-incorporated nominals and compound predicates. Although these three structures are syntactically identical, their interpretations represent a scale of availability of ‘true’ argument properties. The indefinites are the typical arguments while the noun in the compound modifies the verb to form a complex predicate, losing a ‘true’ argument status. For example, the referent of a weak indefinite ‘dog’ in (19a) can be referred back by an overt pronoun. Such anaphoric connection is not found in the pseudo-incorporation structure in (19b) where the referent of ‘book’ is referentially opaque. The compound in (19c) is fully coerced into the semantics of the verb, and do not allow any anaphoric link to the nouns. (19) a. ami baire kukur dekhechi. oTa lej naRchilo. (Indefinite) I outside dog saw that tail wagging-pst ‘I saw dog(s) outside. It was wagging its tail.’ b. ami boi poRechi. #oTa bhalo chilo. (Pseudo-Incorporation) I book read that good was ‘I read some book or other. #It was good.’ c. nei kaj to khoi bhaj. *oTa puriye phelo na. (Compound) have-NEG work PRT rice fry. That burn drop-NEG ‘Idle work.’ We will shortly elaborate on the distinction between bare noun indefinites and pseudo- Incorporation structures, but before that let me point out two more properties of Bangla bare nouns. Bangla bare nouns do not have strong indefinite reading. The following examples are unavailable in terms of a strong indefinite reading, where other readings (i.e. weak indefinite and definite) are possible. 21 (20) a. #gaRi gErej-e ache. car garage-LOC be Intended: ‘A (certain) car is in the garage.’ (Strong indefinite) Interpreted: ‘The car is in the garage.’ (Definite) b. ami apel kinechi. I apple bought Intended: ‘I bought one apple.’ (strong indefinite) Interpreted: ‘I bought apple(s).’ (weak indefinite) The semantic difference between the weak and the strong indefinites involves availability of a partitive reading. Milsark (1977) observed that the indefinite DPs (e.g. two girls), although not marked overtly for partitive, could be ambiguous between ‘two girls’ and ‘two of the girls’. This partitive specificity is generally considered a diagnostic of indefiniteness in numerically quantified DPs (Milsark 1977, Dayal 2014). However, Dobrovie-Sorin et al. (2012, Ch. 4, pg. 136) argues that a non-partitive strong reading is also available for the strong indefinites. While partitive readings presuppose a contextually determined set of entities, non-partitive strong indefinites do not presuppose such a set. These indefinites are not quantificational. They cannot have distributive interpretation and cannot be interpreted as a dependent plural. The most accessible readings of this type of indefinites are the speaker-specific indefinites (e.g. English ‘I saw this man...’), which are again unavailable for Bangla bare nouns. (21) a. ramu *(Ekjon) iNrej-ke biye korte cay. Ramu one-TA British-ACC marry do wants Intended: ‘Ramu wants to marry a British.’ Intermediate scope reading is also unavailable. In the following example, we see that the bare noun ‘paper’ can only have the narrowest possible scope. (22) prottek chatro du-To biSoye paper poReche. each student two-TA subject-LOC paper read 22 (i) ‘Every student read paper(s) in two subjects.’ (ii) ‘Every student read paper(s) in the two subjects.’ (iii) #’Every student read all the paper(s) in two subjects.’ The unavailability of the specific indefinite readings in Bangla bare nouns indicates a lack of a quantification structure in the bare nouns. It suggests that the bare noun indefinites are more restricted in their interpretation. Below I reflect on the interpretations of indefinite bare nouns against the pseudo-incorporated structure. The data show that, despite the superficial similarity in the structures, the pseudo-incorporation cases are semantically different than the indefinites. 2.2.3.1 Indefinites vs. Pseudo-incorporation Various types of incorporation have been discussed in the literature for a long period of time (see Massam 2009 for details). Here I discuss how Bangla bare nouns form pseudo-incorporated structures that are indistinguishable from the weak indefinites. Although pseudo-incorporation cases are superficially similar to weak indefinites and share a lot of properties, not all bare nouns in the object position can be interpreted as incorporated nominals. Pseudo-incorporation has been argued to be phrasal in nature, with morpho-syntactic (e.g., reduced structures, word order) and semantic effects (e.g., low scope, number-neutrality). A detailed description of Bangla pseudo- incorporation structure can be found in Biswas (2011). Following Dayal (2011), I proposed that Bangla pseudo-incorporated nominals have a phrasal syntax, and they are interpreted as properties. Here I report only the properties that set the bare incorporated nouns apart from bare indefinites. Test 1: Human indefinite or definite nouns must have case markers in the object position, whereas pseudo-incorporation cases do not. (23) a. chele-r jonno meye dEkho. (Pseudo-incorporation, no case) boy-GEN for girl see ‘Look for brides for your son.’ 23 b. mohila-ke Dako. (Indefinite, case) woman-ACC call ‘Call the woman.’ Test 2: While indefinite NPs allow anaphoric link to the referent in continuing discourse, pseudo-incorporated nominals do not. Anaphorically linked pronouns result in semantically infelicitous interpretations in the pseudo-incorporated structure (24a). Bare indefinite noun phrases on the other hand, allow discourse anaphora in (24b). (24) a. mira rate boi poRchilo, (Pseudo-incorporation, #pro) Mira night-LOC book was-reading ‘Mira was reading some book or the other at night,’ #o SeTa Tebil-e rekhechilo. she that table-LOC kept ‘She kept it on the table.’ b. mira rate ca kheyechilo, (Indefinite, pro) Mira night-LOC tea ate ‘Mira drank tea at night,’ o SeTa ThanDa kore kheyechilo. she that cold be-PRT drank ‘She drank it cold.’ Although the previous examples show that there is a difference between the bare indefinite nouns and pseudo-incorporated nouns in terms of anaphoric reference, it must be noted that noun-verb compounds in Bangla do not allow anaphoric link at all. I do not know of any tests to distinguish compounds from pseudo-incorporation cases. 24 (25) a. mira protibad korechilo, (Compound) Mira protest did ‘Mira protested.’ *kintu o-Ta khub-Ekta joralo chilo na. but that NPI strong be NEG ‘but it was not strong.’ Test 3: Mohanan (1995) demonstrates that gapping is not possible in pseudo-incorporated structures, while it is allowed in weak indefinites. In the following example, bari kena bEca [house buy sell] ‘real estate business’ is a pseudo-incorporation structure, whereas botol kena bEca [bottle buy sell] does not have any corresponding pseudo-incorporation status. A weak indefinite object ‘bottle’ can be elided while the same precluded in the pseudo-incorporation. The distinction, however, is very subtle. (26) a. #Mir baRi kene ar ram _ bEce. (Pseudo-incorporation) Mir house buys and Ram _ sells ‘Mir does house-buying and Ram does house-selling.’ b. Mir botol kinlo ar ram _ beclo. (Indefinite) Mir bottle bought and Ram _ sold ‘Mir bought bottles books and Ram sold them.’ Test 4: A pseudo-incorporated nominal can only be modified by an adjective when its proto- typical properties are reserved. No such restriction applies to weak indefinites. For example, in the pseudo-incorporation predicate, boi-baMdhano [book-binding] ‘book-binding’, ‘old’ is acceptable as it retains the proto-typicality of books, while ‘square’ does not. In the indefinite scenario, if book is an argument of the predicate ‘picking’, the adjective do not have to retain the proto-typicality of the object argument, and thus no restriction. The availability of modification suggests that pseudo-incorporation cases involve NPs and not nouns (contra Baker 1996). 25 (27) a. Mir {purono/#couko} boi baMdhae. (Pseudo-incorporation) Mir {old/# square} book binds ‘Mir binds old-books.’ b. Mir {purono/bhari} boi tole. (Indefinite) Mir {old/heavy} book picks ‘Mir picks old/heavy books.’ Test 5: Another piece of evidence that pseudo-incorporation cases are different from the weak indefinite noun phrases is that the former is string adjacent to the VP, while the latter is not. Pseudo-incorporation does not allow object shift. No VP-level adverb may intervene the bare noun and the verb in the pseudo-incorporation cases, while objects may be shifted to the front by intervening VP-level adverbs. The unavailability of adverbial modification also suggests that the nominal in pseudo-incorporation cases is adjacent to the VP, while the object arguments are not. This indicates that the former involve a lower position in the VP and hence has to be string adjacent to the VP. (28) a. chele-r jonno taRataRi meye (*taRataRi ) dEkho. (P-Incorporation) boy-GEN for quickly girl see ‘Quickly look for brides for your son.’ b. taRataRi lok (taRataRi) Dako. (Indefinite) quickly man quickly call-IMP ‘Quickly call some people.’ The diagnostic tests in this section suggest that, despite of the lack of surface-structural evidence of difference between pseudo-incorporation and indefinite bare NP structures, they are semantically different. The availability of intervening VP-level adverb presents a clue for the structural difference, as will be elaborated later. The incorporation cases are string adjacent to the VP while the indefinite cases are not. 26 2.2.4 Definiteness In addition to the kind and indefinite interpretations, Bangla bare nouns can be definite in argument positions, if both the speaker and the hearer uniquely identify the referent. The referents of the bare noun definite do not have to be identifiable by means of a previous context or discourse. The referent may be familiar to the speaker if the speaker has a particular referent in mind, but the context does not provide any information about the referent. If the referent is familiar and anaphoric to a context, a bare noun cannot occur in such cases. In the following example, the bare noun kukur ‘dog’ is interpreted as definite; the referent is uniquely identifiable by both speaker and hearer in the context (29a). The same rationale holds for human noun referents in the subject position in (29b). (29) a. kukur biskuT kheyeche. dog biscuit ate ‘The dog ate biscuit(s).’ b. mukkhomontri boktrita dicchen. chief-minister speech giving ‘The chief-minister is giving a speech.’ In the object position of episodic sentences, Bangla bare nouns can be definite if the referent is uniquely identifiable to the interlocutors (Biswas 2014, Simpson & Biswas 2016). The referent ‘key’ in (30a) is definite by identifiability, although they are not anaphoric to the context. ‘The moon’ in (30b) is globally unique and hence identifiable to the interlocutors. (30) a. ami cabi bEg-e rekhechilam. I key bag-LOC put-PST ‘I kept the key in the bag.’ b. ami caMd dekhini. I moon saw-NEG ‘I didn’t see the moon.’ 27 Bangla bare nouns are non-anaphoric unique definites. The distinction associated with the anaphoric link is illustrated in the following example. (31a) shows a bare noun pattern associated with the existence of a unique referent in the context that is previously unfamiliar to the speaker. The referent is unique in the context – neither is it familiar to the interlocutors nor is it identified by means of a previous context. In contrast, (31b) shows, if the speaker has a particular referent in mind, a marked definite form (the ‘bare classifier’ form) must be used (31b). (31) Context: Everyday on planet Varg, a security vehicle delivers one new prisoner to guards from the camp. Today the vehicle drives up, the back doors open out and no one is inside. The guard says: a. kOedi kothae? prisoner where ‘Where is the prisoner?’ (Simpson & Biswas 2016, ex. 27) b. kOedi-(*Ta) kothae? prisoner-TA where ‘Where is the prisoner?’ These examples illustrate that a bare noun definite is licensed when the referent is not anaphoric in the context. A bare noun may be preferred when the referent is unique in the context, or identifiable by the interlocutor by means of activation of the referent in mind or by vicinity, but they are not allowed where the referent is anaphoric to a previous context. Additionally, socio-linguistic factors may enforce the bare noun pattern for definite reference even if other forms of definite are expected in that context. For example, the referents of human nouns that are generally considered highly respected, strongly prefer a bare noun in the definite context. If due to negative attitude or other socio-linguistic factors, the respect is suspended, a bare noun pattern may not be used. This is a strong constraint. In (32), two nouns (i.e., president and janitor), which are generally subjected to different levels of respect, represent definiteness by selection of different forms. In an anaphoric context where a bare noun form is 28 not licensed, the noun with high respect referent (i.e. ‘president’) appear in the bare noun form whereas the other (i.e., ‘janitor’ – low respect), as expected, is in the bare classifier pattern. (32) Ofis-e Ek jon notun presiDenT ar Ek-jon mEthor niyog kOra hoeche. office 1-CL new president and one-CL janitor employ do was president-(*TA) bOyoSko manuS, tObe mEthor-*(Ta)-r bOyeS kOm. president-TA elderly person but janitor-TA-GEN age less ‘The office hired a new president and a janitor. The president is an elderly person, but the janitor is young.’ Similarly, a negative-quality predicate may cause to disallow a bare noun pattern. As in (33), the negative-quality predicate (i.e. ‘being lazy’) with a referent that is generally respected (e.g. ‘officer’) licenses the bare classifier pattern, despite of the expectation of the bare noun pattern. (33) 1990 Sale Ofis-e Ek jon Ofisar niyog kOra-hoechilo. 1990 year office-LOC one-CL officer employ -do-be ‘The office hired an officer in the year 1990.’ kintu officer *(Ta) khub kuMRe tai kono unnoti korte-pareni. but officer-*(CL) very lazy so NPI promotion do-be-able-NEG ‘But the officer could not get promotion because he was very lazy.’ These examples illustrate how the speaker’s perception of a human referent in a particular context may influence and override a default selection of definiteness pattern. A bare noun pattern is generally observed in non-anaphoric unique contexts; however, the constraints may override the general pattern. 29 2.2.5 Summary of observations So far we have seen that the Bangla bare noun is number neutral and is interpreted as weak indefinites in the object position of transitive verbs. It may appear as arguments to generic and kind predicates. The noun can be interpreted as non-anaphoric unique definite in both in subject and object positions. In addition, it appears in pseudo-incorporated structures. The following table summarizes the properties discussed so far. Bare nouns Bangla - are number neutral ✓ - appears in the subject position ✓ - appears in the object position ✓ - are interpreted indefinite ✓ - are interpreted Definite ✓ - are interpreted as strong indefinites ✗ Table 2: Properties of Bangla bare nouns It is difficult to determine the exact nature of the bare nouns as it has a varied range of interpretations. The unavailability of the strong indefinite shows that these bare nouns cannot be interpreted as individuals or generalized quantifiers. This is contradictory because the bare noun in its definite interpretation is necessarily interpreted as an individual. To account for the differences syntactically, one may argue for the involvement of extra functional projections. I will elaborated on this later, but first let us discuss the available and relevant interpretations of bare nouns in related languages. 2.3 How Bangla differs from other classifier languages 2.3.1 Mandarin Chinese Mandarin noun phrases have been discussed in detail in the literature (Cheng & Sybesma 1999, Yang 2001, X. Li 2001, Jiang 2012 among others). Mandarin bare nouns can appear kind-level predicates and in generic sentences. They receive a definite or a narrow scope existential 30 interpretation in episodic sentences. The following examples from Jiang (2012) illustrate the properties of the bare nouns in Mandarin Chinese. (34) a. xiong jue-zhong le. (Kind predicate) bear vanish-king Asp ‘The bear is extinct.’ b. gou hen jiling. (Generic predicate) dog very smart (i) ‘Dogs are intelligent.’ (ii) ‘The dog(s) is/are intelligent.’ c. wo kanjian gou le. (Episodic sentences) I see dog Asp (i) ‘I saw some dog(s).’ (ii) ‘I saw the dog(s).’ d. waimian gou zai jiao outside dog Prog bark (i) ‘Outside, dogs are barking.’ (ii) ‘Outside, the dog(s) are/is barking.’ Mandarin bare nouns can either be interpreted as definite or an existential weak indefinite but no intermediate scope is available with bare nouns. (35) dabufen yuyanxuejia dou kan-guo mei-ge [jiejue wenti de] fenxifangfa. most linguist all look-Asp every-CL solve problem DE analysis (i)‘Most linguists have looked at every analysis that solves the problem.’[definite] (ii) ‘Most linguists have looked at every analysis that solves problems.’ a. Narrow scope: [most > ∀> problems] b. *Intermediate scope: *[most >some problems/a problem> ∀] 31 Intended: ‘For most linguists, there are some problems/there is a problem that they have looked every analysis of’ Similarly, adopting Jiang’s (2012) example, we see that Bangla bare nouns do not have intermediate scope either. In fact, Bangla does not allow a definite reading for the bare noun ‘paper’ either. It only has an existential narrow scope. (36) prottek chatro du-To biSoye paper poReche. each student two-TA subject-LOC paper read (i) ‘Every student read papers in two subjects.’ (ii) ‘Every student read papers in the two subjects.’ But as Jiang (2012, pp. 179-180) mentions, Mandarin bare nouns can anaphorically refer to the maximal plural member in the antecedent clause. Note that such definite construal is not available in Bangla. As the scopal relation between the quantifiers in the above example require a strong definite construal, that Bangla bare nouns do not have a wide scope associated with the definite interpretation can be explained. 2.3.2 Yi Jiang also discusses Yi, a Tibeto-Burman classifier language spoken in China. Yi bare nouns freely appear in argument positions, can be interpreted as kind-terms and appear in generic sentences. These bare nouns can have an existential narrow scope interpretation, as well an anaphoric and referential definite interpretation – just like Mandarin and unlike Bangla. The following Yi examples are from Jiang (2012). (37) a. ne si-hni shyp ngop ddu ggep la go-li, nga khat ox. you girl bring my home play come if, I happy SFP (i) ‘If you bring girls to hang out in my house, I will be happy.’ [if >girls] (Narrow scope Indefinite) (ii) Not: there are some girls, if you bring them to my house, I will be happy.’ 32 *[some girls > if] (16) a. mu lindoo. horse lose SFP ‘The horse(s) got lost.’ (Definite) b. si-hni ma sini sse-vo ma i go nyi, si-hni nyi-zi ku jjo nzi. girl CL and boy CL house sit, girl 20 year have-seem ‘A girl and a boy are sitting in the house, the girl looks like in her 20s.’ (Ana. Def.) 2.3.3 Thai Thai bare nouns can be arguments of kind predicates and generic sentences. As arguments of episodic sentences, a Thai bare noun is interpreted as narrow scope existential indefinite. Referential and anaphoric definite readings are also possible in Thai bare nouns, just as Mandarin and Yi. The following examples are from Jiang (2012, pp. 413-414). (16) a. nuu klai suunpan (Kind predicate) mouse almost extinct ‘Mice are almost extinct.’ b. nuu aasai taam thonaam (Generic) mouse live in sewer ‘Mice live in the sewer.’ c. chan mai hen maew nai hong (Episodic) I NEG see cat in room (i) ‘I didn’t see cats in the room.’ [not > ∃; * ∃>not] (Narrow scope indefinite) (ii) ‘I didn’t see the cat in the room.’ (Referential definite) d. muawaan nuu khaw maa nai khrua 33 yesterday mouse enter come in kitchen wannii nuu haay pai laew (Anaphoric definite) today mouse disappear ASP already ‘Yesterday, a mouse/mice came in the kitchen. Today, the mouse/mice disappear.’ Similar distributions of the bare noun is found in Wu and Min, two relatively less-described classifier languages (see Cheng & Sybesma 2005 for details). Although Cheng & Sybesma claims Cantonese bare nouns to be different from other classifier languages as the bare NPs cannot be interpreted as definite, and can only have a generic and an indefinite reading, later work (Simpson et al. 2011 a.o.), however, shows that bare nouns are available in definite contexts. In an extensive investigation of the bare nouns (and the bare classifier phrases) in different classifier languages, Simpson et al. (2011) reports that in Vietnamese, Cantonese, Hmong and Bangla bare nouns are available for definite contexts. They are however, less preferred in anaphoric contexts. Additionally, Jiang (2015) reports availability of definite, indefinite and kind/generic readings of the bare nouns Min, Japanese, Vietnamese (see Jiang 2015, pp. 328, for a summary of strategies to express (in)definiteness in classifier languages). Below I present a summary of bare nouns and their interpretations in classifier languages discussed in this chapter on the basis of available data (unavailable data is marked NA in the table). Bare nouns Bangla Mandarin Cantonese Thai Yi - number neutral ✓ ✓ ✓ ✓ ✓ - appears in the subject position ✓ ✓ ✓ ✓ ✓ - appears in the object position ✓ ✓ ✓ ✓ ✓ - interpreted indefinite ✓ ✓ ✓ ✓ ✓ - interpreted referential definite ✓ ✓ ✓ ✓ ✓ - interpreted as anaphoric definite ✗ ✓ ? ✓ ✓ - interpreted as strong indefinites ✗ ✗ NA NA NA Table 3 Summary of bare nouns and their properties across classifier languages 34 From our discussion of other classifier languages, we see that Bangla bare nouns show similar properties of bare nouns in other classifier languages. Bangla (and Cantonese) differs from other classifier languages in terms of definiteness construal in anaphoric contexts: while Bangla strongly prefers an alternative marked form in the anaphoric contexts, bare nouns in other languages allow anaphoric definite reference. In the next section, I discuss a well-accepted account for the nominals in classifier languages (Chierchia 1998b). I show that the predictions of this account are not thoroughly borne out in Bangla, and I ultimately take a different route for the basic assumptions. In the following, I first discuss Chierchia’s (1998b) proposal for bare nouns considering Bangla to be a classifier language. This Neo-Carlsonian approach to bare nouns in classifier languages assumes that bare nouns enter the derivation as kind terms. I show that Bangla bare nouns can be arguments to kind predicates, but due to additional interpretations a different set of assumptions may fare better. I assume that bare nouns are properties, and I argue that additional functional projections are required to explain the interpretations. 2.4 Bare nouns: Basic assumptions 2.4.1 Bare nouns as kind terms Chierchia (1998b), in his remarkable attempt to unify the cross-linguistic variation of nominal semantics, argues that bare nouns in classifier languages ‘can, or sometimes even must refer to kinds’ (Chierchia 1998: 343-344). In this neo-Carlsonian school of thought, classifier languages and number-marking languages are different on the basis of what the bare nouns denote and whether their maximal projections are argumental ([+arg]) or predicative ([+pred]). If a language is argumental, bare nouns may appear as arguments (e.g. Chinese, Japanese), whereas determiners are required for predicative languages (e.g, French, Italian). To explain this cross- linguistic variation, Chierchia (1998b) proposes the Nominal Mapping Parameter (NMP). According to the NMP, classifier languages are argumental [+Arg, -Pred], and the bare nouns in classifier languages are kind terms. Kind terms are saturated and they allow bare nouns as arguments. They are functions from the world to pluralities, ‘the sum of all instances of the kind’ (Chierchia 1998b: 349). Due to their reference to pluralities, kinds are mass-like, and contain both atomic and non-atomic entities in the denotation. They do not differentiate between singularities and pluralities. Therefore, classifiers are required for individuation before they 35 combine with numerals. Classifiers individuate the mass-like denotation into units for quantification. Below I discuss properties that fit the NMP and novel observations that show that Bangla does not fit well in terms of the predictions of the NMP as a classifier language. A few properties of Bangla bare nouns are compatible with the claim that bare nouns are kind terms. They are number neutral and can be arguments to kind predicates. However, the fact that Bangla bare nouns appear in argument positions and can be interpreted as both definite and indefinite entities, as discussed earlier, are incompatible for the kind thesis unless an additional semantic apparatus is proposed. The main assumption for the kinds, as put forward by the NMP, requires the denotation of the kind term to be a ‘well-established’ property of being that kind. Shared knowledge of speakers thus makes more impact on defining a kind than the rules of grammar. For example, dog will be considered a kind term, but a mythical creature that has only one instance in any world (e.g., chewbacca) will not be. No corresponding distributional variation is noticed in Bangla. The first argument against the assumption for bare nouns as kind terms involves Bangla human nouns in kind predicates. Bare nouns as kind terms predict that they yield generic reading with individual-level predicates. Generally, this is the case with Bangla bare nouns, except for human nouns, as shown below, that require an associative plural (-ra) when they are arguments to kind predicates. (38) a. DoDo-pakhi-(ra) Obolupto (hoe gEche). dodo-bird-RA extinct become ‘Dodos are extinct.’ b. chatro-*(ra) Obolupto (hoe gEche). student-(RA) extinct become ‘Students are extinct.’ c. RiSi SikkhOk-*(der) 3 sroddha-kOre. 3 -der is the oblique form of –ra. It can be optional with certain predicates. I conjecture that such optionality is due to the nature of the predicate and not associated with the bare nouns properties, per se. 36 Rishi teacher-RA-GEN respect-does ‘Rishi respects teachers.’ –ra is obligatory for human nouns as kind terms. If we assume all nouns in Bangla are kind terms, and we want to accommodate this observation, we will have to assume a different semantics for the human nouns than other nouns. Analogous to the predictions of the NMP, Bangla bare nouns are generally interpreted scope-less i.e., they scope under other scope-bearing operators in the sentence. Number neutral bare nouns are often compared to English bare plurals for this reason. Consider the following examples (originally from Carlson 1977, ch. 2), adapted in Chierchia 1998b, pp. 370), where the English bare plural in (39a) takes narrow scope in the presence of the negation operator, the indefinites (39b) are ambiguous between a wide and a narrow scope interpretation. (39) a. Miles didn’t see policemen. (see > ∃, *∃ > see) b. Miles didn’t see a policeman. (see > ∃, ∃ > see) In comparison, Bangla bare nouns are scope-less too. However, they show an additional array of interpretations. For example, they take wide scope when interpretations analogous to definite descriptions are available. The car in the following examples can have both wide and narrow scope, along with a differentiated scope reading. (40) RiSi gaRi dEkheni. Rishi car saw-NEG ‘Rishi didn’t see cars.’ (Narrow scope) ‘Rishi didn’t see the car.’ (Wide scope) The bare nouns and the singular indefinite in English differ in terms of their interpretations in differentiated scope environments. These are the situations where the referent of the noun phrase must take narrowest possible scope. English bare plurals take the differentiated scope whereas indefinite singulars do not. 37 (41) a. Cars burnt in Chicago and New York. b. *A car burnt in Chicago and New York. The differentiated scope reading is clearer in the following pair of Bangla examples. Along with the opaque or non-referential reading (42a), Bangla bare nouns can also have a differentiated or narrowest possible scope when the adverb scopes over the object in (42b). Availability of the narrowest possible scope in Bangla bare nouns rules out existence of null functional projections. (42) a. RiSi kukur puSte cay. (Opaque reading) Rishi dog to-pet wants ‘Rishi wants to pet dogs.’ b. RiSi barbar kukur puSeche. (Differentiated/narrowest scope) Rishi repeatedly dog pet ‘Rishi repeatedly pet dogs.’ Interestingly, the noun phrases that cannot be considered a kind (e.g., modified NPs) also have similar interpretations. Modified bare nouns in Bangla can be opaque, scope-less and have differentiated scope with respect to other operators in the sentence, whereas similar NPs in English have both opaque and transparent readings, and wide scope. (43) a. RiSI [lOmba lej-wala kukur] puSte cay. (Opaque reading) Rishi long tail-PRT dog to-pet wants ‘Rishi wants to pet long-tailed dogs.’ b. RiSi [lOmba lej-wala kukur] poSeni. (Narrow scope) Rishi long tail-PRT dog pet-NEG ‘Rishi didn’t pet long-tailed dogs.’ c. RiSi barbar [lOmba lej-wala kukur] puSeche. (Differentiated scope) Rishi repeatedly long tail-PRT dog pet 38 ‘Rishi repeatedly pet long-tailed dogs.’ d. RiSi [lOmba lej-wala kukur] khuMjche, meri-o (*oTa-ke) khuMjche. Rishi long tail-PRT dog searching, Mary-too that searching ‘Rishi is looking for long-tailed dogs, Mary is looking for them too.’ Thus, Bangla modified NPs have different properties than English modified bare plurals. The data in (43) show that the bare NPs and modified bare NPs in Bangla have similar characteristics: they both have only opaque readings, scope under negation, require differentiated scope with adverbs and cannot be antecedents to anaphora. Whereas, English modified bare nouns have variable behavior with opacity and have wide scope. Chierchia (1998b) argues that the suspension of the scope-less-ness in English is indicative of existentially quantified NPs. In the same line of thought, Bangla modified bare NPs then lack existential quantification. The scope-less-ness of Bangla modified bare nouns suggests that (i) bare nouns and modified bare nouns are both available for narrowest possible scope, and (ii) wide scope reading requires presence of an overt/covert determiner. Therefore, it is not conclusive that Bangla bare nouns are kind terms. Additionally, the presence of English plural marker –s in ‘more bananas’ forces quantification on the basis of cardinality (i.e. counting), while the absence thereof in ‘more banana’, indicates non-cardinality (i.e. amount) quantification. This suggests availability of both count and mass interpretation in English bare nouns, and that –s here performs the same function as a classifier (cf. Borer 2005). When applied to Bangla, the absence of –Ta does not show amount interpretation at all. The following example shows that in the absence of the classifiers or the plural markers, beSi kOla ‘more banana’ only has a cardinal reading. It combines with predicates of counting. (5) a. ram beSi kOla guneche. Ram more banana counted ‘Ram counted more bananas.’ (Counting/cardinal) 39 b. ram smudi-te beSi kOla diyeche. Ram smoothie-LOC more banana gave ‘Ram gave much banana in the smoothie.’ (Amount/non-cardinal) This discussion suggests that the count mass distinction is present in Bangla bare nouns. It might be present in the lexicon or in the syntax. Wherever the distinction between the count and mass nouns are introduced (later I will argue that it is present in the syntax), this actually suggests that the bare nouns might not be just kind terms. Otherwise, as the NMP suggests, a classifier must be introduced for a count interpretation. Below I summarize the predictions of the NMP and how Bangla fares. We see that some of the predictions are borne out, whereas exceptions are found for the remaining. Predictions of the NMP Bangla Bare nouns are kind terms ? Bare nouns have mass-like denotations ✗ Bare NP argument possible ✓ No morphological singular plural distinction ✗ Mandatory classifier with numeral ✗ Table 4: Prediction of the NMP and Bangla Bare nouns in many classifier languages are compared to English bare plurals (Mandarin, Yang 2000; Thai, Jenks 2011). Bangla fits the description of classifier languages (as shown in chapter 1). Bangla bare nouns are thus predicted to be kind terms. However, as Bangla bare nouns have a range of interpretations: from the narrowest scope with adverbs to a wide scope with definite, suggests possibility of null functional projections. According to the NMP, Bangla is predicted to lack a morphological distinction between singular and plural number markers. However, as we will see in latter chapters, Bangla has number markers. The presence of plural markers does not necessarily argue against Chierchia’s proposal as he claims bare nouns in classifier languages ‘can, and or sometimes even must refer to kinds’ (Chierchia 1998b: 343-344). Such exceptions to the predictions of the NMP are not only found in Bangla, Indonesian also shows several inconsistencies in the predictions of the NMP (see Chung 2000). As Chung suggests, these inconsistencies raise critical questions regarding the rigidity of the NMP. Alternatively, this 40 raises concerns regarding the status of Bangla and what we assume for the bare nominals. This is why I probe into an alternative proposal and follow the assumptions that languages do not differ in terms denotation of the bare nouns. In other words, I assume nouns in all languages are uniformly assumed to be of same type. Thus, bare nouns in Bangla and other languages (e.g., English) nouns do not differ in their lexical properties. The distinction of count and mass nouns are derived in the syntax. 2.4.2 Bare nouns as properties Now let us discuss another set of assumptions that begins with bare nouns as properties. Bare nouns in predicate positions exhibit the classic characteristics of properties. 4 Bangla bare nouns appear in predicate positions regardless of a singular or a plural subject, as shown below in (44a & b). (44) a. RiSi Daktar. Rishi doctor ‘Rishi is a doctor.’ b. RiSi ar Riko Daktar. Rishi and Riko doctor ‘Rishi and Riko are doctors.’ This is also evidence that Bangla bare nouns and English bare plurals are not interpreted in the same way. Only bare plurals (and mass nouns) are allowed in the predicate position, whereas, even in the presence of plural subjects, as in the above example, does not allow bare singular nouns. 4 See for alternative approaches bare nouns as properties (see, Dobrovie-Sorin 1997, van Geenhoven 1998, 1999, van Geenhoven and McNally 2002, Chung and Ladusaw 2003, and Farkas and de Swart 2003). 41 Partee (2002) shows that verbs like become or consider take property-denoting noun phrases as arguments. In the following example, the arguments of consider involves the conjunct and ensures that the two constituents are of the same type <e,t>. (45) a. Mary considers John competent in semantics and an authority on unicorns. b. meri mone-kore jOn semantics-e dOkkho ar unicorn-er biSeSOggo. Mary considers John semantics-LOC competent and unicorn expert ‘Mary considers John competent in semantics and an authority on unicorns.’ Generally nouns of any type (argumental or predicative) are amenable to predicative uses. But if what Chierchia suggests is true then the nouns classifier languages do not require an overt or covert type-shift for appearing in kind predicates. In general, Bangla bare nouns appear in such positions without any morpho-syntactic marker. However, Bangla human nouns in subject positions of generic/characterizing predicates appear with a mandatory associative plural marker (-ra) (briefly mentioned in (38), elaborated below in (46)). Human bare nouns require associative plural marker (-der) for the kind and generic predicates (Biswas 2014). Non-human nouns appear bare in these predicates. (46) a. chele-ra meye(-der) pOchondo kOre. (Human) boy-RA girl-RA-GEN like does ‘Boys like girls.’ b. beRal iMdur mare. (Non-human) cat mouse kills ‘Cats kill mouse.’ As (46a) shows, the mandatory markedness of the human nouns in the subject and object positions of the characterizing predicates suggest that these nouns may not be kind terms. If they were kind terms, bare noun forms would have been expected in these predicates. 42 Another piece of evidence for bare nouns as properties comes from bare nouns in pseudo- incorporation structure. Pseudo-incorporation is different than a weak indefinite noun phrase and cannot be treated as kind terms (cf. Dayal 2011). Dayal argues that syntactically pseudo- incorporated nouns projects in the NumP, and are interpreted as properties. Earlier in this chapter, I discussed how pseudo-incorporation is different from the indefinite noun phrases. 2.5 Proposal: Syntactic individuation and the bare noun 2.5.1 Individuation So far in this chapter, I have shown the disadvantages of a lexicalist approach that assumes a count-mass distinction in the lexicon: it is cross-linguistically inapplicable and produces wrong predictions regarding bare nouns. I therefore have a different assumption for the bare noun. I assume that any lexical item enters the derivation as a root (i.e. without a syntactic category). Once embedded under a functional nominal structure, the roots are interpreted to be nouns. I propose an n 0 that categorizes the roots as nouns. It further contributes to the count mass distinction in nouns. However, instead of a binary distinction, I follow Grimm (2012) in assuming that individuation has scalar properties. Individuation is scalar and is dependent on connectedness (a property that depends on the strength between the atoms or instances in the noun denotation, and follows axioms of mereology) in various degrees. The countability can be gradable according to the status of the minimal noun elements in that language, and thus, allows a systematic approach to incorporate granular, aggregate noun (e.g. sand, fence, stones etc.) along with the proto-typical substance (e.g. water) and individuated (e.g. dog) nouns. Depending on how strongly the individuals are connected, the singular property of a count noun is able to pick out ‘whole’ instances of the object or sums thereof. ‘Strong connection’ between the instances means that one instance of water implies existence of another instance. ‘Locally’ connected indicates that the instance of water is a mass body of water, and not specified in terms of any other shape or container (e.g. glasses, puddles etc.). ‘locally strongly connected’ thus implies that the two instances may contain each other. A n 0 indiv maps the denotation of the root to one of predicate of individuals. Particularly, for a count noun, both atoms and sums are available in the denotation. The n 0 , if unmarked, returns a predicate of individuals, without specifying the structure of noun denotation. This is suitable for substance 43 mass nouns. According to Grimm (2012), each instance of substance nouns is locally strongly connected to the other. Thus, no minimal parts are available for substance masses, and only ‘portions’ of water are available. 2.5.2 Bare nouns The count mass distinction is thus syntactic. If n 0 [+individuated] returns a predicate of atomic and non-atomic individuals, a bare count noun can be interpreted as number-neutral. Syntactically, the nominal is interpreted as mass if the n 0 is not specified by [+individuated] feature (cf. Borer 2005, see Bale & Barner 2009, de Belder 2013 for a similar account, except for a n 0 mass head). Below is a schema for basic count mass distinction as represented in the syntax. (47) a. Count structure: b. Mass structure nP count nP mass n 0 indiv √ n 0 √ Both count and mass nouns are predicates of type <e,t>. When the n 0 indiv is introduced, the denotation of ‘apple’ includes atoms and sums. For example, if a, b, and c are apples then the noun ‘apple’ represents the set {a, b, c, ab, bc, ac, abc}. When n 0 is introduced for ‘water’, the domain is not segregated into atomic entities. Like an identity function, the nP represents water, a set of individuals that are portions of water. (48) a. ⟦apple⟧= λx.apple(x) b. ⟦water⟧= λx.water(x) With this assumption, let us now move to the distinction of indefinite and pseudo-incorporated nouns. Recall that the bare noun in the pseudo-incorporated structure is string adjacent to the predicate. The indefinite allows a VP-level adverb to intervene the noun and the verb. The example below is repeated from (28). 44 (49) a. chele-r jonno taRataRi meye (*taRataRi ) dEkho. (P-Incorporation) boy-GEN for quickly girl see ‘Quickly look for brides for your son.’ b. taRataRi lok (taRataRi) Dako. (Indefinite) quickly man quickly call ‘Quickly call some people.’ This relative immobility of the bare noun in the pseudo-incorporated structure, I argue, is due to the position of the noun, whether they are inside or outside the VP. I argue that the pseudo- incorporated NPs are ‘true’ NPs and are existentially bound in the VP domain. The indefinites are DPs and are existentially bound in the IP domain. This proposal is supported by the existing accounts of pseudo-incorporation (Dayal 2011, van Geenhoven 2003, among others). Generally, the incorporated nominals are existentially bound, and are composed via predicate modification. Dayal (2011) in her work on Hindi pseudo-incorporation proposes that pseudo-incorporation involve a NumP, as the Hindi data show interaction with morphological number marking in pseudo-incorporation cases. Bangla, however, lacks number marking, and Biswas (2011) shows that an NP suffices for the structure of pseudo-incorporation in Bangla. These NPs denote properties. If transitive verbs are predicates of events, then pseudo-incorporating verbs take a predicate as an argument: it pseudo-incorporates a predicative NP object. The subject DP becomes the only one individual argument. I have shown in the previous section that the pseudo- incorporated NPs are VP-internal whereas indefinites are VP-external. Here I am following standard assumptions for indefinite bare nouns for explaining indefiniteness and incorporation (following Diesing’s (1992) Mapping Hypothesis). Diesing applies the mapping hypothesis for weak versus strong indefinite NPs, suggesting that NPs of type <e, t> are interpreted as predicates remain in a VP-internal position and are therefore subject to Existential Closure. NPs of type <<e, t>, t> with a quantificational reading move out. The Existential Closure is applied at the VP domain for the pseudo-incorporation. It is at the IP domain for the indefinite occurrences. Additionally, I argue that the various interpretations depend on the presence of a null determiner. The pseudo-incorporation cases do not project a DP, whereas the indefinite and the 45 definite bare nouns have a null D, resulting in a DP projection (elaborated in the next chapter). The null D 0 requires licensing (under the ECP) and thus the bare noun appears in properly licensed positions. This allows bare noun indefinites in the subject and object positions. If the DP headed by a null determiner if contains a [+definite] feature, the NP undergoes movement, resulting in definite readings. If it is not [+def], the null determiner is semantically equivalent to English ‘a’ or ‘some’ and therefore allows variable scope. Thus we have both syntactic and semantic evidence in favor of a null determiner heading bare nouns. 2.5.3 Bare nouns in classifier languages The incompatibility of the number marking with the classifier languages has been discussed in a lot of languages, and often the literature reports instances of exception in such languages. The common property in these languages, that is often overlooked, is that the bare noun are interpreted as number neutral in all classifier languages. Alternatively, bare nouns in number marking languages are never interpreted as number neutral. Similar to the exceptions for plural marking in classifier languages, noun particles are often found sporadically in number-marking languages. However, the common property is that the bare nouns preclude number neutral denotation. The denotation of the bare noun in the classifier languages always includes atoms and sums, whereas, only atomic elements are available in number marking languages. Thus, a bigger and stronger generalization arises – the basic cross-linguistic distinction between the two types of languages do not involve the presence of overt number marker, rather it is the availability of sums in the denotation of the bare noun. In this dissertation I do not pursue on this clue, and it is open for future work. 2.6 Summary of the chapter Unlike the assumptions of the Neo-Carlsonian view (Chierchia 1998, Dayal 2009, 2014), I assume that bare nouns are devoid of any categorization properties, and enter the derivation as a root. A functional projection n 0 categorizes the roots and the nP denote properties. It also introduces individuation, allowing a division between the count and mass nouns. This chapter shows (i) although bare nouns in Bangla may be interpreted as kind terms, the data suggest 46 various other properties of the bare nouns. I proposed that pseudo-incorporated nominals involve an NP projection, whereas the rest of the distributions of the bare noun involve a DP projection headed by a null D. If the D is interpreted as [+definite], the NP undergoes movement to the DP and is interpreted as a definite description. If not, it is interpreted as an indefinite DP. I also argue that the number neutrality of bare nouns in classifier languages are indicative of the availability of number, and the focus on the incompatibility with plural marking need to be shifted. This chapter presents the basic building blocks of the bare nouns alone. 47 Chapter 3: Definiteness in Bangla Introduction 3.1 The goal of this chapter is to analyze the expression of definiteness in Bangla. As briefly discussed in chapter 2, Bangla bare nouns construe definiteness. However, more marked forms are available in Bangla for definiteness construal. Several concepts such as anaphoricity, familiarity, identifiability, maximality, inclusiveness, and uniqueness have been argued to be key to understanding definiteness. Among the proposals for Bangla in particular, anaphoricity/familiarity, uniqueness (Biswas 2011, Simpson & Biswas 2016), maximality (Dayal 2012, 2014) and inclusiveness (Syed 2016) have been claimed to explain various aspects of definiteness. These, nonetheless, do not provide a complete account of all available morpho- syntactic forms and interpretations. In this chapter, I first present an overview of five different morpho-syntactic representations of definite constructions in Bangla (section 3.2), followed by the licensing contexts (section 3.3). Then I discuss each form in different contexts to explore definiteness construal in Bangla (section 3.4). My syntactic proposal falls along the lines of previous syntactic accounts for Bangla definiteness that involves a DP projection with a null D 0 with a strong [+definite] feature that motivates an NP-movement to the Spec, DP. Additionally, I argue for a lower DP projection for explaining the domain restriction found in some definite constructions (section 3.5). In addition, I explore domain-restricted readings of the quantifiers with –gulo and –Ta, and argue that Bangla involves a lower DP projection (following Zamparelli 2000) (section3.6). I provide a syntactic account for the two types of definite constructions found in Bangla. 48 Bangla definite descriptions 3.2 3.2.1 Bangla definite forms Bangla definites can be expressed in five ways. Among them, three forms (‘bare classifier’ 5 , bare noun and NP-fronted with numerals 6 ) have been sporadically discussed in Bangla literature (Simpson et al. 2011, Biswas 2014a, 2014b; Dayal 2012, 2014; Syed 2015, 2016; Simpson & Biswas 2016). There are additional morpho-syntactic patterns for definiteness construal. The complete array has five forms. (I) N (Bare noun) The previous chapter discussed the bare noun definites. These definite appear in the argument positions of episodic sentences. The referent must be uniquely identifiable to the interlocutors, but it is not anaphorically linked to the context. In the following example, the bare nouns kukur ‘dog’ and cabi ‘key(s)’ are interpreted as definites (29). The referents are uniquely identifiable by both speaker and hearer but the information is not provided by the context. (50) a. Situation: John is talking to Mary about their sick dog: kukur oSudh kheyeche. dog medicine ate ‘The dog took the medicine.’ b. Situation: John and Mary are standing in front of their house, and John is searching for the keys in his bag: ami cabi bEg-e rekhechilam. I key bag-LOC put-PST ‘I kept the key(s) in the bag.’ 5 The term ‘bare classifier’ is used in all works of Andrew Simpson to refer to the noun complex where the classifier appears with a noun in the absence of a numeral (Simpson 2005, Simpson et al. 2011, Simpson & Biswas 2016). This form unambiguously represents definiteness in some classifier languages (e.g., Hmong, Vietnamese, Cantonese). I use this term in this chapter for all occurrences of N-Ta and N-gulo to avoid terminological confusion. However, I would like to emphasize that I do not necessarily pertain to the claim –Ta and –gulo classifiers, per se. 6 The non-canonical word order represented by an NP-fronted order where the noun phrase appears before the numeral has also been referred to as the ‘alternative’, ‘NP-raised’, ‘NP-Object-Shift’. 49 (II) N-Ta/gulo (Bare classifier) The bare classifier form is represented by the noun-classifier form without any numeral. Although -Ta generally appears with a numeral, and the form (e.g., ‘one-TA book’) is interpreted as indefinite. In the absence of a numeral, the –Ta may appear with the noun, in the ‘bare classifier’ form (i.e. book-Ta), and is interpreted as definite singular. Similarly, -gulo in the bare classifier form (i.e. book-gulo) is interpreted as definite plural. These forms are represented below: the definite singular NP-Ta is in (51b), and the definite plural NP-gulo is in (51c). (51) a. Ek-Ta pakhi one-TA bird ‘a bird’ b. pakhi-Ta bird-TA ‘the bird’ c. pakhi-gulo bird- GULO ‘the birds’ The NP-Ta has been discussed extensively in Bangla syntactic literature (Dasgupta 1983, Bhattacharya 1999, 2000, Ghosh 2010, Chacón 2010, Simpson 2005, Dayal 2014, Biswas 2014a, Simpson & Biswas 2016 among others). NP-gulo has been discussed in details in Biswas (2014b). The following examples (52) seem to be in free variation with the examples with bare nouns in (1). However, the examples in (52) involve the bare classifier form, and the referents must be overtly present in immediate discourse. (52) a. rasta-e EkTa kukur dekhlam. kukur-*(Ta) biskuT khacchilo. road-LOC one-TA dog saw Dog-TA biscuit was-eating ‘I saw a dog on the street. The dog was eating biscuit(s).’ 50 b. Situation: John is searching for his keys, and says to Mary: tin-Te cabi ei bEg-e chilo. cabi-(*gulo) khuMje pacchi na. three-TA key this bag-LOC was. Key-GULO find-PPL get NEG ‘Three keys were in this bag. I can’t find the keys.’ (III) NP Num-Ta (NP-fronted numerically quantified DPs) Definiteness in numerically quantified DPs is expressed by a non-canonical word order, where the NP appears before the numeral (i.e. NP Num-Ta). The canonical word order (Num-Ta NP) is interpreted as indefinite. This word order is limited to lower numerals (2-4). Any attempt to raise the phrasal constituent past the numeral greater than 6 is consistently disallowed in Bangla (cf. Syed 2015) (53c). A demonstrative assisted definiteness is used in stead, as shown below. (53) a. ami lal boi du-To/tin-Te/char-Te kinechi. I red book 2-TA/3-TA/4-TA bought ‘I have bought the two/three/four red books.’ b. ?ami lal boi paMch-Ta kinechi . I red book 5-TA bought ‘I have bought the five red books.’ c. *ami lal boi chO-Ta/Sat-Ta/aT-Ta kinechi . I red book 6-TA/7-TA/8-TA bought ‘I have bought the six/seven/either red books.’ (Syed 2015, ex. 22-27) d. ami oi chO-Ta/Sat-Ta/aT-Ta lal boi kinechi . I that 6-TA/7-TA/8-TA red book bought ‘I have bought those six/seven/either red books.’ Syed (2015) and Simpson & Syed (to appear) argue that the discrepancy in the pattern is caused by different base generation locations of the numerals. The low numerals are heads in QP while the high numerals are projected in the specifier of QP. The definiteness-related successive cyclic 51 movement of the NP to the DP is blocked by the higher numeral in the Spec, QP that creates an intervention effect. (IV) Dem Num-Ta NP (Demonstratives with NP in default order) Numerally quantified DPs, where the numerals are higher than 5, can be definite only by means of the demonstrative. The following example (54) shows that definiteness is achieved only by means of the demonstrative when the numeral is higher than 5. (54) a. ei/oi/Sei dOS-Ta chele [Dem with Num > 5 & canonical word order] this/that/that ten-TA boy ‘these/those ten boys’ b. *ei/oi/Sei chele dOS-Ta [*Dem with Num > 5 & non-canonical word order] this/that boy ten-TA ‘these/those ten boys’ (V) Dem NP Num-Ta (Demonstratives with NP-fronted order) Irrespective of the status of the numeral (i.e. high or low), demonstratives can be used for definiteness. While the NP-fronting is precluded in numerals higher than 5, numeral lower than 5 allow both canonical and non-canonical word orders in the presence of a demonstrative, as in (55a & b). The distinction between the (a) and (b) examples in (55) concerns maximality (Dayal 2014) or inclusiveness (Syed 2015), as discussed later. (55) a. ei/oi/Sei 7 du-To chele [Dem with Num ≤ 5 & canonical word order] this/that two-TA boy ‘These/those two boys’ b. ei/oi/Sei chele du-To [Dem with Num ≤ 5 & non-canonical word order] this/that boy two-TA 7 Bangla has three demonstratives – ei, oi and Sei. While ei is proximal demonstrative and oi is the distal demonstrative, Sei is an anaphoric demonstrative that refers back to a previous discourse. oi and Sei are used interchangeably in most contexts. Here I refer to both as ‘that’. 52 ‘These/those two boys’ 3.2.2 Section summary We have seen that Bangla has five ways of expressing definiteness. Numerically quantified DPs have different distribution than their non-numerical counterparts. In the absence of a numeral, a DP can be represented in two forms: the bare noun and the bare classifier. Three forms are found in the numerically quantified DPs. Numerals lower than 5 allows a non-canonical NP-fronted word order as well as definiteness by demonstratives. Demonstratives are the only way to express definiteness in DPs where the numeral is higher than 5. It is cross-linguistically attested that if a language has two ways of expressing definiteness, one form is preferred for a subset of contexts, while the other complements it in its selection of contexts. Schwarz (2009, 2013) shows for two sets of unrelated languages that two different forms are used for two types of contexts. The “strong” or more marked form of definite article is used in anaphoric contexts (e.g. German/Fering – strong article), while the less marked form is used for unique contexts. If a language has one overt determiner and a bare noun form for expressing definiteness (e.g. Akan), the overt determiners are used in anaphoric contexts; elsewhere a bare noun can be used. Simpson (2005) and Simpson et al. (2011) also observe similar pattern in a set of East Asian classifier languages where a bare classifier and a bare noun form is used for similar contexts. Since Bangla has five forms for definiteness construal, a detailed investigation involving different contexts must be in order. In the following section, I adopt and adapt the contexts presented in Simpson and Biswas (2016) as diagnostic tests for Bangla definiteness for exploring these definite forms in Bangla. Licensing contexts for Bangla definite forms 3.3 In this section, I present the set of contexts that I use to investigate Bangla definite forms. These contexts are carefully constructed to cover all contexts of definite occurrences, starting from the most salient anaphoric contexts to non-anaphoric unique occurrences. Different versions of these contexts have been tested in multiple languages (See Simpson 2005, Simpson et al. 2011 for 53 Cantonese, Mandarin, Vietnamese, Bangla and Hmong; see also Schwarz 2009 for similar contexts and applications in another set of languages). Here I present the range of contexts in the order of salience: from anaphoric most salient to non-anaphoric least salient inferred unique cases. Context A: Discourse-Anaphoric definite reference Anaphoric definite reference is established via an overt anaphoric link to the referent in the existing/previous context. The definite description refers back to an overt linguistic expression mentioned in the previous discourse/sentence (Christophersen 1939, Heim 1982, Kamp 1981). A marked form of definite is generally used in this context (cf. Schwarz 2009). (56) John has a dog. The dog is an Alsatian. Context B: Effects of visibility and pointing to a referent Where the referent is clearly visible to the interlocutors, and speaker draws attention to the presence of the referent, for example by pointing, or directing the hearer to look at the referent. A marked form of definite is generally used for this context. (57) Context: In a zoo, John points at a bear. Look! The bear is watching us. Context C: Bridging cross-reference These are the cases of definiteness where the establishment of an entity/individual as definite is by virtue of a link to some other entity/group referred to in the discourse (also called “inferable” in Prince 1981, or definiteness “via association” in Lyons 1999). Schwarz (2009) distinguishes two classes of bridging usage that relate to the context differently: (a) “part-whole bridging” (e.g., crisper-fridge, kitchen-house, trunk-car etc.), and (b) “producer-product bridging” (e.g., author-book, painter-painting etc.). The context has an indirect connection to the discourse. The antecedent in the context is not the referent itself, but has an indirect salient relation with the definite. For example, ‘author’ is related to ‘a book’ in a salient way, but the referent of the ‘book’ is not anaphorically present in the discourse. Cross-linguistically, a weak article is used in 54 the ‘part-whole’ bridging, while a strong article is used in the ‘producer-product’ bridging. Examples corresponding to the two cases, reflecting the article use in German, are given below (from Schwarz 2009, ex. 58-59, pp. 52-53). (58) a. The fridge was so big that the pumpkin could be stowed in the {im weak /#in dem strong } crisper without a problem. b. The play displeased the critic so much that he tore the {#am weak /an dem strong } author to pieces in his review. Context D: Non-anaphoric salient uniqueness As opposed to the previous cases, these are the instances of non-anaphoric salient definiteness construal. The referent may not be familiar to the speaker (e.g. inference unique cases). But if familiar, the referent is not provided by means of the discourse or immediate context. Languages mostly use he unmarked or bare form of definiteness for this context. Context D.1: Global uniques A salient non-anaphoric referent, that is unique for everyone. E.g. the Sun, the Earth. Context D.2: Situational uniques These are the instances where the referent is the only instantiation of a particular type of entity (picked out by a noun) in a certain situation, and it is familiar to both the speaker and the hearer. These are neither globally unique nor inferred to exist in the situation. In such cases, the interlocutors have contextual knowledge that the referent exists and is unique in the situation, but not directly mentioned in the context. As in the following example (Simpson & Biswas 2016, ex. 22), where the speaker and the hearer know that someone present in the context of utterance has brought along a camera, and this is the only camera that is present with the speaker and hearer(s). (59) Where’s the camera? 55 Context D.3: Inferenced uniques These are the cases where the speaker makes an inference that there is a unique entity of a certain type in a particular situation. These may seem similar to the bridging cross-reference cases, however no link to the antecedent is present in the discourse. The bridging cases have an overt antecedent in the context: a group or an entity that is indirectly linked to the definite. The inference cases do not have an overt antecedent. The situation triggers the inference. It is possible that the hearer is not familiar with the referent either. Consider the following example (Simpson & Biswas 2016, ex. 22), ‘the entrance’ is entirely unique in the context. It is inferred that a building has an entrance, but the ‘building’ is not overtly present in the previous discourse. (60) Where’s the entrance? (Asked by a speaker approaching an unfamiliar building) It is a cross-linguistically attested fact that the most marked forms of definites occur in the anaphoric definite contexts. Unmarked forms are often preferred in contextually non-anaphoric ‘unique’ cases. In the following section, I discuss each context, adapted to fit the conditions of the definite forms in Bangla (e.g. numerals higher or lower than 5), to check availability of each form in such context, and how Bangla definiteness could be to definiteness observed cross- linguistically. Bangla definite forms in licensing contexts 3.4 3.4.1 Context A: Discourse-anaphoric definite reference Bare Nouns: Bare nouns in Bangla do not generally express a definite description in an anaphoric context, unless there are other constraints overriding the canonical choices. For example, human noun referents tend to be expressed by bare forms, even in a discourse- anaphoric context. The following example shows that ‘woman’ is more prone to a bare noun pattern than ‘boy’. 56 (61) jOn kalke Ek-jon 8 mohila ar tin-Te chele-ke dekheche. John yesterday 1-CL lady and 3-TA boy-ACC saw ‘John saw a woman and three boys yesterday.’ mohila(?Ta/i) Sari poRechilen, ar chele-*(gulo) dhuti poRechilo. lady-TA/EMPH saree wore-HON, and boy-GULO dhuti wore ‘The woman was wearing a saree, and the boys were wearing dhutis.’ The selection may be influenced by socio-linguistic factors. For example, the referents of human nouns that are generally considered highly respected strongly prefer a bare noun pattern. If due to negative attitude or other socio-linguistic factors, the respect is suspended, a bare noun pattern may not be used. This is a strong constraint. In (62), two nouns (i.e., president and janitor), which are generally subjected to different levels of respect, represent definiteness by selection of different forms. In an anaphoric context where a bare noun form is not licensed, the noun with high respect (i.e. ‘president’) appear in the bare form whereas the other (i.e., ‘janitor’ – low respect), as expected, is in the bare classifier pattern. (62) Ofis-e Ek-jon 9 notun presiDenT ar Ek-jon mEthor niyog kOra hoyeche. office 1-CL new president and 1-CL janitor employ do was ‘The office hired a new president and a janitor.’ presiDenT(*TA) bOyoSko manuS, tObe mEthor-Ta-r bOyeS kOm. president-TA elderly person but janitor-TA-GEN age less ‘The president is an elderly person, but the janitor is young.’ Similarly, a negative-quality predicate may cause to disallow a bare noun pattern. As in (63), the negative-quality predicate (i.e. ‘being lazy’) with a referent that is generally respected (e.g. ‘officer’) licenses the bare classifier pattern, although a bare noun pattern is expected in this 8 -jon is a particle that generally co-occurs with numerals when the complement human noun is accorded respect. It is often considered a human classifier. However, unlike –Ta, it never co-occurs with nouns in a ‘bare classifier’ form (e.g., *chele-jon ‘boy-CL’). Thus, I do not endorse the view that it is a human-classifier. 57 context. In contrast, a referent without respect may observe a bare noun pattern is combined with a positive quality predicate. These examples illustrate how the speaker’s perception of a human referent in a particular context may influence and override a default selection of definiteness pattern. (63) 1990 Sale Ofis-e Ek jon Ofisar niyog kOra-hoechilo. 1990 year office-LOC 1 CL officer employ -do-be ‘The office hired an officer in the year 1990.’ kintu officer *(Ta) khub kuMRe, tai kono unnoti korte-pareni. but officer-*(CL) very lazy, so NPI promotion do-be-able-NEG ‘But the officer could not get promotion because he was very lazy.’ Simpson et al. (2011) conducted an investigation in Bangla involving the selection and rating of the two definite forms – NP-Ta and the bare noun, to find out what Bangla speakers prefer in a given context. From the average of the acceptability ratings of each form, it is evident that Bangla native speakers prefer a bare classifier (NP-Ta) pattern in the discourse-anaphoric context (4.81/5.00). A relatively lower average rating (3.84/5.00) with large standard deviation for the bare noun suggests that a bare noun is not generally preferred in these contexts. Simpson et al. (2011) also suggests that the higher variability in the bare noun form might be caused by other constraints influencing the choice of the form. Bare classifier: In Bangla, the bare classifier forms (e.g. NP-Ta and NP-gulo) are highly preferred in an anaphoric context. (64) jOn kalke Ek-Ta kukur ar tin-Te beRal kineche. John y.day 1-TA dog and 3-TA cat bought ‘Yesterday John bought a dog and three cats.’ kukur-*(Ta) ar beRal-*(gulo) Ekhono kichu khayni. dog-TA and cat-GULO till-now some eat-NEG 58 ‘The dog and the cats have not eaten anything yet.’ Numerically quantified DPs: If the referent is numerically quantified and if the numeral is lower than 5, the NP-fronted (NP Num-Ta) form is preferred in the anaphoric context. Akin to the bare classifier pattern, this word order is strongly preferred in anaphoric contexts. Below I show that this word order is mandatory in the anaphoric contexts. (65) jOn tin-Te boi kineche, boi tin-Te baNla-e lekha. John three-CL book bought, book three-TA Bangla-LOC written ‘John bought three books yesterday, the three books are written in Bangla.’ In addition, the demonstrative may be used to convey definiteness, in both the canonical and non-canonical word order in the complement NP. The subtle difference in their meaning relates to the availability of maximal reading of the definite referent (to be elaborated later). (66) jOn boimEla theke tin-Te boi kineche, John bookfair from 3-CL book bought, ‘John bought three books from the bookfair.’ a. Sei tin-Te boi baNla-e lekha. (Dem + Num-Ta NP - default) that 3-TA book Bangla-LOC written ‘Those three books are written in Bangla.’ b. Sei boi tin-Te baNla-e lekha. (Dem + NP Num-Ta) that book 3-TA Bangla-LOC written ‘Those three books are written in Bangla.’ Summarizing, we see that the discourse-anaphoric context prefer the marked forms of definiteness. The unmarked or the bare noun is only allowed when some socio- or meta- linguistic factors override the default choice. The following is a schema of this finding. Results of the extra-linguistic factors are marked within parenthesis indicating their natural choice. 59 Contexts N NP-Ta/-gulo NP Num-Ta Dem Num-TA NP Dem NP Num-Ta Discourse-anaphoric (✗) ✓ ✓ ✓ ✓ Table 5: Bangla definite descriptions in discourse-anaphoric contexts 3.4.2 Context B: Effects of visibility and pointing to a referent The visibility and deixis require the referent to be present in the physical discourse. When the referent is clearly visible to the speaker and hearer, and the speaker draws attention to the presence of the referent by pointing or directing the hearer to look at the referent, the bare classifier pattern is highly preferred. Similar to the discourse-antecedent context, the referent must be present in the discourse. Specifically, it has to be physically present. Bare nouns are allowed in this context when strong constraints override the bare classifier pattern, for example, the ‘president’ in (67d). (67) Situation: Ramu and his son are repairing a bicycle. Ramu points to a hammer and says: a. amake hatuRi-*(Ta) de-to. I-ACC hammer-TA give-TOP ‘Pass me the hammer.’ b. Situation: At a zoo. Ramu says to his son: oi dEkh! baMdor-*(gulo) amader dekhche. Oh look monkey-GULO us watching ‘Look over there! The monkeys are watching us.’ c. Situation: In a bank, during a robbery. Ramu says to his son: puliS-*{Ta/gulo}-ke dEkh! {or/oder} kache bonduk ache! police-TA/GULO-ACC see his/them near gun has ‘Look at the {policeman/policemen}. {He/they} has/have a guns!’ 60 d. Situation: In a rally for the president. Ramu says to his son: presidenT-(*-Ta)-ke dEkh! Ki Sundor boktrita dicchen! President-TA -ACC see what beautiful speech giving-HON ‘Look at the president. S/he is delivering such a beautiful speech!’ As expected for the immediate presence of the referent, the NP-fronted order is allowed in this context. Similarly, demonstratives are naturally acceptable due to the indexical nature of this context. (68) Situation: during a robbery, Ramu points out to his son: chele du-To-ke dekh! oder kache bonduk ache. boy two-TA-ACC see they near gun is ‘Look over there, the two boys! They have guns.’ (69) oi chO-Ta chele-ke dekh! oder kache bonduk ache. that 6-TA boy-ACC see they near gun is ‘Look over there, the six boys! They have guns.’ (70) a. oi chele du-To-ke dekh! oder kache bonduk ache. that boy 2-TA-ACC see they near gun is ‘Look over there, those two boys! They have guns.’ b. oi du-To chele-ke dekh! oder kache bonduk ache. that 2-TA boy-ACC see they near gun is ‘Look over there, those two boys! They have guns.’ The following presents a schema of the available definite forms in the two contexts discussed so far. The contexts have one similarity: both require the antecedent to be present in the discourse, either in the actual physical discourse or in the context. We see that the distribution of the definite forms is the same in these two contexts: the bare nouns are not generally allowed. 61 Contexts N NP-Ta/-gulo NP Num-Ta Dem Num-TA NP Dem NP Num-Ta Discourse-anaphoric (✗) ✓ ✓ ✓ ✓ Visibility & pointing (✗) ✓ ✓ ✓ ✓ Table 6: Bangla definite descriptions in discourse-anaphoric and visible/pointing contexts 3.4.3 Context C: Bridging cross-reference Recall that the bridging cases provide an indirect link between the definite and the antecedent in the discourse. The bare classifier pattern is preferred in Bangla for the bridging cases. (71) a. jOhn Ek-Ta biSal frij kineche. John one-TA gigantic refrigerator bought doRja-(Ta) du hat-e khulte hOe. Door-TA two hand-LOC open be ‘John bought a gigantic refrigerator. The door needs to be opened by two hands.’ b. ami Ek-Ta gaRi kinechi. caka-*(gulo) moTa-moTa. I one-TA car bought wheel-GULO fat-fat ‘I bought a car. The wheels are wide.’ If a human referent is [+resptected] and contextually presented in a positive way, a bare noun pattern is preferred. For example, ‘the director’ is considered a human referent to whom respect is generally accorded, and it is presented in a positive context, resulting in the selection of a bare noun pattern. (72) Ek-Ta naTok Ekhon khub bikkhEto hoyeche. one-Ta play now very popular became 62 poricalok-(*Ta) Eto-Ta aSa kOren-ni. director-TA so-much hope did-NEG ‘A play has been very popular. The director didn’t expect so much (popularity).’ Just as we have seen in the discourse-anaphoric cases, the bare noun pattern is not preferred in the bridging cross-reference context unless there are other constraints that override the selection of the bare classifier form. Numerically quantified DPs: the numerically quantified DPs where the numeral is lower than 5, the NP-fronted pattern is allowed in the bridging cross-reference cases. The interlocutors must be aware of the numerals used in the definite description. For example, if the context refers to the Taj Mahal, one of the Seven Wonders of the World, the speaker and the hearer must know that it has four towers associated with it. (73) jOn or baik-Ta bikri kore-dicche, caka du-To bheNe-geche John his bike-TA sell do-give wheel 2-TA break-gone ‘John is selling his bike. The two wheels are broken.’ (74) jOn Tajmohol dekhte gechilo. bollo minar car-Te khub uMcu. John Taj Mahal visit-PPL went said tower 4-TA very tall ‘John went to visit the Taj Mahal. He said the four towers are very tall.’ For numerals higher than 5, the use of the demonstratives is not preferred in this context. For example, if a cube is mentioned in the context, and even if the interlocutors know that it has 6 faces/sides, the use of demonstrative is not preferred. Definiteness is construed by the use of a possessor in this context. (75) a. jOn Ek-Ta kaTher kiub baniyeche. *Sei chO-Ta dik ebro-khebro. John 1-Ta wooden cube made. DEM 6-TA side uneven ‘John has made a wooden cube. The six faces are uneven.’ 63 b. jOn Ek-Ta kaTher kiub baniyeche. SeTa-r chO-Ta dik ebro-khebro. John 1-Ta wooden cube made. That-GEN 6-TA side uneven ‘John has made a wooden cube. Its six faces/sides are uneven.’ Demonstratives are not preferred even when the numerals are lower than 5. (76) a. jOn Tajmohol dekhte gechilo. *bollo Sei car-Te minar khub uMcu. John Taj Mahal visit-PPL went said that 4-TA tower very tall ‘John went to visit the Taj Mahal. He said the four towers are very tall.’ b. jOn Tajmohol dekhte gechilo. *bollo Sei minar car-Te khub uMcu. John Taj Mahal visit-PPL went said that tower 4-TA very tall ‘John went to visit the Taj Mahal. He said the four towers are very tall.’ The bridging cross-reference provides an anaphoric link between the antecedent and the definite. Note, however, the link is indirect and the antecedent must not be present in the discourse. The indexical nature of the demonstratives is thus not compatible with the context. Therefore, we see the definite strategies involving demonstratives are not acceptable in this context. The following is a schema of available forms for the contexts we have discussed so far. Contexts N NP-Ta/-gulo NP Num-Ta Dem Num-TA NP Dem NP Num-Ta Discourse-anaphoric (✗) ✓ ✓ ✓ ✓ Visibility & pointing (✗) ✓ ✓ ✓ ✓ Bridging cross-ref. (✗) ✓ ✓ ✗ ✗ Table 7: Bangla definite descriptions in discourse-anaphoric, visible/pointing and bridging contexts These three contexts (i.e. discourse-anaphoric reference, bridging cross-reference and visibility and pointing) strongly prefer the bare classifier pattern. In all these three cases, the referent is highly salient and present in the discourse. This is in line with the cross-linguistic observation that the marked form of the definite utilizes the most salient form of the definite reference. The 64 contexts that we are going to discuss next do not require the referent to be present in the context. We will see that the unmarked form is preferred in these contexts. 3.4.4 Context D: Non-anaphoric salient uniqueness 3.4.4.1 Context D.1 - Global uniques The global uniques are the cases where the referent is unique for everyone (e.g., the sun, the moon etc.). These are generally expressed in the bare noun form. However, in the contexts where they are anchored in a particular point in time, or are used in episodic predicates, the bare classifier form is used. (77) Situation: out of the blue, speaker says to hearer: a. caMd-(*Ta) prithibi-r upogroho. moon earth-GEN satellite ‘The moon is the satellite of the earth.’ b. gOto-rate caMd-(Ta) jhOlmOl korchilo. last night-LOC moon-TA bright do ‘The moon was very bright last night.’ c. ajke khub meghla. Surjo(-Ta) kothay? today very cloudy sun-TA where ‘It’s very cloudy today. Where’s the sun?’ (Simpson & Biswas 2016, ex. 24-26) Before we apply the tests for the salient non-anaphoric uniqueness contexts for the marked definite forms in Bangla, a possible question must be addressed that concerns number marking and uniqueness. Since these constructions are naturally plural, whether these are available in unique contexts concerns their number representations. Unique referents are generally thought to have a singular entity. This, however, does not have to be the case. The treatment of uniqueness in singulars can also be extended to plural entities (Sharvy 1980, Coppock & Beaver 2012). A 65 plural entity may be unique in the context, albeit the instantiation must be the maximal set satisfying the descriptions. Thus, only plural definites representing the maximal set may be unique (Dayal 2014). To consider the applicability of numerically quantified DPs in this context, I consider entities that are generally known to be plural, or required to be multiple in the context. For example, it is known that Mars has two satellites, and we generally have more that one battery (for the searchlight, for example) while hiking, or have two pilots in airplanes. (78) Situation: out of the blue, speaker says to hearer: moNol-er upogroho du-To ajke poriSkar dekha jacche. Mars-GEN satellite two-TA today clearly see go ‘Mars’ two satellites are clearly visible tonight.’ Demonstratives in the global unique context gives rise to an indexical interpretation, hence it is incompatible in this context. And often in the absence of such context or relevant indexing, the construction is ungrammatical. (79) Situation: in broad daylight, speaker says to hearer: a. *moNol-er oi upogroho du-To ajke poriSkar dekha jacche. Mars-GEN DEM satellite two-TA today clearly see go b. *moNol-er oi du-To upogroho ajke poriSkar dekha jacche. Mars-GEN DEM two-TA satellite today clearly see go Contexts N NP-Ta/-gulo NP Num-Ta Dem Num-TA NP Dem NP Num-Ta Discourse-anaphoric (✗) ✓ ✓ ✓ ✓ Visibility & pointing (✗) ✓ ✓ ✓ ✓ Bridging cross-ref. (✗) ✓ ✓ ✗ ✗ Global unique ✓ (✓) (✓) ✗ ✗ Table 8: Bangla definite descriptions in anaphoric and global unique contexts 66 3.4.4.2 Context D.2: Situational uniques Situational uniques are instances where speaker and hearer are aware that there is a unique referent of a certain type in a particular situation, and this is not licensed by bridging cross- reference (via the presence of a linguistic link) or visibility/pointing (which automatically causes the bare classifier pattern to occur). Purely situational uniques are frequently represented with the bare noun pattern. The bare nouns are most natural, however, the bare classifier pattern is acceptable too. (80) Ramu and Mina are walking in a dark wood. Ramu says: tOrc-(Ta) kothay? ami oTa tomar bEkpEk-e rekhechilam. flashlight-TA where I that your backpack-LOC kept ‘Where’s the flashlight? I kept it in your backpack.’ The numerically quantified DPs are also allowed in this context if the interlocutors are aware of the referent of the DP, and the referent is activated in the mind of the interlocutors. (81) A plane has crashed into a field. A team of rescuers arrives at the scene and sees that the front of the plane is still intact. They are concerned about the pilots, who they cannot see but know that there were two pilots in that plane. One rescuer shouts: ei, pilot (du-To) kothae? ami-to kau-ke dekh-te pacchi na! hey pilot two-TA where I-TOP nobody-ACC see-PPL can NEG ‘Hey, where are the two pilots? I can’t see anyone!’ Demonstrative assisted definiteness in numerically quantified DPs are available in this context only if the speaker and the hearer the unique referent in the particular context. (82) Situation - While working in a dark shed, Ramu searches their toolkit for the flashlight. He says: 67 a. tOrc du-To kothay? flashlight two-TA where ‘Where are the two flashlights?’ b. ?oi tOrc du-To kothay? that flashlight 2-TA where ‘Where are those two flashlights?’ c. *oi du-To tOrc kothay? that 2-TA flashlight where ‘Where are those two flashlights?’ d. tOrc kothay? flashlight where ‘Where are the flashlights?’ Contexts N NP-Ta/-gulo NP Num-Ta Dem Num-TA NP Dem NP Num-Ta Discourse-anaphoric (✗) ✓ ✓ ✓ ✓ Visibility & pointing (✗) ✓ ✓ ✓ ✓ Bridging cross-ref. (✗) ✓ ✓ ✗ ✗ Global unique ✓ (✓) (✓) ✗ ✗ Situational unique ✓ (✓) (✓) ✗ ✓ Table 9: Bangla definite descriptions in anaphoric and unique contexts The examples discussed in this section illustrates that the bare classifier form is still available even if the referential connection between the referent and the definite is not straightforward. Both the bare noun and the bare classifier forms are available and may be in free variation in such cases. The availability of the demonstrative-assisted definiteness, along with the free variation of the bare noun and the bare classifier forms in the situational unique cases seem surprising. Simpson & Biswas (2016) argues that the situational unique cases are the cases where the referent of the definite is activated in the mind of the speaker and the hearer. This is in 68 parallel to the use of marked cases in the anaphoric contexts. The activation (aka the presence of the referent in the mental context) somehow requires the use of the marked form. 3.4.4.3 Context D.3: Inferenced uniques When the speaker infers the existence of a unique referent from the context that is previously unfamiliar to the speaker, a bare noun pattern is mostly preferred. Inferenced uniques are thus very similar to the situational uniques: the referents are not present in the discourse in both cases. However, the referent of the situational unique may be familiar to the interlocutors, which may not be the case with the inferenced uniques. Note that this context does not satisfy the familiarity constraint of definite descriptions: the referent may or may not be familiar to both the interlocutors. These are the cases where the referent is inferred in the context. The definite has no (in)direct link to the context (unlike the discourse-anaphoric or bridging cases). The referent is not pointed out by pointing/looking at it. These contexts are created deliberately restricting the inference on non-linguistic evidence. Consider the following examples from Bangla showing definite uses of the bare noun when the referent is uniquely identified. In (83a), the door has to be a unique door, or a door identifiable to the speaker and the hearer (for example, the speaker is standing in front of the door). For the bare noun in (83b), the dog must be a unique dog, known to the interlocutors. The referents in these cases are not identifiable by the context. (83) a. Asked by the speaker standing in front of the door. Mina, dOrja bOndho, cabi kothae? Mina, door closed, key where ‘Mina, the door is locked, where is the key?’ (Simpson et al. 2011, ex. 16c) b. Situation: The interlocutors hear their pet dog barking: kukur bagan-e Dakche. dog garden-loc is-barking ‘The dog is barking in the garden.’ #‘Some dog(s) is/are barking in the garden.’ 69 This intuition is clearly expressed in the following context that restricts the available information to the context. The use of the bare classifier form implies that the referent is identifiable to the speaker, and hence is not compatible to this context. (84) Situation: Everyday on planet Varg, a security vehicle delivers one new prisoner to guards from the camp. Today the vehicle drives up, the back doors open out and no one is inside. The guard says: kOedi-(*Ta) kothae? prisoner where ‘Where is the prisoner?’ (Simpson & Biswas 2016, ex. 27) Numerically quantified noun phrases are not compatible in this context. Only a bare noun form is grammatical in this context. (85) Situation: While walking in a dark wood, Ramu searches Mina’s backpack. He says: a. *tOrc du-To kothay? ogulo tomar bEkpEk-e rekhechilam na? flashlight two-TA where those your backpack-LOC kept NEG ‘Where are the two flashlights? Didn’t I keep those in your backpack?’ b. *oi tOrc tin-Te kothay? ogulo tomar bEkpEk-e rekhechilam na? that flashlight 3-TA where those your backpack-LOC kept NEG ‘Where are those three flashlights? Didn’t I keep those in your backpack?’ c. *oi tin-Te tOrc kothay? ogulo tomar bEkpEk-e rekhechilam na? that 3-TA flashlight where those your backpack-LOC kept NEG ‘Where are those three flashlights? Didn’t I keep those in your backpack?’ d. tOrc kothay? tomar bEkpEk-e rekhechilam na? flashlight where your backpack-LOC kept NEG 70 ‘Where are the flashlights? Didn’t I keep those in your backpack?’ These examples illustrate that a bare noun pattern is licensed when the referent is not anaphoric in the context. The bare noun may be allowed when the referent is not familiar to the interlocutors. Contexts N NP-Ta/-gulo NP Num-Ta Dem Num-TA NP Dem NP Num-Ta Discourse-anaphoric (✗) ✓ ✓ ✓ ✓ Visibility & pointing (✗) ✓ ✓ ✓ ✓ Bridging cross-ref. (✗) ✓ ✓ ✗ ✗ Global unique ✓ (✓) (✓) ✗ ✗ Situational unique ✓ (✓) (✓) ✗ ✓ Inferenced unique ✓ ✗ ✗ ✗ ✗ Table 10: Bangla definite descriptions in anaphoric and non-anaphoric contexts 3.4.5 Summary of observations Five different morpho-syntactic patterns represent definiteness in Bangla. Six different contexts ranging from anaphoric definite reference to salient non-anaphoric uniqueness show that the distribution of the bare noun forms pattern differently from the rest of the forms. Bare nouns are preferred in non-anaphoric unique contexts. The bare classifier form prefers the anaphoric salient referent, albeit it may also appear in salient unique contexts. The other three forms of the numerically quantified noun phrases are the tokens of the same grammatical configuration as the bare classifier form. Their distribution is similar to the distribution of the bare classifier form: these are preferred in anaphoric definite reference, where the referent is visible or salient by means of pointing. And just as the bare classifier form, these are also applicable to certain unique contexts, such as the situational unique where the referent is activated in the mind of the interlocutors. In Simpson & Biswas (2016) we point to a novel aspect of definiteness that explains the selection of forms in unique cases. We argue that the uniqueness and identifiability are the base 71 conditions for the selection of the forms. If a base condition is satisfied, the respective relevant form is most likely to occur. However, if the situation is such that a referent is both identifiable and unique, either form is in principle licensed, and an additional constraint decides the selection of the form. We argue that the level of activation of the mental representation of the referent plays an important role in this case. Where there is a high level of activation, the use of a bare classifier form naturally occurs, whereas a low level of activation will cause the use of a bare noun pattern. Instances of anaphoric definite reference, pointing to a referent and bridging cross- reference are all contexts in which there is high activation of the mental representation of an entity/individual, and consequently result in very strong pressure to use the bare classifier pattern. The activation level of a referent is therefore a property that has very significant effects on the representation of definite individuals and entities in Bangla, and is one which is formally independent of the uniqueness and the identifiability of the referent. It characterizes the current state of referents whose uniqueness and identifiability is otherwise determined, and does not alter such properties. As per the cross-linguistic representation of definiteness in languages with more than one definite determiner, the Bangla marked forms are used in anaphoric contexts, while the non- marked forms are used for unique contexts. Similar alternation between a bare noun and a bare classifier form is common in East Asian classifier languages, and it depends on the stages of grammaticalization the language has undergone in terms of expressing definiteness by means of the bare classifier form (Simpson 2005, Simpson et al. 2011). Our discussion in this section suggests that the referential complexity represented by the form selection is shared underlyingly across languages and is represented by different surface structural ways. In the following section, I present the existing account of Bangla definiteness, followed by my proposal for definiteness in Bangla. Existing theories of Bangla definiteness 3.5 3.5.1 Specificity vs. Definiteness Dasgupta (1983) analyzes –Ta as a classifier and proposes that classifiers can be markers of specificity. Later work (Dasgupta 1985, Dasgupta & Bhattacharya 1993) deliberates between 72 definiteness and specificity associated with this word order. Dasgupta (1983) discussed NP-Ta form and associated it to a definite interpretation. Bhattacharya (1999 and subsequent work) provides a major syntactic account of the Bangla DP. He discusses the numerically quantified DPs and suggests that the non-canonical word order has a specific reading. In this feature-driven account, an optional, strong [Specific] feature on the noun regulates the word order variations. If the feature is strong, it motivates the movement of the NP to the Spec, QP while the weak feature yields an indefinite reading in the numerical constructions. The following tree structure shows the structure, the base position and the movement site of the NP as per Bhattacharya (1999). (86) a. tin-Te [lal bOl] three-TA red ball ‘three red balls.’ b. [lal bOl] tin-Te red ball three-TA ‘the three red balls.’ (87) The non-canonical order in the numerically quantifier DPs (NP Num-Ta) is also the derivation for the bare classifier forms (NP-Ta). The NP moves to the front, landing either in the specifier of a QP (Bhattacharya 1999 and subsequent work) or the DP (Dasgupta 1983). Bhattacharya refers to this movement as the NP-“object”-shift, owing to the similarity of this movement to that of the VP-fronting. Tanmoy Bhattacharya 4 Num-Cla as well. 2.2. Q-Float Further evidence that the middle head in the Bangla DP is indeed a quantifier head can be found from investigating the phenomenon of Q-float. The Num-Cla complex shows a Q-float like effect, Q/Num-Cla sequences are, therefore, to be thought of as belonging to the domain of QP. This is evident further from their behaviour as Floating Quantifiers (FQ). Miyagawa (1988) assumes that the Numeral-Quantifier must be in a relation of mutual c-command with the DP it quantifies over at D-structure. He observed that a numeral-quantifier occurring to the right of the DP it modifies could be dislocated from it if the DP is a subject of an unaccusative or a passive verb but that the DO may not intervene between the transitive subject and a Num-Q. This holds for Bangla too: (7)a. chatro i aj tin-Te t i eSechilo studentto day 3-CLA came ‘three students came today’ b.* chatro boi tin-Te enechilo student book 3-CLA bought ‘three students had bought books today’ c. gaRi i cor dara car-Te t i curi gEche car thief by 4-CLA theft gone-PASS ‘four cars were stolen by thief’ These examples prove beyond doubt that the Num-Cla constituents in Bangla are like FQs and are therefore by definition, Qs. 3. Difference between all and non-all quantifiers (NAQ) If the Num and the Cla are part of a single complex head then the following is a plausible structure: (8) QP 2 NP Q’ 2 Q [specific] NP 2 Q Cla [specific] tin ‘three’ -Ta kichu ‘some’ -Ta In the rest of the paper, I discuss this structure in detail. First, notice the behaviour of the Q SOb ‘all’ in the following pair: (9)a. SOb gulo chele aSbe 73 In recent works, Chacón (2011) claims that this word order involves definiteness (after Dasgupta 1983, see also Ghosh 2010, Simpson et al. 2011, and Dayal 2012). He argues that the NP moves to the Spec, DP to check a strong [Def] feature. Following Simpson (2005) he argues against a combined head analysis of Num and Cl (cf. Bhattacharya 1999). In the numerically quantified DPs (NP Num-Ta) the numeral and the classifier, as separate heads, project their own functional projections. The NP moves past the NumP and ClP to the specifier of DP. In NP-Ta, he assumes (following Dasgupta 1983), the numeral is ‘one’, the definite form [NP-Ta] undergoes ‘one’ deletion to yield the definite singular interpretation. The following structure is the proposed structure for the definiteness in Bangla where the NP moves to the Spec, DP (cf. Chacón 2011). (88) Dayal (2012) sides with Chacón in assuming definiteness, for the non-canonical word order in numerically quantified Bangla DPs. To rule out the specific indefinite reading, Dayal further argues that intermediate scope variability is not observed in the marked word orders. They do not enter the scope relations. In the following set, the NP-raised word order (a example) can only mean that every student had to study the exact same subjects. The indefinite (b) example have intermediate readings. (89) a. prottek chatro biSOe du-To-te Sob pepar poRlo. every student subject two-TA-LOC all paper read DP ei D’ ei D 0 NumP ei Num ClP Tin ‘three’ ei Cl 0 NP -TA [ NP lal bOl] ‘red ball’ 74 ‘Every student read all papers on the two subjects.’ = [there are two subjects] such that [every student] read [all papers] in the two subjects. = *[for every student] [there are two subjects] such that he read [all papers]. = *[for every student] he read [all papers] in [two subjects]. b. prottek chatro du-To biSOe Sob pepar poRlo. every student two-TA subject-LOC all paper read ‘Every student read all papers on the two subjects.’ = *[there are two subjects] such that [every student] read [all papers] in those subjects. = [for every student] [there are two subjects] such that he read [all papers]. = [for every student] he read [all papers] in [two subjects]. The tests that hold for NP-Ta also hold for NP-gulo, making the two forms equivalent in terms of their definiteness construal (Biswas 2014b). Quantificational determiners cross-linguistically do not appear with definite DPs (except Matthewson 1992), and this test holds for both NP-Ta and NP-gulo, suggesting that NP-gulo is also a definite description, as shown below. (90) a. *{every/most/the/many/three} the boy b. all the boys (91) a. {prottek/beSirbhag/Onek} chele-(*gulo) each/most/many boy-GULO ‘every boy’ ‘most boys’ ‘many boys’ b. SOb chele-(gulo) 10 all boy- GULO ‘all (the) boys’ 10 The optionality is available due to the scopal nature of ‘all’. 75 3.5.2 Familiarity, anaphoricity & uniqueness Among multiple approaches to definiteness, familiarity and uniqueness are most discussed in the philosophical and formal linguistic literature. In the philosophical literature, the ‘familiarity condition’ is satisfied for a definite description when the speaker believes that the hearer is familiar with the individual satisfying the description's nominal or the description's denotation (Pupa 2010). The classic familiarity approach (Christophersen 1939, Heim 1982) to definite determiner holds when the definite description is anaphoric to a discourse referent present in the discourse context. Within the familiarity approach, Roberts (2003) contrasts strong familiarity, which usually involves explicit previous mention of the entity in question, with weak familiarity where the existence of the entity is entailed in the linguistic or non-linguistic context (e.g. perceptually accessed information). Dayal presents ‘familiarity’ as the key concept and the diagnostic of definiteness. The referent must be familiar to the speaker and the hearer for a definite reading, and this distinguishes the specific cases from the definite ones. The NP-raised order is familiar to the speaker and hearer, and thus rules out the specific readings in this word order. NP-Ta NP Num-Ta Num-Ta NP Familiar to speaker ✓ ✓ (✓) Familiar to hearer ✓ ✓ ✗ Interpretations: Definite(+Specific) Definite(+Specific) Indefinite(± specific) Table 2: Identifiability of Bangla definite descriptions This diagnostic supports our observation that the marked forms involve the knowledge of the speaker and the hearer, even in the cases of global unique and situational uniques. The discourse anaphoric referents are cases of strong familiarity, whereas the pointing or the visibility of the referent and the bridging cross-reference may be categorized under the weak familiarity. The classic uniqueness approach (Frege 1892, Russel 1905, Strawson 1950) holds for a definite determiner ‘the’ only if the property denoted by its complement is uniquely present to the speaker and the hearer. No reference to the prior discourse is required. Uniqueness is also satisfied by the criterion of anaphoricity, except that the anaphoric antecedent must be present in 76 the immediate discourse, while uniqueness involves maximal reference with respect to the current context. A definite description in an anaphoric situation is also unique; however, anaphoricity is not an absolute criterion for uniqueness. For example, the referent of global unique (e.g., “the Prime Minister”) matches the descriptive content of the definite description without any reference to the previous context. From our observations in the previous section we saw that the unmarked bare noun form are the most appropriate for a unique context. 3.5.3 Maximality and uniqueness presupposition Earlier I mentioned that Dayal (2102) considers NP-Ta as definite depending on the concept of familiarity. Along with this syntactic criterion, Dayal (2012, 2014) provides semantic extensions for definiteness in Bangla. The covert application of the iota maximizes over a context/situation or time interval yielding a definite reading. If it applies to a singleton set in a situation or context, a singular definite reading (with -Ta) is achieved. If it applies to a plural set in the situation or context, it achieves a plural definite reading (with -gulo). 11 That is, the definite entity must be unique and must represent a maximal set in the context for definite readings. For example, if there are only three objects a, b, and c in the denotation of the noun, chele-gulo ‘boy-GULO’ is maximal. By maximality, it contains only the largest sum {a+b+c} in its denotation. If NP-gulo is maximal and it contains only the largest sum {a+b+c} in its denotation, a distributive operator may not be compatible with the sentence. Brisson (1998) argues that ‘all’ is a modifier of distributive operator and if a sentence allows ‘all’ it must be distributive. Thus, ‘all’ removes ‘imprecision’ from the semantics of the distributive operator. This, for example, results into ‘all the boys’ being inappropriate in the following context, because ‘all’ + plural NP requires all ten boys to satisfy the predicate. See the following example (adapted from Brisson 1998). (92) Context: There are 10 boys, and only 9 of them jumped into the lake. a. The boys jumped into the lake. Except John, who was sick. b. #All the boys jumped into the lake. Except John, who was sick. 11 Jiang (2012: 406) proposes singular and plural iotas for singular and plural definite descriptions in French. Although this may be a feasible idea, I am not sure if that is necessary, given that the –Ta and –gulo determines the nature of the set. 77 Pragmatic weakening (i.e. when not every member takes part in the event denoted by the VP) is possible in the English definite plural (the boys) in the subject position, because it incorporates the non-maximal reading. That is, along with {a+b+c}, it includes {a+b}, {b+c} etc., leading (92a) to be true. It is not possible with ‘all the boys’ because ‘all’ does not tolerate any exception. She argues, following the idea of Schwarzschild (1993), that the distinction can be explained in terms of the ‘good-fitting’ and ‘ill-fitting’ covers. A Cover is a partition over atomic entities where no members overlap with each other. The function of all is to make sure that a good-fitting cover is assigned to the denotation of the definite DP, leading (92b) to be ill formed. Pragmatic weakening is not possible is Bangla. As shown below, all includes extra (somewhat redundant) information regarding the reference of the definite plural chele-gulo [boy- GULO]. The context in (92) does not allow boy-GULO in the subject position of the sentence. All ten boys are required for the sentence to be true. In that case, ‘all’ is redundant; leading to support the claim that boy-GULO only has the largest sum in its denotation. (93) Context: There are 10 boys, and only 9 of them jumped into the lake. chele-gulo (SObai) jOle jhaMp dilo. boy-gulo (all) water-locs jump give ‘The boys (all) jumped into the lake.’ The concept of maximality explains a related reading in Bangla. Recall that in the numerals lower than 5, both the default and the non-canonical orders are possible. Dayal (2012) argues that the distinction between the word orders in the (94a) and (94b) involve maximality. The NP- raised version is only possible when the DP refers to “the full set of entities that the description applies to.” The canonical order or the base order can be used to “pick out a subset of a larger group of entities to which the description applies.” (94) a. ei/oi/Sei du-To chele [Default word order] this/that two-TA boy ‘These/those two boys’ 78 b. ei/oi/Sei chele du-To [Non-canonical word order] this/that boy two-TA ‘These/those two boys’ Syed (2016) argues that the distinction between the orders in (94a) and (94b) is that of inclusiveness. He acknowledges that the definition of Inclusiveness is similar to Maximality. Syed proposes that the non-canonical word order in Bangla is interpreted as definite by inclusiveness. Following Hawkins (1974), he defines inclusiveness as a property by which the hearer infers a shared set (by the speaker and the hearer) containing the definite referent as the only entity. It can be the definite singular (atomic entity) or the definite plural (the largest sum). Thus, according to Syed, the distinction between (94a) and (94b) involves definiteness: (94b) infers that there are only two boys in the set, whereas, there could be more boys in the set in (94a). Consider the following example from Syed (2016) (ex. 8). (95a) is infelicitous in this context due to failure of inclusiveness. The context specifies four shoes, whereas the NP-raised order is for three shoes. This violates inclusivity that predicts four shoes instead of three shoes. Only the non-raised indexical version (95b) is felicitous in this context as the denotation is inclusive (i.e., allow three shoes). (95) Context: Ram has bought four shoes. And then he returned all of them. Next day, while walking past the shoe store with Shyam, he notices three of the shoes that he returned on the showcase. He points to them, and tells Shyam: a. #ami oi juto tin-Te ferot diye diyechi. I that shoe three-TA return give-PPL gave b. ami oi tin-Te juto ferot diye diyechi. I that three-TA shoe return give-PPL gave ‘I have returned those three shoes.’ 79 The examples are minimally different by the word order (i.e. NP-fronting). It is important to tease apart the effect of the demonstrative from the definiteness achieved by the word order. Definite determiners and demonstratives share similar semantics. Wolter (2006) states that definite descriptions refer uniquely relative to the situation corresponding to the discourse context. Demonstrative descriptions refer uniquely relative to a salient situation distinct from the discourse context, and this is achieved by pointing or indexicality. Although they both presuppose uniqueness/maximality within the pragmatically restricted domains, the demonstrative additionally creates an immediate salient situation, by means of pointing gesture or visibility. Thus, the indexical function of the demonstrative in some way also represents definiteness in the canonical word order cases. Note that the contexts discussed earlier in this chapter, the visibility and pointing cases (96), the definiteness associated with these constructions is the same as the concerned examples here. (96) Ramu and his son are repairing a bicycle. Ramu points to a hammer and says: amake hatuRi-*(Ta) de-to. I-ACC hammer-TA give-TOP ‘Pass me the hammer.’ The question that concerns us here is the role of pointing, and its representation – overt (96) or covert (95), which makes the presence of the demonstrative optional, but, mandates the bare classifier pattern or the NP-raised order. Without going into the details of how demonstratives are associated with the definite interpretations, I would like to conjecture that there is more to maximality for explaining these contexts. The role of the demonstratives needs to be explained in detail. This optionality does not appear with the higher numerals, as the demonstrative is the only medium of expressing definiteness in that context. 3.5.4 Section summary This section provides a summary of previous literature for Bangla definiteness. We see that there are multiple ways of explaining definiteness in Bangla. Anaphoricity, identifiability and familiarity have been associated with the definiteness associated with overt morphological forms 80 of definiteness. Thus, the over forms of definiteness (as opposed to the bare noun definites) can be explained by the theories that requires the interlocutors to know the referent of the definite description, or anaphorically linked to the referent, either in previous sentential context or by visibility and pointing. The uniqueness-based approaches are applicable only to the bare noun patterns. Thus, the novel point that can be added to the existing documentation of Bangla definiteness is that Bangla definiteness must not always be a result of strong [+definite] feature resulting in the morpho-syntactic alternations. Definiteness by uniqueness does not involve any morphological alternations. From this discussion we clearly distinguish the two types of definiteness involving syntactic characteristics: the definite by anaphoricity or familiarity require morpho-syntactic support, possibly due to a strong definite feature associated with the D 0 . The definite by uniqueness does not require morpho-syntactic support. The latter does not involve a strong D-feature in the DP. In the next section, I follow up with the idea of two types of definiteness by proposing two syntactic positions for definiteness. Along the lines of these observations, I elaborate on the requirement of the anaphoricity, which appears to be providing contextual domain restriction. I argue that a lower DP projection is associated with this interpretation. I present data from Salish in support of observations similar to Bangla. This leads to the novel proposal of this chapter: two DP projections resulting in two different types of definiteness. While the lower one is associated with providing contextual restriction, the higher one contributes to definiteness involving an iota. Towards the proposal: Domain restriction and definiteness 3.6 Contextual restriction can be expressed in three ways – grammatical, pragmatic and semantic (Stanley & Szabo 2000). In the semantic approach, the domain is contextually restricted by a covert variable – either contextual domain variable (C) or a situation variable/ choice function (f(x)), which is bound in a context. Usually, determiners are attributed to providing the C (Giannakidou 2004, Etxeberria and Giannakidou 2010). The definite descriptions contain phonologically null situation variables that can refer to particular situations (von Fintel 1994). As we will see below, nouns and quantifiers with –Ta and –gulo are contextually 81 restricted, and this novel observation significantly contributes to our proposal of two definite DPs in Bangla. 3.6.1 Domain restricted quantifiers with –Ta and –gulo Recall that NP-Ta and NP-gulo represent singular definite and plural definite entities respectively. Earlier in this chapter, we established that the NP-fronted word order (NP Num-Ta) has the exact same distribution, and they all are anaphoric in the context. Additionally, -Ta and – gulo appear with degree quantifiers, and surprisingly, turn them into adnominal quantifiers, as in (97). Note that (97d) modifies a verb. But as the meaning stands, it modifies a elided cognate object (i.e. sleep). (97) a. [Onek-(gulo) bachcha] khelche. (only adnominal with -gulo) Onek-(GULO) kid playing ‘Many kids are playing.’ (=Onek) ‘Many (of the) kids are playing.’ (=Onekgulo) b. ram [Onek-(*gulo) lOmba] hoyeche. (*Onek-gulo + adjective) Ram Onek-(gulo) tall became ‘Ram has become very tall.’ c. ajke ami [Onek-(*gulo) ghumiyechi] (*Onek-gulo + verb) today I Onek-(gulo) slept ‘I slept a lot today.’ d. ajke ami [Onek-(Ta) ghumiyechi] (Onek-Ta + verb) today I Onek-(TA) slept ‘I slept a lot today.’ These quantifier expressions with –Ta and –gulo are presuppositional and are interpreted as strong quantifiers. They are restricted to the discourse-context. The domain of the strong 82 quantifier is always restricted in context, and the context can be restricted either implicitly or explicitly. (98) {SOb/Onek/kOto}-gulo phul Sada. (Presuppositional reading with Q-gulo) all/a lot/few -GULO flower white ‘All (of the) flowers are white.’ ‘Many (of the) flowers are white.’ ‘Few (of the) flowers are white.’ (99) kichu chatro porikkha diyechilo, few student test gave ‘Few students took the test.’ Onek-#(gulo) chatro paS koreche. (Strong reading with -gulo) Onek-(-gulo) student pass did ‘Many of the students passed.’ Below I present a list of degree and adnominal quantifiers that are compatible with –Ta and – gulo. As we will see, only degree quantifiers co-occur with –gulo and are interpreted as an adnominal count quantifier. These quantifiers with –Ta is interpreted as a mass quantifier, denoting a mass measure amount. 12 Whenever –Ta or –gulo co-occur with these quantifiers, they are restricted in the domain. The adnominal quantifiers with –Ta are interpreted mandatorily presuppositional. Quantifiers Domain Restricted availability of Q-Ta/-gulo & count-mass measures Domain Restricted Degree Quantifiers SOb ‘all’ No -Ta = Mass, -gulo = Count Yes Eto/Oto/kOto ‘this/that/how much/many’ No -Ta = Mass, -gulo = Count Yes 12 The count-mass distinction in the quantifier will be discussed in next two chapters. 83 Onek ‘many/much’/ ‘a lot’ No -Ta = Mass, -gulo = Count Yes Adnominal Quantifier kichu ‘a few/some’ No -Ta = Mass, ✗-gulo Yes prottek ‘each’ No -Ta = Count, ✗-gulo Yes kO(y) ‘how many’ No -Ta = Count, ✗-gulo Yes Quantifiers mandatory with -Ta Numeral No -Ta = Count - khanik-Ta No - - koyek-Ta No - - Table 11: Quantifiers with classifiers and count-mass selection 3.6.2 Domain restriction and DPs Bangla definiteness in discourse-anaphoric and pointing contexts precisely require the bare- classifier and non-canonical word order definite. The most conventional cases of contextual dependence involve pointing to a referent. This also extends to the anaphoric reference cases where the anaphoric link to a referent in a pre-existing sentence establishes the referential dependence. The strong, presuppositional interpretations, in the DPs and the quantifiers with – Ta/-gulo seem to suggest that these constructions involve contextual domain restriction. Generally definite determiners are the prime elements that contribute to domain restriction. Since nominals with –Ta or –gulo are also interpreted as definite descriptions, one may conjecture that these particles are definite determiners. However, demonstratives share similar semantics, and – Ta and –gulo co-occur with demonstratives. In fact, they are the only means to pluralize demonstratives. (100) a. ei gaRi-Ta bhalo chOle. this car{-Ta/-gulo} good run ‘This/these cars run(s) well.’ 84 b. ei{-Ta/-gulo} bhalo chOle. this-{-Ta/-gulo} good run ‘This/these run(s) well.’ This set of data reveals two facts – (i) –Ta and –gulo are not indexical by nature, and (ii) –Ta and –gulo themselves are not definite determiners. Wolter (2006) states that demonstrative descriptions refer uniquely relative to a salient situation distinct from the discourse context, while definite descriptions refer uniquely relative to the situation corresponding to the discourse context. Although they both presuppose uniqueness/maximality within the pragmatically restricted domains, the demonstrative additionally creates an immediately salient situation, by means of pointing gesture or visibility. If –Ta/-gulo co-occur with a demonstrative, it is apparent that they are not demonstratives or definite determiners themselves. The domain restriction is thus apparent on the presence of a definite determiner. 3.6.3 Two DPs in Bangla Previous syntactic accounts of Bangla definiteness assume that Bangla projects a single DP. A covert definite determiner with a strong definite feature is responsible for the definite interpretations of NP-Ta/-gulo and NP Num-Ta. The strong definite feature of this D requires the NP to move to Spec DP when matched, resulting in the NP-fronted NP-Ta/-gulo and NP Num- Ta. If we assume this definite determiner to be responsible for the domain restriction in definite DPs, one point remains unexplained: the strong quantificational interpretations of Q-Ta/-gulo, because no syntactic movement is apparent in the quantifiers. And, the D is syntactically in a much higher position for the quantifiers. To explain the contextual domain restriction in these quantifiers and also the DPs, I propose that a lower D head projects with –Ta and –gulo. I follow Zamparelli’s (2000) layered DP structure for this proposal. Zamparelli claims, cross-linguistically, different interpretations of the nominals, determiners and quantifiers can be unified under one DP structure if we assume that the DP has multiple layers. Based on Italian data, Zamparelli proposes three different positions for the DP, where the lexically filled highest position (SD max ) is responsible for strong 85 quantifier interpretations, while the lower DPs are reserved for adjectival and predicative interpretations respectively (Zamparelli 2000, pp. 233). The two-layered DP structure in Bangla is particularly relevant for the mandatory adnominal and presuppositional quantification properties with Q-Ta/-gulo. These quantifiers must be restricted in context. Since the adnominal quantifiers are adjectival in nature, I argue that these quantifiers lexically fill the specifier of a lower DP position. A null determiner, the functional head of this DP projection is responsible for the process of domain restriction. It also triggers existence presupposition. also The following is a schema for the DPs in Bangla. (101) I argue that the existence presupposition in the quantifiers co-occurring with –Ta and –gulo is a result of the contextual domain restriction provided by this lower DP. The higher DP is the stronger DP (following Zamparelli 2000). This is the position for strong definite markers (i.e., pronouns). The filled higher DP triggers maximality presupposition. 3.6.3.1 Anaphoric DPs The NP-Ta or NP-gulo is interpreted maximally definite and they are mandatorily context- bound. I argue that the first DP projects after the NumP. The anaphoric interpretation of these nominal DPs (i.e., NP-Ta or NP-gulo) is due to the movement of the nP to the Spec DP. A domain restrictor binds the denotation of the noun t the context. This DP also projects in cases of degree quantifiers. Thus, the Q-Ta/-gulo are interpreted as obligatorily presuppositional. DP strong wo wo … DP weak wo QP wo D 0 weak NP 86 (102) DP D 0 NumP -Ta/-gulo nP count n 0 indiv √apel ‘apple’ ‘water’ In addition, the higher DP also projects in NP-Ta/-gulo. This D 0 is a strong DP and due to the strong EPP feature, the NP further moves to the Spec, DP. Two factors. domain restriction by a lower D 0 and strong definite feature of a higher D 0 , are responsible for the strong anaphoric interpretations of the bare classifier forms. 3.6.3.2 Unique DPs Recall that the bare noun form represents the unique definite cases. I argue that only the higher DP projects in this case. The syntactic mechanism of this projection is not different from previous proposals for definiteness in Bangla. The novel proposal in this account is the separate projection for the anaphoric DPs. Note that the bare classifier form (i.e. NP-Ta) have both projections. Thus, it may appear in unique cases, as well, it must appear in anaphoric contexts. And, consequently, since the bare noun definite lacks the higher DP, they are unavailable in anaphoric contexts. In the absence of the higher D 0 , the quantifiers are can be only restricted to the context, due to the domain restriction by the lower D 0 . The semantics of the higher strong D 0 involve an iota. Following Dayal’s (2014) proposal, the strong D in this account encodes an iota operator. The application of the iota returns the maximal singular (NP-Ta) or plural individual (NP-gulo). Dayal (2014) defines the iota as a function that returns the maximal individual from the predicate it takes as an argument. 87 (103) (ι) <<e, t>, e>> : λP. ιP S , if there exists a unique maximal entity in P in situation s, undefined otherwise. Due to the null D 0 with a strong definite feature, the complements of –Ta and –gulo move to specifier DP. For the semantics of the definite bare nouns that are only possible in uniqueness contexts, I follow Roberts (2003) in assuming that all definites have presuppositions of existence of a discourse referent to satisfy their descriptive content, as well as of (informational) uniqueness relative to that descriptive content. Bare nouns satisfy these presuppositions for definite interpretations. 3.6.3.3 Bangla free choice items and –gulo Supporting evidence for this proposal comes from Bangla free choice items. These FCIs with – Ta and –gulo return wh-elements that are presuppositional (e.g. which). The FCI and presuppositional wh-words in Bangla are partially homophonous. Without going into further details (See Ghosh Dastidar 2015 for details), it may be conjectured that they are derived from the same root and the presence of the –Ta and –gulo precludes free choice by anchoring to a domain. For example, –gulo co-occurs with presuppositional wh-elements such as which, and may appear either with the complement NP (e.g. book in (a) below), or with the pronominal (e.g., kon ‘which’ in (b)), but it is not compatible with what. (104) a. kon boi-gulo bhalo? which book-GULO good ‘Which (of the) books are good?’ b. kon-gulo bhalo? which-GULO good ‘Which are good?’ (must be uttered in a context with pre-existing objects) 88 c. *ki-gulo bhalo? what-GULO good 3.6.4 Section Summary In this section I provided syntactic and semantic descriptions of Bangla definites. I have argued that Bangla has two DP projections. A covert higher D 0 has a strong definite feature that must satisfy by checking and moving the complement of the NP to the specifier DP. A lower DP projects when –Ta and –gulo are present with nouns and adnominal quantifiers. This combination results in contextual domain restriction over the complement NP, resulting in a mandatory discourse-anaphoric interpretation in the bare classifier forms and the presuppositional interpretation in the quantifies that combine with –Ta or -gulo. The higher DP projection on the top of the lower DP results in a strong anaphoric interpretation in NP-Ta or NP- gulo. This explains the preference for the bare classifier forms in anaphoric and indexing contexts. Additionally, I side with Dayal (2014) in suggesting an iota operator for the definite projection in the higher DP. This comes with maximality presuppositions. They are interpreted as anaphoric strong definites that are anchored to a previous context, or salient by visibility or pointing. For the definite bare nouns, I argue that only the higher DP is projected. For the semantics, I follow Roberts (2003) in assuming unique entities in context have existence presupposition. Hence the uniqueness readings of the bare nouns are satisfied. Conclusions 3.7 This comprehensive account present all possible definite constructions in Bangla. Five different constructions for definiteness construal are tested against six different contexts to find out the compatibility of the forms with different contexts. It broadly classifies the definite constructions into two groups: definiteness based on anaphoricity/familiarity, and definiteness based on uniqueness. Anaphoricity /familiarity/ identifiability based strong definite features employ a 89 higher DP projection, whereas the uniqueness based weaker definite, that prefer the bare nouns are associated with a lower DP projection. 90 4 Chapter 4: –Ta Bangla linguistic literature, ranging from prescriptive grammars to modern generative linguistic literature, treats -Ta as a ‘default’ classifier (Chatterji 1926; Dasgupta 1983, 1985, 2005; Bhattacharya 1999, 2000, 2001; Ghosh 2010; Dayal 2012, 2014 among others). –Ta is claimed to be one of the few classifiers in Bangla, and the most apparent reason for this claim is that numeral constructions generally require –Ta. However, -Ta is not confined only to the numerals, it can be cliticized onto several other nominal elements. Nouns, for example, when co-occurs with –Ta, without the presence of a numeral, are interpreted as singular definites. Quantifiers co- occurring with -Ta result in presuppositional strong quantifiers. –Ta also co-occurs with adjectives (and even verbs). The contribution of –Ta in the interpretation of these associates are unlikely for that of a classifier. In this chapter, I discuss the syntactic and semantic properties associated with the presence and absence of –Ta. I first discuss the distributions of –Ta with various nominal elemets (section 4.1). Then I discuss novel observations concerning the presence and absence of –Ta in various constructions (sections 4.2). These observations reveal a nature of –Ta and its relation to counting that the existing theories of –Ta does not satisfactorily explain. After discussing available accounts (section 4.3), I present the proposal and discuss the implications (section 4.4). I argue that -Ta introduces a singular partition on its complement. Distributions of –Ta 4.1 4.1.1 –Ta with numerals The particle –Ta is obligatory with numerals in a numerical construction. True to its properties corresponding to clitics, it is suffixed to the numeral. The default order of the numerically quantified NPs are [Num–Ta NP], and is interpreted as an indefinite. The count-mass distinction is coerced in this structure. Whenever a noun appears with a numeral, count and mass nouns alike occur with –Ta, and the result is a count expression. Thus, am ‘mango’ - a count noun in 91 (105a), is syntactically not different jOl ‘water’ in (105b), a prototypical mass noun, and the latter is interpreted as ‘3 bottles of water’. Default classifiers in proto-typical classifier languages (e.g. Mandarin Chinese) do not co-occur with prototypical mass nouns (105) a. tin-*(Te) 13 am (Bangla) three-TA mango ‘three mangos’ b. *tin-Te jOl (Bangla) three-TA water ‘three bottles of water’ c. *yi ge cha (Chinese) one cl tea 4.1.2 –Ta with nouns Although –Ta appears to be mandatory with numerals, it is not restricted to the presence of a numeral. It can be cliticized to a numeral (106a) and a noun without a numeral (106b), with differences in interpretations. The ‘bare classifier’ pattern (cf. Simpson 2005), where the classifier appears with bare nouns without the presence of numerals, are not very common in classifier languages. Only a handful of them show this property (e.g., Cantonese, Min, Yi etc.). (106) a. Ek-Ta pakhi ghOr-e Dhuklo. One-TA bird room-LOC entered ‘A bird entered a room.’ b. pakhi-Ta ghOr-e Dhuklo. bird-TA room-LOC entered ‘The bird entered the room.’ 13 -Te, –To and -Ti are phonological variations of –Ta. For pragmatic (and semantic) nuances associated with the forms, see Dasgupta (2009). 92 In the absence of a numeral, -Ta can cliticize on to mass nouns without changing their prototypical properties. No mandatory count interpretations arise in this form. (107) a. pakhi-Ta ghOre Dhuklo. bird-TA room.at entered ‘The bird entered a room.’ b. dudh-Ta poRe gElo. milk-Ta drop-PPL went ‘The milk got spilled.’ In addition, these structures, i.e., nouns with –Ta without the numeral, are interpreted as definites. If –Ta combines with a count noun, it is interpreted as a definite singular. If it appears with a mass noun, it is interpreted as a definite mass entity. The other less-productive classifier forms also show this trait (e.g., -khana). (108) a. am-Ta kothay? (Count noun + -Ta, Bangla) mango-TA where ‘Where is the mango?’ b. jOl-Ta kothay? (Mass noun + -Ta, Bangla) water-TA where ‘Where is the water?’ c. boi-khana kothay? (Count noun + -khana, Bangla) book-KHANA where ‘Where is the (damned) book?’ As discussed in the previous chapter, the NP-movement to the Spec, DP explains the definite interpretation of this form. However, the obligatory number marking remains to be a question. 93 Relevant cross-linguistic literature considers that classifiers are number markers (cf. Nomoto 2013). Unlike the general perception that classifier languages lack number marking, Nomoto proposes that classifier languages do not lack number morphology, and in fact, they distinguish between three types of basic number categories: singular, dual and general. As we will see, this account will be pivotal in understanding the Bangla phenomena, and we will modify the assumptions and the proposal as we proceed. 4.1.3 –Ta with quantifiers -Ta is compatible with most quantifiers, and is mandatory with some quantifiers (e.g. numerals). The strongest empirical evidence against its proto-typical classifier properties comes from the interpretations involving some of these quantifiers. Some quantifiers in Bangla are unspecified for count or mass, (e.g., Onek ‘a lot, much/many’). The presence of -Ta with such quantifiers yields an obligatory mass reading (i.e. Onek-Ta ‘much’). Such quantifiers with –gulo, the plural counterpart of –Ta, are interpreted to have an obligatory plural interpretation (Onek-gulo ‘many’). With other quantifiers that are specified for count, -Ta is mandatory and renders only a count interpretation (kOek-Ta, few/several). If –Ta were a proto-typical classifier, such dichotomous count and mass interpretation would not be predicted; only a count interpretation would have been predicted instead. (109) a. RiSi Onek {pakhi / jOl} dekhlo. Rishi a lot bird/ water saw ‘Rishi saw many birds.’ ‘Rishi saw much water.’ b. RiSi Onek-Ta jOl dekhlo. Rishi a lot-TA water saw ‘Rishi saw much water.’ c. *RiSi Onek-Ta pakhi dekhlo. Rishi a lot-TA bird saw 94 Interpreted: ‘Rishi saw much bird-stuff.’ d. RiSi Onek-gulo pakhi dekhlo. Rishi many-gulo bird saw ‘Rick saw many birds.’ With quantifiers specified for count (e.g., every, few, each etc.), only a count interpretation arises with -Ta. (110) RiSi kOyek-*(Ta) pakhi dekh-lo. Rishi few-TA bird saw ‘Rishi saw several birds.’ These unique characteristic properties suggest that a deeper intervention is required for categorizing –Ta. Below I present novel observations involving –Ta that add to the problematic categorization of –Ta and leads to a better understanding of the nominal structure. New observations: Interpretations in the absence of –Ta 4.2 4.2.1 Numerals without –Ta 4.2.1.1 Approximation So far we have seen that –Ta is mandatory with numerals. In O’Connor & Biswas (under revision) we first reported that relatively smaller numbers mandatorily retain –Ta, whereas, –Ta is optional with large round numbers. Any large round numbers above 100 optionally require – Ta, and in fact, the presence of –Ta gets increasingly unacceptable with higher numbers involving thousand. This is represented in the following examples: in (111a & b) –Ta is optional with a large round number like 100, whereas it is obligatory with precise numbers like 8 or 108 (112). Any number below one hundred must have –Ta in a numerical construction. Mass nouns with -Ta in numeral structures show similar properties. 95 (111) a. Ek-So-Ta am one-hundred-TA mango ‘exactly one hundred mangoes’ b. Ek-So am one-hundred mango ‘roughly one hundred mangoes’ (112) a. aT-*(Ta) am eight-TA mango ‘eight mangoes’ b. Ek-So aT-*(Ta) am one-hundred eight-TA mango ‘one hundred and eight mangoes’ Large round numbers (100 and higher) are suitable for approximation. The absence of –Ta in such cases represents a lack of precise counting mechanism. By ‘precise counting’ I mean counting individuated items one by one. The presence of –Ta implies precise counting. –Ta is mandatory in higher precise numbers (e.g., 164). Large round numbers allow and prefer omission of –Ta, while small numbers do not allow omission of –Ta. A small number like eight is easily countable, and allows an easy access to precise numeral interpretation. -Ta is absent in other syntactic mechanisms for approximation. Small numbers allow omission of –Ta only when they are morpho-syntactically marked approximate. This approximation construction has a circumfix-like structure. The approximated numeral is preceded by goTa ‘whole’ and then optionally followed by –ek ‘one’. It is restricted to sub- decimal numbers and multiples of ten, numbers easily amenable to approximation. The -ek is phonologically constrained for numbers ending in a palatal sibilant [S]. The examples in (113a) shows that –Ta is ungrammatical in a sub-decimal approximate construction. Numbers higher than ten do not employ the suffix –ek ‘one’ (113b), unless the number ends in a [S], as in (113c). 96 (113) a. goTa paMch-ek-(*Ta) am appx.PRT five-one-TA mango ‘approximately five mangoes’ b. goTa ponero-(*ek) am appx.PRT fifteen-(*one) mango ‘approximately fifteen mangoes’ b. goTa tiriS-(ek) am appx.PRT thirty-(one) mango ‘approximately thirty mangoes’ The approximation strategy for numerals higher than 10 involves numeral doubling, and –Ta appears only with the second numeral. This strategy generally involves two consecutive numerals or two multiples of 10. (114) a. Egaro-baro-Ta am eleven-twelve-TA mango ‘approximately eleven-twelve mangoes’ b. Sottor-aSi-Ta am seventy-eighty-TA mango ‘approximately seventy-eighty mangoes’ The presence of –Ta implies the presence of precise counting. The absence of –Ta is represents an approximate outcome. Whether the underlying procedure for this outcome is precise counting or estimation is not clear at this point. But, the presence and absence of –Ta present a general pattern that reflects its correlation with precise vs. imprecise counting mechanisms. Classifier optionality is found in other classifier languages too. In Burmese, classifiers do not occur with multiples of ten. Unless individuation is implied, Thai classifiers do not co-occur 97 with large round numbers like 1000. Assamese has around ten independent lexemes that serve as numeral classifiers and they combine on the basis of sex, animacy and politeness (e.g., zOn (respected human male), zOni (disrespectful human male), zOna (high status humans) and gOraki (respectful human). Thus, it is not unlikely that an approximated reading arises in the absence of a classifier. 4.2.1.2 Attributive modifiers Numeral constructions without –Ta serve as attributive modifiers. For example, ‘a four-wheeled car’ involves a numeral ‘four’ without –Ta, as in (115a). The presence of –Ta makes the noun phrase a numerically quantified one (115b). (115) a. char chaka-r gaRi four wheel-GEN car ‘a four-wheeler’ b. char-Te chaka-r gaRi four-TA wheel-GEN car ‘four cars carrying/of wheels’ Analogous interpretations are available with measure nouns. Some dimensional nouns, such as diameter, are interpreted non-monotonically on the part-whole structure, while length is interpreted monotonic in the part-whole structure (See Schwarzschild 2006 for details). Thus, 2- inch cable (diameter) is non-monotonic and cannot be quantified by any numeral, whereas 2 inches of cable (length) is monotonic on the part-whole structure (i.e. adding more cable increases the length). In Bangla, a genitive must appear in the non-monotonic measure, as in (116a) and it can further be quantified with a [Num-Ta] construction (e.g., four-Ta [four-kg-GEN apples]). Thus, it is amenable for attributive modification, just like the numerals without –Ta. (116b), on the other hand, is monotonic on the part-whole structure, and cannot be further quantified by numerals. 98 (116) a. char keji-r apel [non-monotonic] four kg-GEN apple ‘a 4-kg apple’ b. char keji apel [monotonic] four kg apple ‘4kgs of apples’ The similarity of the bare numerals (i.e. -Ta-less numerals) with the non-monotonic measure phrases suggests that the absence of –Ta in numerals contributes attributive modification which the presence of –Ta precludes. The two measure phrases discussed above are also syntactically different. The monotonic phrases (QP, AP or NP) appear higher (than the partitive of) in the structure, while the non-monotonic attributive ones appear lower in the tree (Schwarzschild 2006, pp. 92). If this generalization is correct, absence of –Ta (or the attributive adjectives) appears to have a lower position in the DP. 4.2.2 Measure NPs without –Ta: Weak noun phrases Related to the previous discussion, another structure that mandates the absence of –Ta is the measure pseudo-partitive construction (e.g., three bottles of wine – amount of wine). This construction is interpreted as a weak noun phrase, and does not participate in scope ambiguities. The noun phrase denotes an amount measure. (117) a. aT-(*Ta) botol wain (Measure pseudo-partitive) eight-(*TA) bottle wine ‘eight bottles of wine’ b. aT-(*Ta) bakSo boi eight(*TA) box book ‘eight boxes of books’ 99 On a similar note, it must be noted that Bangla does not allow pseudo-partitive ambiguities, as observed in English and some other languages. Due to the inherent counting nature in the individual pseudo-partitives (e.g. three bottles of wine – number of bottles), it must involve –Ta and thus have a different syntactic structure. The structure (similar to three wine’s bottles) does not correspond to the English pseudo-partitive constructions. Like other weak predicative noun phrases, noun phrases without –Ta do not allow the inverse scope reading, which its counterpart with –Ta does. As discussed in the previous chapter, the adnominal quantifiers such as each project a lower DP that is responsible for the domain restriction. Here the presence of –Ta makes the quantifier phrase a strong one, and the absence thereof results in a weak one. (118) a. prottek chhele EkSo sinema dekhechhe. each boy one-hundred movie saw ‘Every boy watched 100 movies.’ Every boy > 100 movies *100 movies > every boy b. prottek-Ta chhele EkSo sinema dekhechhe. each-TA boy one-hundred movie saw ‘Every boy watched 100 movies.’ Every boy > 100 movies 100 movies > every boy 4.2.3 Quantifiers with -Ta: Exact measure Noun phrases with quantifiers that are compatible with –Ta denote a precise or exact amount. Similarly, measure demonstratives attached to –Ta (e.g. this much-Ta) involves a specific quantity in their interpretation. It is important to note at this point that the third person 100 demonstratives in Bangla turn to pronouns when –Ta is attached to them. They are interpreted as anaphoric definites. 14 Bangla correlative measure quantifiers must denote the exact amount in the relative and the correlative segments when –Ta co-occurs with them. Without –Ta, no mandatory matching of amount is imposed. This is illustrated in the following example. If someone drinks 200ml of water, the quantifier with –Ta in (119b) implies that they sweated the exact amount (i.e. 200 ml), and thus renders it implausible. The quantifiers without –Ta in (119a) however, does not have the same implications. (119) a. jOto jOl kha-be, tOto gham-be. REL-much water drink-FUT CREL-much sweat-FUT ‘The more you’ll drink water, the more you’ll sweat.’ b. # jOto-Ta jOl kha-be, tOto-Ta gham-be. REL-TA-much water drink-FUT CREL-TA-much sweat-FUT Interpreted: ‘The exact amount of water you’ll drink, you’ll sweat the exact amount.’ Similar interpretations arise with other quantifiers. Particularly those that require an exact count, -Ta is mandatory with them. Thus, count quantifiers (including numerals in the general sense), that mandatorily require access to atoms, requires –Ta. For example, kOyek ‘several’ is a count quantifier and requires access to atoms, hence –Ta is mandatory with it (120a). Certain quantifiers are compatible only to episodic sentences when they appear with –Ta. For example, prottek ‘each’ may appear in a generic or characterizing sentence without –Ta. With –Ta it appears in episodic sentences. (120b) shows that prottek is implausible in an episodic sentence while with –Ta (120c) it is perfectly compatible in such sentences. 14 The following chart shows a selected list of demonstratives and pronouns in Bangla. Demonstratives Pronouns Anaphoric Se(i) N Se-Ta Proximal e(i) N e-Ta Distal o(i) N o-Ta Relative-correlative je N-Se N je-Ta, Se-Ta Question kon N Kon-Ta 101 (120) a. kOyek-*(Ta) phOl pOcha. few-Ta fruit rotten ‘A few fruits are rotten.’ b. *prottek phOl pOcha. few fruit rotten ‘Every fruits are rotten.’ c. prottek-Ta phOl pOcha. few fruit rotten ‘Each fruit is rotten.’ The exact count implication with –Ta is also evident when it appears with inalienable possessions. Adding –Ta takes the inalienable meaning away, perhaps due to the explicit counting implication that is redundant for inalienable possessions. The same is the case for idiom constructions. Semantically bleached elements relating to inalienable parts naturally involve their count, adding –Ta forces a counting measurement, as in (121c), and often renders it unacceptable. (121) a. amar du-hat lOmba. [Inalienable construction] my two-hand tall ‘My (two) hands are longer.’ b. du-To hat bhanNa. two-TA hand broken ‘Two hands are broken (preferably not mine)’ c. du-(*To) kan kata. [idiom] two(*TA) ear cut ‘two-ear-cut person’ (=shameless) 102 4.2.4 Optionality of classifiers cross-linguistically Optionality of classifiers with certain numerals is not unheard of, but it is definitely not widely discussed. Nomoto (2013) lists a few cases where classifier optionality is present in classifier languages. 15 Korean classifiers, for example, are optional for a limited number of animate nouns (e.g. salam ‘person’, kay ‘dog’, haksyan ‘student’). Vietnamese classifiers are not compatible with abstract nouns. Classifiers are often optional for, and sometimes even unacceptable for large and non-specific numbers. As Nomoto cites examples of classifiers from Nung and Burmese, they are optional for multiples of ten. Large numbers such as 1000 in Thai also drop classifiers (Aikhenvald 2000: 100). Classifiers are also optional in Mandarin vague quantity expressions such as many and roughly 100. Similar trait is also observed in Japanese when non-specific quantifiers fill the slot of the numeral. In Japanese, even numbers as low as 15 can be made approximate with dropping the classifier. In Persian, classifier absence is more acceptable with larger numbers (~20) than smaller numbers (~3). (122) a. slam pam (áhn) hon three hundred CLF house ‘three hundred houses’ (Nung, Saul & Wilson 1980) b. ni san (nin)-no gakusei two three CL-PRT student ‘two or three/a few students’ 15 In Japanese and Vietnamese, speaker variability exists for the optionality. Classifiers are optional in lists. Classifiers are not used in Vietnamese when placing orders in food stalls or waiters repeating their order. In obligatory classifier languages like Japanese and Thai, elimination of the classifier in a ‘NP-case Num CL’ form gives rise to subkind readings, along with the already existing object reading. (i) Uti-de-wa petto-o zenbude kyuu atukatteiru keredo, neko-ga itiban ureteiru. we-at-TOP pet-ACC in.total nine deal.in and cat-NOM most sell ‘We have nine pets in total, and cats sell the most.’ (ii) Uti-de-wa petto-o zenbude kyuu hiki atukatteiru keredo, we-at-TOP pet-ACC in.total nine CL deal.in and ‘We have nine pets in total, and’ ??neko-ga itiban ureteiru. cat-NOM most sell ‘and cats sell the most.’ (Nomoto 2013: 24, ex. 29-30) 103 c. {san *(ko)/ kyuu ?(ko)/ juu-go (ko)}-no gengo {three CL/ nine CL/ fifteen CL}-PRT language ‘three/nine/fifteen languages’ d. hóu dō (jī) bāt very many CL pen ‘many pens’ (Cantonese, Matthews and Yip 2011:111) Cross-linguistically, classifiers cannot be dropped in counting contexts. Moreover, in Thai, the classifier can be optional in a ‘Dem N’ construction and similar readings arise as observed in Japanese. (123) a. rót níi [car this] (i) ‘this particular car’, (ii) ‘this kind of car’ b. rót khan níi car CLF this (i) ‘this particular car’, (ii) *‘this kind of car’ (Nomoto 2013: 25-25, ex. 34) Similarly, in Japanese, kazu ‘number’ is incompatible in counting constructions and vague quantity expressions. This morpheme, however, appears in quantifier constructions, like many/ a lot, just like –gulo (Onek vs. Onek-gulo) in Bangla. 16 4.2.5 Section summary In this section, we observed that quantifier constructions with –Ta imply that the output involve a precise measure, either derived by precise counting or have the exact measure. It does not allow 16 Watanabe (2010) does not motivate any change of interpretation for the two competing forms of many in Japanese. Many is an adjective on par with its pair with few. (i) a. John-wa ooku-no ronbun-o kaita. John-top many-link paper-acc wrote b. John-wa kazu ooku-no ronbun-o kaita. John-top number many-link paper-acc wrote ‘John wrote a large number of papers.’ 104 any estimation in the output. Large precise numbers (e.g., 789) thus must co-occur with –Ta, while –Ta is optional with large round numbers (e.g., 800). This is supported by the fact that –Ta is blocked in larger round numbers (e.g. 10000) and in approximate structures. The attributive reading in the absence of –Ta also indicates the lack of precise counting. Thus, weak readings and measure readings are found in pseudo-partitives or measure constructions. –Ta is also not preferred in idioms, inalienable possession and numeral reduplications, indicating relevance to its referential property of the denotation. Additionally, -Ta with nouns in a non-numeric construction must denote a definite singular number. All these data seem to suggest that –Ta is associated with a precise count, and most probably with number marking. Cross-linguistic observation also reveals the correlation of the absence of classifiers with vagueness in numeric constructions. Classifiers cannot be optional when counting is mandatory. However, how and why this is the case is not clear in the existing literature. Additionally, -Ta in Bangla also suggests the evidence of singular number marking. Recall that Bangla is considered a classifier language, and classifier languages generally lack number marking. However, recent work in classifier languages explores possibilities of number marking. Next I briefly discuss the existing literature on number marking in classifier languages and numerals. Then I present my account of –Ta and its implications. Previous work on classifiers and number marking 4.3 Among the extensive list of previous literature on numeral classifiers and related constructions (Li 1998, 1999; Cheng & Sybesma 1999, 2005; Watanabe 2006, Jenks 2011, Simpson 1995 among many others), I discuss two recent studies as that particularly explores the possibilities of number marking in these languages and are particularly relevant to the nature of my work. 4.3.1 Classifiers function between numeral and atomic entities (Jiang 2012) Jiang (2012) follows Neo-Carlsonian assumptions of the nominal domain: bare nouns in classifier languages are kind-denoting nouns <e k >, while those in number marking languages are 105 property-denoting <e,t>. 17 She proposes that a classifier performs three functions – (i) converts kinds to properties, (ii) checks atomicity of the noun, and (iii) takes a numeral argument to create a numeral construction. The classifier takes two arguments, analogous to a transitive verb, a kind and a numeral. Numerals are syntactically phrasal, and are non-restrictive modifiers of type <<e,t>, <e,t>> (following Ionin & Matushansky 2006). Jiang furthers the idea by proposing that numerals in the Universal Grammar are ambiguous, depending on whether they incorporate a choice function or not, as in (124a) and (b) respectively. The choice function is instrumental for the indefinite interpretation of numeral constructions. She also assumes that the structure of the numeral construction across languages is similar. Following are the definitions of the numerals. (124) a. Numeral <<e, t>, e> = λPf∃(n P ) (Indefinite reading) b. Numeral <<e, t>, <e, t>> = λP[n P ] Syntactically, the classifier has its own functional projection above the NP. After checking atomicity of the noun, it combines the noun with the numeral projected above the ClP to achieve a set of individuals or sets. The details of the atomicity checking are not discussed, but I conjecture that it is done by restricting only atoms in the denotation of the noun. If the numeral has a choice function which turns the property-denoting NP to an entity (124a), and the NumP achieves an indefinite reading by existential closure. The numeral without the choice function remains property-denoting (124b), and can be further selected by other functional elements like determiners (e.g., the three apples). Following Dayal (2011), Jiang considers Bangla a classifier language (N <e> ) with a D. Bangla has a null definite D, albeit without a lexical definite determiner, which blocks any covert type shifting from predicative NPs to definite entities (i.e. ARG in syntax and semantics). Definiteness is marked by NP-fronting in numeral(less)-classifier phrases. –Ta is considered the general classifier for count nouns. It is a function of type <e, <n, <e, t>>>. It takes the kind entity and returns an atomic property with an unsaturated n-argument. Next it combines with a numeral to saturate the n-argument and returns a property with the n-number of atoms in the denotation. 17 Jiang (2012) follows the insights presented by Dayal (2012 & p.c.). This account heavily relies on the data and resembles the proposals of Dayal for Bangla (2012, and later 2014). Therefore, I exclusively discuss Jiang (2012). 106 (125) ⟦−𝑇𝑎⟧= 𝜆𝑘𝜆𝑛(𝑛(𝐴𝑇( 𝑘 ∪ ))) Either of the numerals in (124a) or (124b) combines with the syntactic object [–Ta NP]. Depending on the nature of the numeral, the numeral construction receives an existential indefinite reading, or is subjected to further modifications. Jiang considers the bare classifier (NP-Ta) construction originates from a numerical construction. A lexical detransitivisation rule applies to –Ta after the numeral has merged, and nullifies the requirement for a numeral argument of the classifier. In that case, cover type shift is blocked for definiteness, and the NP moves to the DP for satisfying a definite interpretation. 4.3.1.1 Limitations Although Jiang’s proposal accounts for Bangla interpretations very well, there are a few observational and theory-internal ramifications that must be addressed. First of all, the Lexical detransitivization rule associated with one-deletion in Bangla (and Yi), allows a numeral- containing classifier phrase to change to a numeral-less classifier phrase, as in (126). However, it applies only to the numeral one (hence one-deletion). Since Jiang does not provide a different semantics for one and the other numerals, theoretically, there is no way of blocking other numerals to undergo the same parameterization rule. However, that is not the predicted outcome. (126) a. Ek-Ta chhele b. chhele i Ek-Ta t i one-TA boy boy one-TA ‘one boy’ ‘the boy’ Secondly, Jiang’s proposal accounts for only indefinite interpretations of the bare nouns. In chapter 3 of this dissertation, I showed that definite bare nouns are allowed in unique contexts, when the referent is inferred to be unique in the context. Jiang’s proposal addresses Situation Restriction (SR) that converts kind-denoting nominals directly to definite entities (<e k , e i >). However, it is not preferred in Bangla, as Bangla exercises a no covert-type shifting rule, generally applied for languages with functional null D. If a null D is present, it projects a DP in the syntax in order to make property-denoting nominals arguments. If the restriction on covert- 107 type shifting were not available, the Situation Restriction (<e k ,e>) could type-shift bare kind- denoting nouns to definite entities to make a definite interpretation available, with undesirable consequences. Third, since –Ta always checks for atomicity in the complement noun, it does not combine with mass nouns. Even if it does, it might be conjectured that the mass nouns are coerced into count ones for such cases. Therefore, when the detransitivization rule applies to the ClP, it returns only coerced atomic interpretation of a mass noun. This predicts a wrong interpretation for water-TA. The correct interpretation of water-Ta does not involve such an interpretation (e.g. ‘the one (bottle of) water’). It is generally interpreted a definite mass portion, as ‘the water’, without any assumption of number. It retains its proto-typical mass properties in such constructions. (127) a. Ek-Ta jOl b. jOl i Ek-Ta t i one-TA water water one-TA ‘one (bottle of) water’ ‘the (bottle of) water’ Furthermore, due to the claim that –Ta introduces a numerical argument, it can never compose with non-numerical quantifiers. And, as we have seen in section 4.1.3, -Ta shows interesting distributional patterns involving count-mass measures when co-occurring quantifiers. Consequently, we are also unable to explain the exact measure implication of –Ta, and the estimation interpretation we get in the absence thereof. 4.3.2 Classifiers as number markers (Nomoto 2013) Unlike the Neo-Carlsonian assumptions, Nomoto (2013) assumes that bare nouns in classifier and number marking languages denote properties. Generic classifiers are singular number makers. More specific classifiers incorporate specification/classification properties along with the singular number marking. A classifier comes with two functions: it restricts noun denotation to singularities, and it specifies the characteristics of the combining noun. Classifiers are functions from number-neutral properties to properties that hold with singularities alone. Its semantic type is <<e,t>, <e,t>>. 108 (128) 𝐶𝐿 = 𝜆𝑃𝜆𝑥 𝑃 𝑥 ∧ ¬∃𝑦∈𝑃 𝑦< 𝑥 t 𝜆𝑃[𝑃 ⊆𝐶𝐿𝐴𝑆𝑆], Where CLASS denotes conjoined properties of the noun and the relevant classifier. t is conjunction. [Nomoto (2013: 82)] One of the strong points of this account is that it explains how the nature of classifier and noun mismatch result in infelicitous sentences. In some cases, use of a different classifier may point to speaker’s attitude (e.g. condemnation for using non-human classifier with a human noun) and is in accordance with conventional implicature. This is also found when –Ta co-occurs with nouns that generally hold high values. For example, professor-Ta has a derogatory meaning. This meaning does not interfere when it is in the scope of negation; it cannot be bound and can pass through belief-contexts (see Nomoto 2013 for details). Nomoto argues that classifiers are singular number markers, and they are different from plural number markers only by how they restrict the noun denotations. Classifiers restrict the denotation to singularities, while plurals restrict to the domain to pluralities. Any numeral-less bare classifier form thus denotes a singular entity (e.g. water-Ta). Other than Bangla, such bare classifier forms exist in Mandarin (Iljic 1994), Cantonese (Cheng & Sybesma 1999), Thai (Bisang 1999), Hmong and Vietnamese (Simpson et al. 2011), and Nomoto claims that his proposal account for such interpretations. The semantic functions of the classifier and the plural marker are given below. (129) a. 𝐶𝐿 = 𝜆𝑃𝜆𝑥 𝑃 𝑥 ∧ ¬∃𝑦∈𝑃 𝑦< 𝑥 t 𝜆𝑃[𝑃 ⊆𝐶𝐿𝐴𝑆𝑆] b. 𝑃𝐿 = 𝜆𝑃𝜆𝑥 𝑃 𝑥 ∧ ∃𝑦∈𝑃 𝑦< 𝑥 t 𝜆𝑃[𝑃 ⊆𝐶𝐿𝐴𝑆𝑆] 4.3.2.1 Two types of numerals cross-linguistically Recall that Jiang (2012) assumed the existence of two types of numerals in the Universal Grammar. Nomoto (2013) too assumes two types of numerals, but he assumes that the numeral constructions vary cross-linguistically. He pursues an idea that numerals are restrictive modifiers (following Bale et al. (2011)). Similar to the adjectives, there are two types of modifiers 109 (numerals): subsective or intersective. When these numerals combine with number-neutral bare nouns, they select the same individual in different ways. For example, for the number three, the subsective numeral restricts the noun denotation to a subset of it such that the cardinality of its element is three (e.g., similar to big restricting the noun denotation to a subset that satisfies the adjective description). The intersective numeral intersects the noun denotation with the set of plural individuals whose cardinality is three. The semantics of the subsective and intersective numerals are given below (from Nomonot 2013). (130) a. 3 ! = 𝜆𝑃𝜆𝑥.∃𝑌[Π 𝑌 𝑥 ∧ 𝑌 = 3∧∀𝑧∈𝑌 𝑧∈𝑀𝐼𝑁 𝑃 ] (Subsective) b. 3 ! =𝜆𝑃.𝑃∩ {𝑥: ∃𝑌[Π 𝑌 𝑥 ∧ 𝑌 = 3∧∀𝑧∈𝑌[𝐴𝑇𝑂𝑀(𝑧)]]} (Intersective) Where MIN(P) is defined iff no two smallest elements in P overlap (∀𝑥,𝑦∈𝑃 ¬𝑧∈𝑃 𝑧< 𝑥 ∨𝑧<𝑦 → 𝑥∩𝑦 ≠∅ . ATOM(x) = 1 iff x is an atom in the domain of the model. The subsective interpretation of three represents that the subsective numeral is a function from the denotations that are closed under the sum operator to a subset that consists of all and only groups that are composed of three non-overlapping minimal parts. The MIN function is applied on the nominal predicate to ensure that the minimal overlapping units are defined by the nominal predicate, so that it may vary according to the predicate (e.g. different sets in big elephant vs. big mouse). The partition (Π) of a plural individual is a set of plural individuals where the sum of the non-overlapping element equals to the plural individual. The intersective numeral three represents the set of all aggregates/groups consisting of two atoms each. If in a language the numerals greater than one must combine with plural marked nominals (e.g. English three apples) then it is suggested that the language must have intersective numerals. On the other hand, if the numeral must not combine with plural nouns (e.g. Persian, Turkish), the language is suggested to have subsective numerals. Nomoto presents evidence of languages containing both subsective and intersective numerals (e.g. Malay) where the numeral must combine with a bare (number-neutral) noun or a noun with a classifier (singular-marked), but the language also allows the numerals to combine with plural-marked nouns (See Nomoto 2012: 92 for details). According to this diagnostic, and if 110 –Ta is a classifier, then Bangla numerals are subsective numerals as they do not combine with plural-marked numerals (e.g., *three-gulo apple). Let us now consider each definition numerals and evaluate the proposal for Bangla numerals. Recall that Bangla bare nouns are number neutral. This indicates that the noun denotation contains both atoms and sums in the denotation. For example, if a, b, and c are three apples, then 𝑎𝑝𝑝𝑙𝑒𝑠 = {a, b, c, ab, bc, ac, abc}. If –Ta is a classifier, then according to Nomoto’s definition, –Ta merged with apple restricts the nominal denotation to only atoms (i.e. {a, b, c}). After the subsective three is merged, it returns a subset that is composed of groups of cardinality three. This is not defined according to Bale et al. (2011), while it is defined under Nomoto’s account, who claims that Bale et al. might have overlooked this possibility. Nomoto argues that the output is a completely different set than the input: the denotation of resulting set is composed of sums of the atomic units (i.e. abc by a, b, c) represented by the denotation of the syntactic object –Ta NP, which is closed under a sum operator. The details are given below. (131) a. 𝑎𝑝𝑝𝑙𝑒 = {𝑎,𝑏,𝑐,𝑎𝑏,𝑏𝑐,𝑎𝑐,𝑎𝑏𝑐} b. –𝑇𝑎 𝑎𝑝𝑝𝑙𝑒 = {𝑎,𝑏,𝑐} c. 𝑡𝑖𝑛−𝑇𝑎 𝑎𝑝𝑝𝑙𝑒 = 𝑡𝑖𝑛 𝑎,𝑏,𝑐 = 𝑎𝑏𝑐 4.3.2.2 Limitations There are two issues with this account. First, -Ta appears with any numeral in Bangla, irrespective of singular and plural. When –Ta combines with the numeral one, the numeral function is vacuously applied: the partition contains no element. Secondly, there are a few other languages that have number neutral bare nouns, and numerals higher than one do not combine with plurals (e.g. Persian - Nomoto 2012, Turkish – Bale et al. 2011). These languages allow number neutral bare nouns to combine with the numerals. Specifically, Persian has a classifier tâ that is optional with numerals. 18 The optionality is predicted in the account. Given that Bangla also has number neutral bare nouns and the classifier -Ta, just as Persian, it should allow bare 18 Bangla –Ta might have been borrowed from Persian –Ta during Persian invasions. Middle Indo Aryan texts, which coincide with the invasion, shows first use of –Ta, which was not found in Old Indo Aryan texts (Chatterji 1926). 111 nouns to appear with –Ta without any change in meaning. The mandatory appearance of –Ta with numerals is thus not accounted for. Additionally, both subsective and intersective numerals may combine with a number neutral bare noun; hence, even if we assume that Bangla has intersective numerals, it will falsely predict a numeral construction with a bare noun (i.e. Num NP and *Num-Ta NP), without any change in meaning. Like Jiang (2012), Nomoto also does not involve the discussion of quantifiers with –Ta and their various interpretations. From the current semantics, no quantifiers are predicted to co-occur with –Ta. 4.3.3 Numerals as non-restrictive modifiers (Ionin & Matushansky 2006) Both Jiang and Nomoto assume two types of numerals to account for the facts that seem to be more closely related to the core properties of the nominal. Therefore, assuming a unified account of the numeral, both cross-linguistically and within a language, may lead to better explanation of the observation. Ionin & Matushansky (2006) mostly concerned about the complex numerals, and the simple numeral semantics are provided as building blocks to the complex ones. In this account, numerals are modifiers of type <<e,t>, <e,t>>. Numerals cannot be determiners, as such an account falsely predicts no numeral as restrictor of definites (e.g., *the three apples in English). The lexical NP complement of the numeral requires an atomic set, the elements of this needs to be semantically singular. The ‘atomizer’ can either be part of the lexical entry of the numeral, or it can be a classifier. Syntactically, a numeral may appear in both specifier and the head, however, the specifier structure is preferred for the localization of the numeral. In these cases a classifier is expected for making the lexical NP atomic. The semantics of the numeral is given below (Ionin & Matushansky 2006). (132) 3 = λP∈ D <e,t > . λx ∈ D <d> . ∃S ∈D <et> [Π(S)(x) & |S|=3 & ∀s ∈ S P(s)] Where S is a partition of an entity with non-overlapping cells and the cardinality of S is the numeral. Π(S)(x) = 1 iff S is a cover of x and ∀z, y ∈ S [z = y ∨ ¬∃a [a ≤ z ∧ a ≤ y]] A set of individuals C is a cover of a plural individual X iff X is the sum of all elements in C, such that ⋃𝐶=𝑋 112 For two apples, this function takes a predicate consisting of atomic elements (singular apple), and returns a set that contains partitions/covers with 2 elements each, and each element is an apple. I later show that the concept of partition plays an important role in explaining Bangla numeral constructions. Little modifications to the numeral semantics lead to the best fit of model for understanding the semantics of the numeral constructions in Bangla. 4.3.4 Section summary In this section I showed that both Jiang (2012) and Nomoto’s (2013) accounts explain the numeral constructions of Bangla to some extent. The categorization and function of –Ta, however, still remains unaccounted for, particularly regarding what happens in the absence of - Ta. Although the basic assumptions in these two frameworks are different, basic observational facts of these constructions are well accounted for in these two constructions. Among a few important observations, Jiang cannot account for the mass interpretation of the NP-Ta, while Nomoto cannot explain why –Ta is mandatory with numeral constructions. Additionally, a common concern in both accounts involves the assumption that different types of numerals exist, both within a language and cross-linguistically. I prefer a unified account for numerals (following Ionin & Matushansky 2006), rather than different numeral semantics, and believe that such an approach may be more appropriate cross-linguistically. With this assumption, next I present my proposal. I first remind the reader about the generalization regarding the distribution of –Ta. Then I proceed to relate this distribution cross-linguistically, followed by my proposal, where I present the syntax and semantics of –Ta, and discuss the implications and consequences associated with it. Associating precise counting with partition: Towards the proposal 4.4 I have shown that the absence of –Ta is associated with lack of a counting mechanism (section 4.2). The numeral constructions without –Ta are interpreted as vague or approximated constructions. The absence of –Ta leads to non-monotonic attributive modification in the 113 numeral constructions. –Ta-less measure constructions are interpreted as weak noun phrases and are compatible only with measure pseudo-partitives. The presence of –Ta leads to outputs that require precise counting. Numeral constructions with –Ta are quantifier constructions, and are interpreted to be strong/existential. Moreover, the presence of –Ta in different quantifier structures has been shown to have different effects: counting with count quantifiers, mass measure with the unspecified ones. With correlative structures, the amount denoted by the relative and the correlative quantifiers has to be exactly the same when –Ta appears with the relative pronouns. All these indicate that the presence of –Ta contributes to some kind of exact measurement in its complements. In the following section, I show how a novel proposal for –Ta is mandatory for explaining these puzzling interpretations. Existing accounts of –Ta that converts a kind term to a property (Dayal 2012, Jiang 2012) or, that restricts the denotation of the complement noun to atomic individuals (Nomoto 2013) do not successfully discuss the observations presented in this chapter. Thus, –Ta may not be categorized to be a proto-typical classifier. I propose that it introduces a partition on the complement noun, and restricts the portion to be a singular portion. 4.4.1 -Ta is required for precise counting In O’Connor & Biswas (under revision), using data from pseudo-partitive constructions we show that –Ta is mandatory whenever individuation is implied. Approximation is associated with the absence of –Ta. Li & Rothstein (2011) observe an association between the measure syntax and cardinality approximation in Mandarin Chinese. Measure readings obligatorily include a particle de which is precluded in precise counting. Certain sortal classifiers co-occur with de only when the cardinality is approximate. Rothstein (2011, citing Doetjes 1997) also observes that in Dutch, the individual reading and the measure reading differ by presence of a plural marker. Only the individual reading obtains in the presence of a plural marker. (133) a. twintig liter frisdrank (Measure) twenty liter soda ‘twenty liters worth of soda’ 114 b. twintig liters frisdrank (Individuating only) twenty liter-PL soda ‘twenty liter-units of soda’ c. twintig zak-*(ken) pluimen (Individuating only) twenty sack-*(PL) plums ‘twenty sacks of plums’ These two measurements correspond to two different syntactic structures. (134) a. (Measure) b. (Individuating) According to Rothstein (2009, 2010, 2011), the measure reading have extra syntactic constituent, a measure word such as –ful/-worth, that covertly or covertly composes with the measure noun bottle, and then combines with the numeral, whereas, the individuating reading takes an e-type kind denoting term as an argument of the measure word bottle. 4.4.2 Proposal: Introducing partition Recall that I have the following assumptions: any lexical item enters the derivation as a root, i.e. without a syntactic category. A n 0 indiv maps the denotation of the root to one of predicate of individuals. Particularly, for a count noun, both atoms and sums are available in the denotation. 115 The n 0 , if unmarked, returns a predicate of individuals, suitable for substance mass nouns without containing atomic and non-atomic individuals. (135) a. Count structure: b. Mass structure nP count nP mass n 0 indiv √ n 0 √ I argue that -Ta merges as a Num 0 . It has [+singular] feature and when combined with a predicate, it returns one portion of the predicate. Count nouns, apple for example, allows both atoms and sums in the denotation. when, -Ta merges with the nP, the denotation of ‘-Ta apple’ includes apple singularities, and returns only a set of atomic apples {a, b, c}. ‘-Ta water’ on the other hand, restricts the denotation of water to a singular portion. Since no minimal parts are available for substance masses, only the ‘portions’ of water are available for the partition. The function of –Ta is to partition the denotation of the nouns into relevant portions. For individuated count nouns, (i.e. apple) the denotation is portioned into equal-sized non-overlapping atomic units, where each portion has exactly one atom. For substance mass nouns, the partition contains one portion of the substance (e.g. water). The partition is defined to be a function that introduces non-overlapping covers on pluralities. Following Ionin & Matushansky (2006), the partition function is defined as a function that takes an <e, t> predicate and a plural individual x, yielding an expression true if the predicate is a cover of x with no overlapping parts: (136) Π(S)(x) = 1 iff S is a cover of x, and ∀z, y ∈ S [z = y ∨ ¬∃a[a ≤ z ∧ a ≤ y]] (137) A set of individuals S is a cover of a plural individual x iff ∪S = x (Ionin & Matushansky 2006, pp. 318) The meaning of -Ta can then be defined as shown introducing a partition over an individual with each cell in the partition corresponding to one atom. 19 19 This semantics is different from the classifier semantics of Nomoto (2013). Classifiers are defined as functions from properties to properties that hold singularities alone, while –Ta introduces partitions on the properties. although 116 (138) −𝑇𝑎 = 𝜆𝑃𝜆𝑥 ∃𝑆 Π 𝑆 𝑥 ∧ 𝑆 = 1 ∧ ∀𝑠∈ 𝑆 𝑠 𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑠 𝑜𝑛𝑒 𝑎𝑡𝑜𝑚 &𝑃 𝑠 The syntax of the count noun is as follows. The n 0 indiv categorizes the root to a noun apple, that represents strongly connected individuals divided into a set of atoms and sums. –Ta merges and picks out non-overlapping partitions of apples where each partition contains one apple atom. (139) a. Count structure: [ NumP -Ta[ − plural] [ nPcount [ NP √apel]]] NumP -Ta nP count n 0 indiv √apel ‘apple’ b. −𝑇𝑎 𝑎𝑝𝑒𝑙 = 𝜆𝑥 ∃𝑆 Π 𝑆 𝑥 ∧ 𝑆 = 1∧∀𝑠∈ 𝑆 𝑠 𝑖𝑠 𝑎𝑛 𝑎𝑡𝑜𝑚 𝑜𝑓 𝜆𝑦.𝑎𝑝𝑒𝑙(𝑦) For the substance mass noun, the n 0 categorizes the root to a substance mass noun water, that represents locally strongly connected instances of water. –Ta merges and picks out non- overlapping partitions of water containing one portion of water. (140) a. Mass structure: [ NumP -Ta [ − plural] [ NP jOl]]] NumP -Ta nP mass n 0 √jOl ‘water’ I do not elaborate on it, I assume that the conventional implicature proposed for classifiers will work with the current definition which introduces a conjoined properties denoted by the predicate and the classifier [ λP[P ⊆ CLASS]], where CLASS denotes the conjoined property denoted by the classifier and the noun (Nomoto 2013 pp. 82). This will be helpful in explaining the infelicitous appearances of –Ta in respected human nouns. 117 b. −𝑇𝑎 𝑗𝑂𝑙 = 𝜆𝑥 ∃𝑆 Π 𝑆 𝑥 ∧ 𝑆 = 1∧∀𝑠∈ 𝑆 𝑠 𝑖𝑠 𝑜𝑛𝑒 𝑝𝑜𝑟𝑡𝑖𝑜𝑛 𝑜𝑓 𝜆𝑦.𝑗𝑂𝑙(𝑦) 4.4.3 Continuation: Introducing units in numeral constructions Let us now recall that Bangla overtly dissociates two modes of cardinality measurement: –Ta is found only in the context of the syntax of counting, while the absence of it indicates the absence thereof. Cardinality counting is typically assumed to involve the division of a plurality into atomic parts, which are then added up to determine a value for that plurality along the scale of natural numbers. Counting partitions pluralities into concrete parts and then counts those parts one by one. Cardinality counting is the most natural in situations where the atomic elements of the NP are highly salient. For example, when there is a small and easily countable number of individuals and each atomic part is exhaustively paired with natural numbers, the resulting expression is exact counting (Solt 2016). Constructions without -Ta correspond to interpretations involving no specific representation for the individual atomic elements. The context is most felicitous when individual elements are abstracted away, treating them as a group of items or as a whole. To introduce the numeral into the noun phrase, I assume, following O’Connor & Biswas (Under revision), a phonologically null word unit that combines with the numeral. We proposed that the unit alternates between providing a partition on individuals (counting) and a partition on degrees (approximation). In either case, it partitions cells that correspond to atoms. 20 The semantics of the two types of unit is presented below (See O’Connor & Biswas (under revision) for more details). (141) a. unit 𝐶𝑂𝑈𝑁𝑇 = λP ∈ D <e, t> . λn ∈ D n . λx ∈ D e . ∃S ∈ D <e, t> [Π(S)(x) & |S|=n & ∀s ∈ S [s is an atomic part of P]] 20 This amounts to a claim that it is possible for a set of degrees to have the measure of an atom in quantity. This captures our intuition that it is possible to understand what it means to be an atom at the most abstract of levels – completely independent of the nature of the noun phrase. 118 b. unit 𝐸𝑆𝑇 = λD ∈ D <d,t> . λn ∈ D n . λD’ ∈ D <<d,t>,t> . ∃S ∈ D <dt, t> [Π(S)(D’) & |S|=n & ∀s ∈ S [s measures one atom & s is a proper part of a scale of cardinality]] 4.4.3.1 Numeral constructions with and without -Ta For the counting reading, the unit takes an <e, t> predicate (P) and partitions it into n cells each corresponding to atomic parts of P. In the estimation reading, the partition cell corresponds to a set of degrees, and the noun remains an unindividuated plurality of individuals with a projected collective value in units. In English cardinality constructions, when unit combines with a predicate like apples, the result is a count on a partition into cells corresponding to atomic apples. When unit combines with a predicate that has already been partitioned, as is the case in pounds of apples, the cells correspond to pound-sized units of apples. Similarly in Bangla, when –Ta is introduced, the NP denotation is already partitioned into atomic individuals (for count nouns). Thus, when unit combines with [–Ta apple] the cells correspond to atomic apples, and the numeral (merged after the unit) counts the number of such cells. (142) a. Num-Ta NP: [ NumP unit -Ta[ − plural] [ nP count [ NP root]]] NumP unit COUNT -Ta nP count n 0 indiv √apel ‘apple’ √jOl ‘water’ b. unit 𝐶𝑂𝑈𝑁𝑇 −𝑇𝑎 𝑎𝑝𝑒𝑙 = λn. λx . ∃S [Π(S)(x) & |S|=n & ∀s ∈ S [s is an atomic part of {𝜆𝑦 ∃𝑆[Π 𝑆 𝑥 ∧ 𝑆 = 1∧∀𝑠∈ 𝑆 𝑠 𝑖𝑠 𝑎𝑛 𝑎𝑡𝑜𝑚 𝑜𝑓 𝜆𝑧.𝑎𝑝𝑒𝑙(𝑧) ]}] 119 The numeral three merged with unit COUNT returns three partitions, each of which is an atomic apple. (143) a. Num-Ta NP: [ NumP Num unit COUNT -Ta[ − plural] [ nP count [ NP root]]] NumP three unit COUNT -Ta nP count n 0 indiv √apel ‘apple’ √jOl ‘water’ b. 𝑡ℎ𝑟𝑒𝑒 unit 𝐶𝑂𝑈𝑁𝑇 −𝑇𝑎 𝑎𝑝𝑒𝑙 = λx . ∃S [Π(S)(x) & |S|= 3 & ∀s ∈ S [s is an atomic part of {𝜆𝑦 ∃𝑆[Π 𝑆 𝑥 ∧ 𝑆 = 1∧∀𝑠∈ 𝑆 𝑠 𝑖𝑠 𝑎𝑛 𝑎𝑡𝑜𝑚 𝑜𝑓 𝜆𝑧.𝑎𝑝𝑒𝑙(𝑧) ]}] The [three unit –Ta apel] can be paraphrased as, given an individual x, there is a partition of three individuals, each of which can be portioned into one individual, each of which is an apple. When the individual apple atoms are not directly attended in the absence of -Ta, the unit EST is at work. The partition still measures an atom and is measured on a cardinality scale. But, the atomic structure is used only to form abstract units on a dense scale. The value on this scale must be estimated for the plurality. Such constructions are approximations of cardinality, and therefore the syntax associated with this reading is appropriate in situations where the individual atomic elements are not salient or easily countable. I conjecture that the nature of the abstract units determines whether it selects a –Ta NP or not. The unit COUNT must have atoms and mandates the [–Ta NP] in the complement, whereas, unit EST does not require an atomic structure, results in an estimated outcome. Supporting evidence come from agreement patterns with cardinal containing NPs when the subject is an abstract measure NP (Ionin & Matushansky 2006). Two hours in an alien ship does not have any reference point for measuring the length. 120 Both singular and plural agreement is possible when counting is not allowed, or abstract measure is intended. (144) Two hours in the alien ship was/were clearly not enough. (Ionin & Matushansky 2006: (76)) 4.4.3.2 Optionality of –Ta with large round numbers Bangla large round numbers (starting from 100) allow optionality of –Ta. The presence indicates precise outcome whereas the absence indicates an approximated interpretation/ outcome. Large, round numbers introduce a coarse granularity scale, whereas smaller numbers provide a fine granularity scale. Large numbers may also appear in a fine-grained scale. (145) Figure 1: Granularity of scales Scale-granularity allows the differences in interpretations that are likely to come from the mapping of the nominals on to the scales. When mapping onto a coarse-grained scale, large round numbers may suggest that the quantity is an approximation. Thus, they are likely to be sub-optimal under the syntax of regular numeral constructions in Bangla (i.e. Num-Ta NP). –Ta is required from numerically quantified NP involving numerals one to ninety nine. The numeral SO ‘hundred’ is semi-lexical element, and it naturally allows a coarse scale, without the requirement to access the atoms. In the counting reading, the individual apple atoms are directly accessed and a value on a scale of natural numbers is obtained by counting them one by one. Only –Ta allows access to atoms in the denotation of the noun and such constructions involve exact enumeration, i.e., counting. The syntax associated with the reading is most felicitous in cases where such an operation is plausible (e.g., with lower numbers). Though both 16 Figure 5. (a) A course granularity scale for the round interpretation of 100, and (b) a fine- granularity scale for an exact interpretation of 100 The position of a numeral on its respective scale is relevant for the ANS due to its characteristic ratio-sensitivity. Assuming the scale in Figure 5a is zero-bounded, the first and second values 100 and 200 can be easily discriminated by the ANS, given the 1:2 ratio. By contrast, the difference between the tenth and eleventh values, 1000 and 1100, will be difficult for the ANS to detect at a 10:11 ratio. Similarly, with respect to the fine-grained scale in Figure 5b, the first and second values 1 and 2 may be easily distinguishable for the ANS, but the units 100 and 101 are impossible due to the 100:101 ratio. In other words, the ANS becomes rapidly less able to discriminate two adjacent cardinalities as their position on the scale increases. Because round numbers like 100 and 1000 are underspecified with respect to their scalar granularity, they may either rank very low or very high on a scale, depending on how they are interpreted. This means that whether the ANS can discriminate one cardinality from its next closest value depends on which scale structure is invoked: if 100 is interpreted with respect to the scale in Figure 5a, then it seems feasible that the ANS could discriminate that value from its closest near-neighbors, 0 and 200. But if 100 is interpreted with respect to the scale in Figure 5b, there is simply no way that the ANS could reliably distinguish among 100 and its surrounding values, 99 and 101. How does this relate to judgments of expressions with and without -Ta/-gulo? First, let us momentarily assume a felicity condition on cardinality expressions, such that a cardinality is considered appropriate when we are relatively confident that it is possible to discriminate between that number and its closest relevant near-neighbors; if I say I have 22 apples, one can be fairly confident that I have a good reason for selecting 22 instead of 21 or 23. If I say I have 100 apples, I have good reason for selecting 100 instead of 200, or (on its precise use) 100 from 101. If estimation syntax is associated with cardinality estimation by way of the ANS, we predict its distribution will be constrained by Weber ratios: -Ta will be optional for a cardinality like 100, somewhat more felicitous with a number like 1100, and fully obligatory for a cardinality like 1101, since these numbers vary in terms of their position on the relevant scale and thus in terms of the extent that they can be reliably discriminated. These predictions are confirmed by the data: the inclusion of -Ta is heavily mediated by the extent to which a numeral can be considered an estimate, and that in turn depends on the numeral’s position on a scale. As we have already seen in (8) (repeated in (43)), -Ta can be optionally omitted from a numeral like 100 if the context makes the scale structure in Figure 5a salient, since in such a case a speaker could plausibly distinguish 100 from its closest neighboring values (0 and 200). Page 16 of 37 https://mc.manuscriptcentral.com/js Manuscripts submitted to Journal of Semantics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 121 structures are in principle available, the preference for one over another depends on the meaning and availability of suitable context. 4.4.3.3 Measure constructions We now shift our attention to the obligatory measure interpretations of –Ta-less numerals, and its compatibility with measure readings only. In the absence of –Ta, the unit EST is the most suitable choice. The estimation reading involves mapping a plurality onto a coarsely partitioned scale. For unit EST we defined a degree partition function as shown in (146). Just as the partition on individuals (as in unit COUNT ), the partition S is a partition on a set of degrees D if S covers D and no degree is a member of two elements of S. (146) Π(S)(D)= 1 iff S is a cover of D, and ∀D’, D” ∈ S [D’ = D” ∨ ¬∃d [d ∈ D’ ∧ d ∈ D”]] Measure pseudo-partitives like three pound of apples do quantity estimation, which requires no individuation into atomic apples or pound-sized apple units. The apples remain an unindividuated plurality of individuals with a projected collective value in units. In English, usually the first argument of these measure words is the measure morpheme -worth or -ful, which is typically optionally overt on the estimation reading (and often obligatorily so for non- canonical unit words, e.g. bottles-worth, bowls-worth): (147) [three pounds(worth)] (of) apples In O’Connor & Biswas, we suggest that this morpheme introduces a scale of quantity – possibly a universal scale of measurement (Bale 2008) – and pound provides the relevant information about the dimension, in this case weight. As mentioned, the function of the unit word pound is to provide a partition over the first argument; in this case, pound partitions a scale of weight into equal-sized, non-overlapping pound units. The collective value of the elements of each unit and the granularity of the scale essentially results in the estimation reading. The numeral three provides the number of partitions, and the plurality of apples is mapped onto that scale segment. 122 The resulting expression provides a non-cardinality measure: weight of apples in pounds. Since quantity estimation requires no access to atomic elements, it follows from our account that –Ta is not allowed in measure construction. And, consequently, only the unit EST is merged with the numeral. Independent evidence that number markers are blocked in measure constructions comes from Dutch, as in (133) (repeated below as (148)). (148) a. twintig liter frisdrank (Measure) twenty liter soda ‘twenty liters worth of soda’ b. twintig liter-s frisdrank (Individuating only) twenty liter-PL soda ‘twenty liters of soda’ In O’Connor & Biswas (under revision), we assumed, following standard assumptions, the noun phrase first must compose with a null measure morpheme µ, which introduces a degree argument for the measure phrase to saturate. Thus, Bangla measure constructions can only have measure readings in Bangla. (149) a. tin botol jOl three bottle water ‘three bottles(worth) of water’ b. NumP three unit EST {kilo, liTar}-worth (µ) nP n 0 √apel ‘apple’ √jOl ‘water’ 123 4.4.4 Implications: Explaining the observations 4.4.4.1 -Ta-less numeral constructions as attributive adjectives Similar to the obligatory measure constructions, I asserted that the absence of –Ta in numeral constructions allows them to be non-monotonic attributive modifiers They are non-monotonic on the part-whole structure, i.e., changes in the part-whole structure of the noun phrase do not monotonically correlate with changes to the measure expression. The structure of such constructions is similar to that of the attributive measure constructions. The relevant examples are repeated below as (150) and (151). (150) a. char-(*Te) chaka-r gaRi (attributive) four(*-TA) wheel-GEN car ‘a four-wheeler’ b. aT-(*Ta) pOkeT-er pEnT (attributive) eight-(*TA) pocket-GEN pant ‘a pant with eight pockets’ (151) a. char keji-r apel (attributive) four kg-GEN apple ‘a 4-kg apple’ b. char keji apel (pseudo-partitive) four kg apple ‘4 kgs of apples’ These attributive constructions are structurally similar to English pseudo-partitive structures: a numeral or a weak quantifier is followed by a measure expression, followed by a genitive 124 (similar to English of) and a noun. 21 However, the Bangla attributives are not limited to dimension nouns. Common nouns that cannot generally be measures (i.e. wheel for cars) can be interpreted as measure expressions (150). They are structurally the same as the measure constructions, lacking a –Ta. However, they lack monotonicity on the part-whole structure. They are interpreted as distributive, and a dimensional adjective may substitute the numeral (e.g. bhari chaka-r gaRi ‘heavy wheel-ed car’). They do not involve a measure construction. In the absence of –Ta, the unavailability of atomic structures in the noun phrase allows partitions that are not precise. They compose with the unit COUNT returning a count on imprecise partition of the noun phrase. This numeral construction then enters an English pseudo-partitive like structure, and is interpreted as a measure pseudo-partitive due to the lack of precisely individuated entities in the denotation. It enables common nouns to denote measures, as pocket and wheel in (150). 4.4.4.2 -Ta with quantifiers I remind the reader briefly that Bangla has two types of quantifiers. 22 The following chart represents their combination with –Ta and –gulo, and available interpretations of the quantifier phrase. Quantifiers -Ta -gulo Degree Quantifiers SOb ‘all’ -Ta = Mass -gulo= count Eto/Oto/kOto ‘this/that/how much/many’ -Ta = Mass -gulo= count Onek ‘many/much’/ ‘a lot’ -Ta = Mass -gulo= count Adnominal Quantifier kichu ‘a few/some’ -Ta = Mass ✗ prottek ‘each’ -Ta = Count ✗ kO(y) ‘how many’ -Ta = Count ✗ 21 The distinction between Bangla attributive and (pseudo-)partitive structures is the mirror image of that of English: the appearance of the genitive case (of) is switched. English pseudo-partitive has the of, while Bangla attributive has the analogous genitive case marking following the measure expression. I conjecture that this distribution is due to what the bare nouns denote, such that it allows the complement of the measure phrase in English to be a bare noun (obligatory singular), while Bangla disallows it, and combines the two in a partitive like structure (similar to #colored women vs. women of color). I leave it open for future work. 22 The detail of the two types of quantifiers requires extensive work, and is out of focus of this dissertation. 125 Quantifiers mandatory with -Ta khanik-Ta ✗ ✗ koyek-Ta ✗ ✗ Numeral -- ✗ Table 12: Quantifiers with classifiers and count-mass selection The first type of quantifiers (i.e. degree) is unspecified for mass/count. In English they are specified for non-count, but in Bulgarian, or in Thai the quantifier ‘much/many’ is unspecified for count or mass, like Bangla. (152) a. many boys b. much water b. mnogo momcheta b’. mnogo voda (Bulgarian) ‘many boys’ ‘much water’ c. Onek chele b. Onek jOl (Bangla) ‘many boys’ ‘much water’ However, when -Ta appears with these quantifiers, the quantifiers are restricted only to selection of mass nouns, i.e., the NP receives a mandatory mass interpretation. These quantifiers with – gulo restricts the denotation to only count nouns. The NP receives a mandatory count interpretation, and even mass nouns are also interpreted in measure units, e.g., ‘bottles’ for ‘water’, ‘cups’ for ‘tea’. (153) a. *Onek-Ta chele b. Onek-Ta jOl much/many-Ta boy much/many-Ta water ‘many boys’ ‘much water’ (154) a. Onek-gulo chele b. Onek-gulo jOl much/many-gulo boy much/many-gulo water ‘many boys’ ≈‘many bottles of water’ 126 These quantifiers share properties of degree quantifiers: they appear as adverbial and as adnominal quantifiers (cf. Doetjes 1997). The adverbial reading is unavailable with –gulo. Interestingly, while the bare degree quantifier Onek, without –Ta, quantifies over different ranges of duration, the combination with –Ta allows for only a single duration. I conjecture that sleeping as a verb involves an implicit object ‘sleep’ that is quantified by –Ta and assigned a singular count, resulting in the episodic ‘one stretch’ interpretation for the verb. This is independent supporting evidence that –Ta restricts the denotation to singularities. If it were a classifier, this combination with quantifiers would not be possible. This also suggests that this group of quantifiers is degree quantifiers. (155) a. aj Saradin bacca-Ta Onek ghumiyechhe. today whole-day baby-TA much/many slept ‘The baby slept a lot the whole day today.’ (in different intervals) b. aj Saradin bacca-Ta Onek-Ta ghumiyechhe. today whole-day baby-Ta much/many-Ta slept ‘The baby slept a lot the whole day today.’ (at a stretch) The other group of quantifiers is marked for count nouns (e.g., every, few, each, three etc.), -Ta is obligatory with this group and the combination only allows for a count interpretation. (156) a. *kOyek pakhi b. *kOyek jOl few bird few water c. kOyek-Ta pakhi d ? kOyek-Ta jOl few-Ta bird few-Ta water ‘some birds’ ≈ ‘a few containers of water’ The adnominal quantifiers have an inbuilt unit with them. They are cardinal quantifiers. This line of thought was reflected in Bhattacharya (2000), who proposed these quantifiers incorporate a 127 vague numeral one. The morphological forms of some of these quantifiers show indications of sandhi or concatenation of two morphemes, and they are compositional in nature. Although, the quantifier onek morphologically similar to these group of quantifiers, it is non-compositional in nature (157c). However, etymologically On- in Onek is a negative prefix, and the addition of it leads to ‘not one’ compositionally for the meaning of many. (157) a. kOyek ‘few’= kOy ‘how many’+ ek ‘one’. b. prottek ‘each’ = proti ‘each’ + ek ‘one’ c. Onek ‘much/many’≠ On + ek ‘one’ –Ta is generally mandatory with these quantifiers. Further support comes from their incompatibility with the plural marker –gulo, as does the numeral constructions (e.g., *du-gulo apel ‘two-pl apple’). I argue that Bangla unspecified degree quantifiers ((153)-(155), which can occur both as adnominal and adverbial quantifiers and is underspecified for count/mass, introduces a non- cardinality scale. The adnominal quantifier ((156)-(157)) introduces a cardinality scale. The degree quantifier, as in (153)-(155), can appear regardless of -Ta, while the other requires obligatory presence of -Ta for the expression of measure. I propose that these quantifiers merge at the lower DP domain, in the specifier of the DP. (158) Unspecified Quantifiers a. Onek chhele many/much boy ‘many boys’ 128 Count structure: [ DP Onek [ nP count [ NP root]]] DP ‘many/much’ nP count Onek n 0 indiv √chele b. Onek jOl many/much water ‘much water’ Mass structure: [ DP [ NP root]] QP ‘many/much’ nP mass Onek n 0 √jOl In the presence of the –Ta and –gulo, which project at the NumP domain, the complement NP has a different denotation. Depending on the syntactic structure involved with the count or mass noun, the quantifier construction has different interpretations. (159) Adnominal Quantifiers a. kOyek-Ta chhele few-TA boy ‘few boys’ 129 Q Num -Ta NP: [ DP kOyek [ NumP -Ta [ − plural] [ nP count [ NP root]]] DP NumP kOyek -Ta nP count n 0 indiv √chele √jOl 4.4.4.3 Quantifiers and obligatory mass interpretation Recall that –Ta returns an obligatory mass measure with the degree quantifiers. Similarly, -gulo yields an obligatory count measure. The examples are repeated below. (160) a. *Onek-Ta chele b. Onek-Ta jOl much/many-Ta boy much/many-Ta water ‘many boys’ ‘much water’ (161) a. Onek-gulo chele b. Onek-gulo jOl much/many-gulo boy much/many-gulo water ‘many boys’ ≈‘many bottles of water’ This interpretation is dependent on what –Ta takes as a complement. Since the semantics of –Ta portions out a singular atom or an instance, it has to be compatible with the semantics of many/much. According to Hackl (2001), the lexical entry of many is d-MANY requires it to range over pluralities. When a semantic plural (e.g., Bangla –gulo) is composed with such quantifier, the underlying MANY returns a plural count reading. (162) count-Onek NP: [ QP Onek[ NumP -gulo[+ plural]] [ nP count [ NP root]]]]] 130 QP ‘many /much’ NumP Onek -gulo nP count n 0 indiv √chele √jOl Since these quantifiers are unspecified for many or much, similar to the ‘–MANY’ component, they must have a ‘–MUCH’ component with these unspecified quantifiers. Thus, when –Ta combines with a mass noun, it returns a set of entities that are true of non-pluralities. Combined with the quantifier it returns a mass measure reading. A particularly relevant question that arises at this point concerns the difference in the interpretation of Onek jOl vs. Onek-Ta jOl ‘much water’. Syntactically, the distinction involves the presence of –Ta. Intuitively, the former represent much water which could be a combination of several small bodies of water, while the latter (Onek-Ta jOl) must represent a large body of water. Crucially, this distinction arises because -Ta is a [+sg] number marker, and when combined with mass nouns like water, it returns only a single portion of water, quantified over a non-cardinality scale. Similarly, Onek-gulo is obligatorily count because of the presence of – gulo. In the absence of –Ta or –gulo, no precise count mechanism is implied. Both structures involve a scale of cardinality, while with –Ta/-gulo, it involves a precise mechanism, without them it involves estimation. This is analogous to English measure construction ‘how much toys’, which is only felicitous when the context expects an answer in measure phrases (e.g. ten- truckloads or ten tons). It is infelicitous when a numerical answer is expected from the context. Related to this discussion, what precludes Onek-Ta boy is not clear. Since no mass interpretation is intuitively present with either Onek or –Ta, how does the combination of the two gets an obligatory mass measure? Note that, I argued that –Ta implies the existence of unit COUNT 131 or precise count. Therefore, if –Ta takes a count noun complement, it returns a singular atom, which may be compatible with Onek (except for the semantics for the many in Onek requreis plurality), however, the source of a mass measure is not clear as yet. (163) mass-Onek NP:[ QP Onek [ NumP -Ta[-plural]] [ NP root]]] QP ‘many /much’ NumP Onek -Ta nP mass n 0 *√chele √jOl The prediction that whenever –Ta is present with an unspecified quantifier, it must represent a ‘non-plural’ restriction is borne out in the adverbial Onek-(Ta) cases. In (164) (repeated from (155)) Onek functions as an adverbial, similarly to the use of time adverbial ‘Sunday.’ Even in such cases, we see that when –Ta is absent, multiple time intervals are possible (164a), while the same is blocked in the latter (164b). I argue that this structure has an elided nP, that in the presence (or absence) of –Ta is interpreted differently, percolating in the interpretation of the quantifier. (164) a. aj Saradin bacca-Ta Onek ghumiyeche today whole-day baby-Ta much/many slept ‘The baby slept a lot the whole day today.’ (in different intervals) b. aj Saradin bacca-Ta Onek-Ta ghumiyeche today whole-day baby-Ta much/many-Ta slept ‘The slept a lot the whole day today.’ (at a stretch) 132 (165) Adverbial Onek: [ VP VP [ QP [ NumP –(-Ta)[-plural]] [ nP [ NP Ø]]]]]] VP QP VP Onek NumP (-Ta) nP count 4.4.4.4 Bare classifier and definiteness Bare classifier patterns display cross-linguistic restriction (from Jiang, JL, 2015). 23 As I discussed in chapter 3, in the presence of a strong definite feature, a null definite determiner binds the individual variable. It requires the goal to move to Spec, DP for licensing of the strong definiteness feature and hence we obtain definite readings in (166). This also explains the unavailability of (167a-b). (166) a. jOl-Ta b. chhele-Ta ‘the water’ ‘the boy’ (167) a. *-Ta jOl b. *-Ta chele -Ta water -Ta boy 4.4.4.5 Approximation with numeral doubling 23 Languages with bare classifiers: SVO languages – (i) Min - *bare CL, (ii) Mandarin – indefinite object, (iii) Wu (Fuyang) - definite subject and indefinite object, (iv) Cantonese – definite subject and object. SOV languages – (i) Japanese and Korean - *bare CL, (ii) Yi – Indefinite subject and object, (iii) Bangla – definite subject and object. 133 I argue that these are classic cases of complex numerals, as seen in Ionin & Matushansky (2006). The approximation has nothing to do with the structure of the numerals. The approximation is a result of two different numerals combining in syndetic coordination. Consequently, when the same numeral is coordinated, the resulting interpretation is that of precise counting. The unavailability of –Ta on the first conjunct provide support to the fact that –Ta is not a numeral classifier, else it would coincide with both numerals. (168) a. du-tin-Te apel two-three-TA apple ‘two or three apples’ b. tin-tin-Te apel three-three-TA apple ‘exactly three apples’ 4.4.4.6 The exact measure in correlative demonstratives Since –Ta marked mass nouns are nouns with a singular amount measure, when it composes with correlative pronouns, they output the same amount for both pronouns. This also explains the reason behind the change in category for demonstratives and pronouns. Recall, –Ta with demonstratives create pronouns in Bangla (as discussed in 4.2.3). Since –Ta is a singular marker, when composed with a demonstrative; it restricts the denotation to a singular entity. This, in the context of a demonstrative refers to only one entity, changing the function to a pronoun. The associated demonstrative functions (i.e., deixis, question or anaphoric) remain the same. Summary 4.5 In this chapter, I argue that –Ta is portions out singular portions from its complement. Depending on the nature of the complement, the portion may be an atomic individual (count nouns) or a singular portion (mass nouns). Standard null unit heads are introduced to explain 134 counting and estimation associated with the presence and absence of –Ta. Following our previous work (O’Connor & Biswas, under revision), I show that cardinality constructions and measure construction have similar underlying structures and interpretations, both being available through two different mechanisms: counting and estimation. To achieve this parallelism, a null unit morpheme is introduced above the NumP that provides partitions on individuals and degrees alike, and must be available for combining with numerals. The combination allows the following readings. The interpretations are dependent on meanings and are verifiable upon availability that could be pragmatically obscured. Measurement by counting [pounds/atoms] Measurement by estimating [pounds/atoms] Dimension: cardinality 10 cherries 10(ish) cherries Dimension: non-cardinality 10 [pounds of cherries] [10 pounds(worth)] of cherries Table 13 Combining dimensions and measurements I further show that the current interpretation of –Ta allows us to explain various interpretations, crucially depending on the availability of accessible atoms. The interpretations associated with its presence/absence with various nominal elements like quantifiers, demonstratives may have various interpretations depending on the accessibility. The next chapter discusses the plural counterpart of –Ta and shows that –gulo more or less holds all the properties of –Ta. Thus, these two chapters provide the most comprehensive understanding of these particles that is not available in the literature. 135 5 Chapter 5: Plurality in Bangla Introduction 5.1 This chapter investigates the syntax and semantics of Bangla ‘plural’ classifiers: –gulo and –ra (Dasgupta 1983, Dayal, 2013, 2014). In the previous chapter, I have established that Bangla ‘default’ classifier –Ta introduces a partition containing an atom on the count nouns. Following the same line of thought, I argue that –gulo is the plural counterpart of –Ta. –ra, on the other hand, is an associative plural. The proposal that –gulo is a plural counterpart of –Ta, goes back to typological literature. What is unique in this account is the function that –gulo performs, and it is different from –ra. I argue Bangla has two different plural markers, and these two plural markers are categorically different and their availability is independent of each other (section 5.2). I differ from previous accounts in proposing two different types of plural markers in Bangla. Based on novel observations regarding that the nature of plural marking by –gulo and –ra, I argue that – gulo introduces plural partitions on the noun denotation (section 5.4). –ra on the other hand, is an associative plural and as adjunct on the DP (section 5.5). This proposal does not only inform us about plural marking in Bangla, it also provides clue to how individuation takes place in this language (section 5.7). Difference between the two plurals 5.2 Bangla has two markers of plurality, and they both are generally considered plural classifiers (Dayal 2014 and previous work). The distinction relates what they select: -gulo selects all nouns while –ra is restricted to only human nouns. Typologically –ra resembles the type of plural marker that is often found in other classifier languages (see Iljic 1994, Kurafuji 2004, Li 1999, Nakanishi & Tomioka 2004, for Chinese and Japanese; Chung 2000 for Indonesian, Jenks 2011 for Thai associative plurals). Associative plurals generally attach to proper nouns and are interpreted as a group containing the referent and its associates. These plurals with common nouns are interpreted as regular plurals (e.g. Mandarin men). Similar to Mandarin Chinese men, 136 or Japanese –tati, –ra with a proper noun results in a group reading (Chacón 2011, Biswas 2013). It functions as a regular plural with common nouns. Relevant examples are in (169). (169) a. rik-ra skul-e gElo. Rick-RA school-LOC went ‘Rick and others (his friends) went to school.’ b. bachcha-ra skul-e gElo. child-RA school-LOC went ‘(The) children went to school.’ Plurality with -gulo appears to be similar to the plural interpretations in regular number marking languages. Two exceptions: obligatory definiteness and incompatibility with numeral set –gulo apart from English -s, for example. With common nouns, -gulo results in a definite semantic plural. It is considered semantic plural as it denotes a maximum set of plural entities, excluding the atoms from the domain. The interpretation also holds for proper nouns. Unlike the associative reading of -ra, -gulo with proper nouns are interpreted as multiple individuals with the same name. Relevant examples are in (170). They also differ in terms of their definite interpretation – NP-gulo is obligatorily definite, while NP-ra can be both definite and indefinite. (170) a. rik-gulo skul-e gElo. Rick-GULO school-LOC went ‘The Ricks (Rick and his clones) went to school.’ b. bachcha-gulo skul-e gElo. child-GULO school-LOC went ‘The children went to school.’ These markers have been described as noun markers (Chatterjee 1926) or classifiers (Dasgupta 1983, 1985; Bhattacharya 1999a, 2001; Ghosh 2010; Dayal 2012, 2014). Dissociation between the two forms has been proposed in recent work (Biswas 2013, 2014b). I have argued that –gulo 137 is an exclusive plural marker and –ra is an associative plural. Dayal (2014) defends the ‘classifier’ thesis for both. –gulo and –ra are plural classifiers merged onto the classifier phrase (ClP). –gulo is different from –ra in how they restrict the denotation of the complement noun: – ra is a classifier restricted to animate nouns while –gulo does not. The main objection towards the classifier thesis presented in Dayal (2014) is the incompatibility of –gulo and –ra with numerals. The incompatibility of –gulo with numerals is explained as –gulo excludes atoms from the denotation of the NP, and numerals require atomic elements (cf. Ionin & Matushaksky 2006). However, the incompatibility of–ra with the numeral remains unclear. Secondly, when -gulo appears with a noun, a definite D requires the NP to move to the spec of DP for a strong definite construal in NP-gulo (*–gulo-NP). This movement has also been attributed to an enclitic-like property of –gulo. However, the adjacency requirement of a clitic to its host is a morpho-phonological requirement, and is unlikely to trigger movement in the syntax. Additionally, -gulo is compatible with some quantifiers, while -ra is incompatible with all. Even if we assume feature-specific compatibilities (for example, -ra only compatible with +human features), the incompatibility still remains unexplained. The plurality thesis (Biswas 2013, 2014b) posits –gulo as a plural number marker, while – ra is proposed to be an associative plural. –gulo excludes atomic entities from the denotation of the NP while –ra does not. However, this account wrongly predicts the co-occurrence of –gulo and –ra. In this work, I predominantly adhere to my previous proposal, fine-tuning the specific details based on new observations. I reanalyze –gulo as introducing a plural partition on the denotation of the noun. It does not exclude the atomic entities from the denotation. The apparent semantic plural interpretation comes from the obligatory definiteness associated with –gulo. I elaborate on the ideas in the following section. First I present an overview of the properties of – gulo as per the existing literature (Biswas 2012a, b; Dayal 2014), followed by new observations and the proposal. –gulo 5.3 5.3.1 –gulo and plurality The syntactic proposal varies: -gulo merges at the head of the classifier phrase (ClP) in the classifier thesis (Biswas 2012a, b), whereas, it merges at the head of the number phrase (NumP) 138 in the plurality thesis (Dayal 2014). In the latter account, -gulo takes a kind-level individual (<e k >) and returns an object-level plural individual (<e o >). (171) 〚-gulo 〛 = λx k λy [ ⋃ x(y) ∧ ¬AT(y)] (Dayal 2014: (21a)) Plurality is built into the meaning of –gulo. It has been widely claimed to be an exclusive (i.e. semantic) plural that excludes atomic elements from the denotation of the plural individuals (Biswas 2012a,b, Dayal 2014, among others). This claim comes from the infelicity of NP-gulo in the context of negating a singular. (172) a. A: Please bring your children to the party. B: #I can’t come. I have only one child. b. A: apnar bacca-gulo-ke party-te niye aSben. your child-GULO-ACC party-LOC bring come ‘Please bring your children to the party’ B: tahole ami aSbona, amar-to Ekta-i bacca. then I come-FUT-NEG. my-PRT one-TA-emph child ‘I can’t come then. I have only one child.’ (negation OK in Bangla) (173) a. Have you seen horses in this meadow? b. #kOkhono ei math-e ghoRa-gulo dekhecho? ever this field-LOC horse-GULO saw ‘Have you ever seen the horses in this field?’ However, unlike English bare plurals, the plural interpretation of NP-gulo is not the result of an implicature. NP-gulo is not interpreted as plural due to the presence of an unambiguous singular NP-Ta; rather, the plurality is part of the semantics of -gulo. The English bare plural is only pragmatically interpreted as exclusive plural (e.g. McCawley 1968, Krifka 1989, Rullmann and 139 You 2006, Zweig 2009, Farkas and de Swart 2010 and others). It is an inclusive plural, including both atomic and non-atomic entities in its denotation. Negating the existence of only the non- atomic entities is thus infelicitous. Thus, a question involving bare plural, “do you have children?” could be answered by “Yes, one”, but not by “No, only one.” Similar interpretations arise in the dialogue in (172a) or the question in (173a). In contrast, Bangla NP-gulo excludes the atomic elements from the reference set; hence negating the exclusive plural is felicitous, as shown in (172b). The questions in (173) would render a positive answer in the case of the English bare plural if Sam has seen a single horse in the meadow, Bangla would be negative in the same situation (173b). Only seeing multiple horses would lead to a positive answer. These examples suggest that NP-gulo refers only to a set of plural individuals, and excludes any atomic entities from its denotation. 5.3.2 –gulo and definiteness -gulo marked nominals are interpreted as anaphoric definite and they are strong NPs. They do not appear in predicate positions and do not enter scope relations. (174) a. ora bhalo chatro-(*gulo). they good student-(*GULO) ‘They are good students.’ b. oi haSpatal-Ta nars-gulo-ke khujche. that hospital-TA nurse-GULO-ACC looking-for ‘That hospital is looking for nurses’ It has been observed that plurality in several classifier languages encode definiteness (Jenks 2011, Li 1993, Cheng & Sybesma 2005). Nomoto notes that plural markers in classifier languages may combine with quantifiers, with a subtle meaning difference, and sentences involving such quantifiers may sound ‘awkward’. Consider the following examples from Japanese (see Nomoto 2013: 118, also for Malay and Mandarin). 140 (175) a. Oozei/Hotondo-no gakusei-tati-ga siken-ni oti-ta. many/most-LINK student-PL-NOM exam-DAT fail-PST ‘Many/Most students failed the exam.’ b. Doredake-no gakusei-tati-ga siken-ni oti-ta-no? how.many-LINK student-PL-NOM exam-DAT fail-PST-Q ‘How many students failed the exam?’ However, Nomoto (2013) shows that definiteness is not necessarily associated with plurality. For example, Persian plural –ha is interpreted as definite when a marked indefinite marker (-i) overriding definiteness is absent. The plural marker is licensed either under an indefinite marker –i or under a null definite determiner. Since the definite marker is null, it appears that –ha contributes definiteness. (176) a. bachche-ha-ye bahuS unja bazi mi-kard-an. child-PL-MOD clever there play DUR-do.PST-3PL ‘The clever children were playing there.’ b. bachche-ha-ye bahuS-i unja bazi mi-kard-an. child-PL-MOD clever-INDEF there play DUR-do.PST-3PL ‘Clever children were playing there.’ (Gomeshi 2013, ex. 59-60) The combination of –gulo with quantifiers is domain restricted, and the quantifiers are interpreted as strong quantifiers, However, they do not sound “awkward”. In the presence of a previous context, –gulo with degree quantifiers result in adnominal quantifiers with an obligatory count measure. (177) Onek-(gulo) bacca khelche. Onek-(GULO) kid playing ‘Many (of the) kids are playing.’ 141 As this pattern has been mentioned in previous chapter, here summarize the properties of the quantifiers and their compatibility with –gulo with weak and strong readings. Quantifier Meaning Q-strength Combination Q-gulo strength English equivalent SOb all Strong ✓ Strong ‘the all’ prottek every/each Strong ✗ NA NA beSirbhag 24 most Strong ✗ NA NA numerals 1,2,3… Weak ✗ NA NA Onek a lot, lots of Weak ✓ Strong ‘the many’ kOtok-gulo several Weak ✓ Strong ‘the few’ Table 14: Comparison of quantifier strength with and without -gulo The table suggests that the contribution of definiteness associated with -gulo is similar to the English definite determiner the: It combines with both strong and weak quantifiers, and returns a strong noun phrase. Note that although the definiteness has always been claimed to be a part of the meaning of –gulo, however it might also be due to a syntactic condition. However, the obligatory definiteness associated with –gulo may be syntactic, as proposed earlier in chapter 3. Recall that I proposed two projections for definiteness, and dissociated the anaphoric reading from the unique definite reading. Definiteness is achieved via NP-movement through a lower DP to the Spec, DP to satisfy a [+definite] feature of the null D 0 . Dayal argues, a covert iota operator in the definite DP returns a maximal set of plural individuals. I argue that the lower DP projection introduces contextual domain restriction and is responsible for the discourse- anaphoric reading of NP-gulo. So far, our discussion has contributed to our understanding of exclusive plurality and definiteness in –gulo. In the next section I will present new observations regarding the plurality that suggest that semantic plurality may not be a contribution of –gulo. Rather, the combination of the plurality and obligatory definiteness, particularly the maximality presupposition associated with NP-gulo is responsible for such interpretation. 24 The semantics of most in Bangla is compositional. beSirbhag ‘most’, being a strong quantifier, combines with – gulo only in the ellipsis of the complement noun. E.g., (i) beSirbhag boi-gulo ‘most book-gulo’ (ii) *beSirbhag-gulo boi ‘most-gulo book’. 142 5.3.3 Novel observations 5.3.3.1 Compatibility with sub-kind predicates Recall that NP-gulo denotes at the object-level (cf. Dayal 2014). It is thus incompatible with kind-level predicates. As expected of its object-level definite plural semantics (cf. Dayal 2014), NP-gulo does not appear in characterizing sentences either. Surprisingly, NP-gulo appears in predicates involving sub-kind readings. (178) a. *DoDo-gulo bilupto hoye gEche. (Kind) Dodo-GULO extinct have went ‘Dodos are extinct.’ b. Context: There are two sub-kinds of dinosaurs: X-saurous and Y-saurous (Subkind) x-saurous-gulo bilupto hoye gEche, kintu X-saurous-GULO extinct have went, but y-saurous-gulo-ke Ekhono ekhane-okhane dEkha-jay. Y-saurous-GULO-ACC still here-and-there seen-go ‘X-saurouses are extinct, but Y-saurouses are still seen here and there.’ Nomoto (2012) observed that sub-kind readings of nominal expressions with classifiers are generally precluded in classifier languages because the classifier noun constructions more ‘faithfully’ represent an object reading, as is also predicted of the classifier thesis. The novel observation that –gulo is compatible in subkind predicates thus remains to be explained. The incompatibility with the kind predicates is explained by the contextual salience of the referents. The salience blocks the reference to kind terms. Such salience should also hold for the sub-kind readings, as they must sufficiently represent a kind although not in as much detail as the kind terms. One advantage, if –gulo is a plural marker, is that no such restriction appears to be a problem, and the incompatibility could be delegated to the incompatibility of definiteness and kind terms in Bangla. 143 5.3.3.2 Distributive interpretations and access to atomic elements If NP-gulo excludes atomic entities, it is predicted to have only a collective reading, as any distributive reading would be precluded in the absence of atomic entities. The availability of the distributive reading along with a collective one suggests availability of atomic entities in the denotation. Below I show that the NP-gulo is compatible with many predicates that allow distributive interpretation. Mixed predicates: these predicates allow both collective and distributive interpretations with plural agents. For example, the boys ate a cake can be interpreted as distributive (each boy ate a cake) or collective (the boys ate one cake together). Bangla NP-gulo is compatible with mixed predicates, as shown below. (179) chele-gulo kek kheyeche. boy-GULO cake ate ‘The boys (together) ate a cake.’ (Collective) ‘The boys (each) ate a cake.’ (Distributive) Individual level predicates: These are true of atomic individuals, and they are mandatorily interpreted as distributive. NP-gulo is compatible with such predicates. If atomic elements are not accessible, NP-gulo is predicted to be incompatible with such predicates. (180) a. chele-gulo-r cokh nil. boy-GULO-GEN eye blue ‘The boys have blue eyes.’ b. chele-gulo ghumocche. boy-GULO sleeping ‘The boys are sleeping.’ 144 Distributive sub-entailment: Distributive sub-entailment (DSE) properties hold for a collective predicate when every member of the subject takes part in an event denoted by the predicate (cf. Dowty 1987). For example, gather entails that each member of the plural subject gathers individually (i.e., come to meet at a designated place). If a collective predicate doesn’t have distributive sub-entailment properties, then the event does not break down to each member for taking part in the event individually. For example, a predicate like be a big group or be numerous, which Dowty defined as pure cardinality predicates, does not allow accessing individuals to take part in the event. Thus, a predicate with distributive sub-entailment properties must access the atomic entities in its subject’s denotation. If NP-gulo is compatible with the predicates with distributive sub-entailments, it must allow access to the atomic individuals. The following examples show that Bangla NP-gulo is compatible with predicates with distributive sub-entailment, which in turn suggests that NP-gulo allows access to the atomic entities. NP-gulo is compatible with a collective predicate gather, as shown in (181a). NP-gulo is not fully acceptable with predicates without distributive sub-entailment properties, as in (181b). This is because the obligatory definiteness makes the availability of the atomic parts opaque. (181) a. chatro-gulo klas-e SOmobeto holo. (DSE predicates) student-GULO class-LOC gather be ‘The students gathered in the classroom.’ b. ?chatro-gulo kuRi-jon-er dOl. (no DSE) student-GULO twenty-CLA-GEN group ‘The students are a group of twenty.’ Overt distributive operator: In addition to various distributive readings, NP-gulo is compatible with an overt distributive operator ‘each’ and distributive constructions ‘one by one’. Distributive operators provide distributive construal, and occur with singular nouns. If a distributive operator is compatible with a plural noun, it accesses those atomic entities in its denotation. 145 (182) chele-gulo {protteke/Eke-Eke} ghumiye poRlo. boy-GULO each/one-by-one sleep dropped ‘The boys each fell asleep.’ ‘The boys fell asleep one by one.’ The duality of collective and distributive readings is often attributed to the distributivity operator (cf. Link 1983). Link proposes a covert distributivity operator (D-operator) to account for the distributive readings of the predicates that are ambiguous between collective and distributive readings. For example, the sentence the girls built a raft is ambiguous between a collective and a distributive interpretation. It also accounts for the distributive interpretation in distributive predicates with plural agents (e.g. wear a dress, slept etc.). As a result of the application of the D-operator, the sentence, the girls are wearing a dress, can be interpreted as a sum of girls where each of the girls wear a dress. It takes a predicate P of individuals and returns a predicate that applies to any individual whose atomic parts each satisfies P (Champollion 2012). There have been several modifications of the distributivity operator (D) in the literature; however it has always been implemented for atomic distributivity. It distributes over atoms or singular individuals (Lasersohn 1998, Winter 2001; Champollion 2012). Although, it can be argued that the duality of interpretation in Bangla, as seen above, is due to the operator D, the crucial point is that it needs to access the atomic elements of NP-gulo. This, in turn, suggests that NP-gulo does retain the atomic entities in its denotation. 5.3.4 Implications of accessibility of atomic elements in NP–gulo Now that I have put forward the idea that –gulo retains the atomic entities in the denotation of NP-gulo, implications of such proposal must be in order. If atomic entities are allowed in the denotation of NP-gulo, why does it not combine with numerals, just as –Ta does? To answer this question, I remind that numerals obligatorily require atomic entities in the denotation of the complement NP (also in Dayal 2014). Hence, -gulo does not return a predicate of atomic individuals, numerals do not co-occur with [–gulo NP]. However, recall that I argued ithat numerals may combine with bare nouns denoting atoms and sums, and may receive a vague reading. This account does not explain, why, in the same logic, [-gulo NP] would not be 146 compatible with numerals to yield a vague reading. I conjecture that the vague reading is probably incompatible with the obligatory definite reading associated with –gulo. 5.3.4.1 Scalar count-mass distinction If –gulo is a classifier, it individuates the denotation of substance mass nouns, turning it into a count one. -gulo co-occurs with both prototypical count and mass nouns, and as expected, mass nouns are coerced into count nouns when they co-occur with -gulo. For example, -gulo with a substance mass noun cha ‘tea’ or jOl ‘water’ is interpreted as plurality of a presumed container (e.g., cups of tea, glasses or bottles of water). However, granular aggregates (e.g. flour, rice, sand, sugar etc.) and collective aggregates (e.g. bees, grapes, veggies etc.) behave differently. The referents of these nouns show different degrees of contiguity in the individuated items. Although referents of both granular aggregate and collective aggregate allow access to minimal parts, how the referents are perceived varies by language (cf. Grimm 2012). For collective aggregates, where the designated entities are more “saliently perceived as collectivities, rather than individuals” (Stolz 2001: 65), NP–gulo can be interpreted as referring to the individual items. The granular aggregates on the other hand, can be interpreted as an aggregate of the grains or in terms of measured units, just as the substance mass nouns, shown in (183b-c). Prototypical count nouns provide salient access to individuals, and are pluralized with –gulo, as in (183c). (183) a. jOl-gulo [substance] water- GULO ‘the *(bottles of) water’ b. gOm-gulo [granular aggregate] wheat- GULO ‘the cups of wheat’ ‘the grains of wheat’ c. moumachi-gulo [collective aggregate] bee-GULO ‘the bees’ 147 d. chele-gulo [individuals] boy-GULO ‘the boys’ In a related discussion, Dayal (2014) shows that in the context of three bowls of rice (granular aggregate), bhaat-gulo ‘rice-gulo’ has a counting reading referring to the largest sum of three bowls. In the absence of the bowls in the context, –gulo is precluded. Thus, she arrives at the generalization that the absence of -gulo indicates the absence of individuation. However, modifying the context, I show that such generalization holds only for the substance mass nouns. ‘Granular’ mass nouns (e.g., rice, pebble, pasta, peanut etc.) may occur with –gulo even when a presupposed container is absent. (184) Context: I have excess rice on a plate and some milk in a bowl. I am throwing both away. a. bhaat-gulo phele dicchi. (granular aggregate) rice-GULO throw giving ‘I am going to throw away the rice.’ b. *dudh-gulo phele dicchi. (substance mass) milk-GULO throw giving Intended: ‘I am going to throw away the milk.’ In (184), in the absence of a presupposed container, only the granulated mass nouns are compatible with –gulo (184a), while it is not with the substance mass nouns (184b), showing that individuation is not a function of –gulo. The different interpretations of -gulo with substance mass nouns and ‘granular’ aggregates suggest that these two types of nouns are treated differently in Bangla than in English. English granular mass nouns are classified as mass nouns in English, whereas they show properties of count nouns in Bangla. This parametric distinction alludes to the perception of natural individuation. The ‘distinguishability’ account of Wierzbicka (1988) seems to be relevant in this case. Wierzbicka argues that entities whose constituents are easily distinguishable are more likely to correspond to count nouns. Thus, the count mass 148 distinction depends on a relative scale of distinguishability. This idea has been corroborated in experimental research (Barner & Snedeker 2005 and subsequent work). Middleton et al. (2004) shows that spatial proximity (near or apart) of nonsensical words in visual world paradigm prompts participants’ choice of count or mass syntax. Close proximity inhibits distinguishability, and forces a selection of mass syntax. Similar ideas are found in Grimm (2012), who proposes a scalar mechanism for the count-mass distinction. The availability of both substance-mass and count readings in granular aggregates in Bangla reveals that individuation does not have a binary count-mass distinction in Bangla either. Such interpretation is independent of functional categories like –gulo. If -gulo does not always contribute individuation, then most likely it comes from the n 0 indiv. , the functional head that categorize the noun. The proposal: –gulo introduces a partition containing pluralities 5.4 In Biswas (2014a), I argued that –gulo is a portmanteau morpheme with two functions: plurality and definiteness. -gulo has [+plural] feature, and when combined with a predicate, it returns a predicate that is true of pluralities (i.e. excludes the atomic entities from the domain). Here I modify my previous proposal and argue that –gulo introduces a partition on the appropriate count noun but it does not restrict the denotation to atomic entities (unlike -Ta). I also suggest that the definite interpretation of –gulo is not a part of meaning of –gulo (as a portmanteau morpheme involving semantics plurality and definiteness). Rather, it is purely a syntactic phenomenon associated with the projection of the DPs. Recall that the basic structure of Bangla noun phrases is the following. The count/mass distinction is introduced in the syntax. A n 0 indiv individuates the denotation of the root to create a count noun (including granular and collective aggregates), while a null head returns a mass noun. Note that for –gulo to combine with any noun, it requires being in a count structure. Thus, we conjecture that proto-typical mass nouns are coerced into count ones when –gulo appears with them (e.g. water-gulo ‘the bottles of water’). 149 (185) a. Count structure: b. Mass structure nP count nP mass n 0 indiv √root n 0 √root (186) [ NumP -gulo[+plural] [ nPcount [ NP √apel]]] NumP -gulo nP count n 0 indiv √apel ‘apple’ / ‘water’ Recall that numerals in this account are proposed to merge with unit COUNT that mandates the availability of singular partitions (Ionin and Matushansky 2006). When –Ta is unavailable, atomic elements are unavailable too. In stead, –gulo provides partition on pluralities, without precluding the atomic entities. Thus, in the absence of singular individuals, numerals are incompatible with –gulo. I argued in chapter 3 that Bangla involves two DP projections, as shown below: I argued that the lower DP provides contextual domain restriction. This domain restriction is responsible for the discourse-anaphoric properties of NP-gulo and the existence presupposition in quantifiers with –gulo. The filled higher DP triggers maximality presupposition. Thus, the NP- DP strong wo wo … DP weak wo QP wo D 0 weak NP 150 gulo is interpreted maximally definite. Thus, two factors: domain restriction by a lower D 0 and strong definite feature of a higher D 0 are responsible for the strong anaphoric interpretations of the bare classifier forms. In the absence of the higher D 0 , the quantifiers are can be only restricted to the context, due to the domain restriction by the lower D 0 . 5.4.1 Section summary The first half of this chapter discusses –gulo. I argue in this section that –gulo introduces a partition that is not restricted to an atomic element. In effect, it allows both atomic and non- atomic individuals in the denotation of the noun. I dissociate the definiteness from the meaning of –gulo. I argue that the apparent definiteness associated with –gulo is result of a lower DP projection. I conjecture that –gulo ensures that a lower DP is projected with –gulo. -ra 5.5 In the second half of this chapter, I discuss the structure and interpretations of nominals with –ra. -ra has an associative interpretation with proper nouns (Chacón 2011, Biswas 2013). It acts as a human plural marker when it appears with common nouns. I argue that it adjoins to the right edge of the host DP, and modifies the meaning to yield an associative plural interpretation. It restricts the reference of the DP to (human) animate nouns and contributes to a ‘group’ interpretation where the host is the participant in focus. Semantically, it takes a set of entities, and modifies the denotation to a group that is associated with the focal referent. The animacy restriction of -ra comes as a part of the associative meaning, as cross-linguistically attested for associative plurals. 5.5.1 Associative plural, in Bangla and cross-linguistically An associative plural is defined as designating a set that is referentially heterogeneous: it refers to a group comprising of one focal referent (the named individual of the group) and one or more referents that are associated with the individuals, the ‘associates.’ Associative plurals have a strong preference towards animate nouns, especially human nouns. The associative plurality is 151 similar to a group interpretation in English. 25 Although –ra is mostly referred to as a noun marker, Dasgupta (1985) first observed its similarity with Japanese –tati in the ‘collective’ interpretation. Chacón (2011) defined –ra as an associative plural. In Biswas (2013), I show that –ra has several properties similar to Mandarin men and Japanese –tati. Below I present a comparison of Bangla with Chinese and Japanese, which shows that Bangla –ra shares several properties with associative plurals of Chinese and Japanese (see Iljic 1994; Li 1999 for Mandarin Chinese, also see Jiang 2012. For Japanese, see Iljic 1994; Li 1999 and others). The following chart is updated from Chacon (2011). (187) Properties Chinese men Japanese -tati Bangla -ra Compatible with - Human animate nouns Yes Yes Yes Non-human nouns No No optional in generic sentences Inanimate nouns No Yes No Proper nouns and Pronouns Yes Yes Yes Interpretations - Associative interpretation Proper names and pronouns Proper names and pronouns Proper names and pronouns Additive interpretation Count nouns Count nouns Count nouns (In)definiteness (In)definite (In)definite (In)definite Co-occurrence with Num-classifier Appositive Yes Appositive NPs fronted when co-occurs with Num-Cla Yes (Yes) Yes Can be predicated of - Predicative NP No Yes Yes Existential predicates No Yes Yes Generic and Kind predicates No Optional Mandatory for human kinds Optional for non-human kinds 25 Associative plurality is different from Similative plurality, which denotes a class of objects sharing similar features rather than a group of closely related associates. For example, the gazelles ‘any animal sharing the genus gazella’ or puli-gili ‘tigers and such’ in Telugu, do not have any group-like interpretation. 152 Takes wide scope relative to Negation No Yes Yes Intensional predicates No Yes Yes Subject-object asymmetry No No Yes Table 15: Comparison of associative plurals in Mandarin Chinese, Japanese and Bangla 5.5.2 Differences between the plurals I briefly discussed the difference between the two plurals (i.e. NP-gulo and NP-ra) in the beginning of this chapter. The following table summarizes the properties, differences and similarities, of the two plural forms: NP-gulo and NP-ra (adopted from Dayal 2014). NP-gulo NP-ra Differences Plurality by entailment Plurality by implicature Definite Definite and indefinite Incompatible with generic and kind predicates Compatible with them Similarities Incompatible with numerals Incompatible with numerals Does not appear in predicative position Does not appear in predicative position Table 16 - Comparison NP-gulo with NP-ra Below I present an overview of the available accounts (Dayal 2014 and Biswas 2013) with novel observations supporting (or not) the proposals for –ra. In Dayal (2014), on the basis of the similarities between NPs with -gulo and –ra, these markers are defined to be plural classifiers. Dayal (2012, 2014) extensively discusses the function and interpretation –ra, arguing for a classifier status of –ra (188a). In the context of kind terms, however, she proposes that –ra as an identity function on kind terms that enables them to appear the kind-terms (188b). (188) a. 〚-ra〛= 𝜆x o 𝜆Q ∃X:∀z [z ≤ i X à animate(z).[x ≤ i X ∧ Q(X)] <e k , <e o , t>> b. 〚-ra〛= 𝜆x o 𝜆X:∀z [z ≤ i X à animate(z).[x ≤ i X] <e k , e k > 153 In its first incarnation, the classifier function (<e k , <e o , t>) in (188a) takes a kind individual (<e k >), and returns a set of individuals that are not specified for atomic or non-atomic units. When NP-ra appears in episodic contexts, it undergoes a type shift from a kind level entity to an object level entity through Derived Kind Predication (DKP) so that it can combine with object- level predicates. The NP raises to the Spec of DP and undergoes a type shift by iota, and the NP- ra gets a definite interpretation. The DKP presupposes animacy built into it, and thus –ra is only compatible with human/animate nouns. In the identity function (<e k , e k >) definition of –ra (188b), it takes a kind predicate and returns another kind predicate. Proposal 5.6 In Biswas (2013) I proposed that –ra is an associative plural, categorially different from classifiers and number markers, situated above the DP. In this account, -ra projects a functional projection and takes a DP complement in a functional maximal projection -raP. The DP moves to the Spec, -raP for phonological reasons. It has been claimed in the literature that Bangla -ra has clitic-like properties and requires a lexical item to its left (Ghosh 2010, Dayal 2012, to appear). Unlike NP-gulo, NP-ra can be both indefinite and definite. For the definite reading of NP-ra, the NP moves to the Spec, DP when the D 0 has a strong [+def] feature. The DP then moves to the Spec, -raP to satisfy the phonological requirement. If the D 0 is not specified with a strong definite feature, no definiteness related movement takes place. When the DP moves to Spec, -raP, an indefinite reading arises from an existential closure. Below I present the tree structures corresponding to two readings of the NP-ra (from Biswas 2013). 154 The associative plural interpretation of –ra is coherent as –ra pluralizes pronouns. Pronouns are interpreted to be inherently associated with a hierarchy of speaker-hearer association (Corbett 2000, Moravcsik 2003). Thus, plural pronouns are inherently associative (e.g. we = I and you/others) as –ra is mandatory for pluralization of the Bangla pronouns. (189) a. ami a’. am-ra I we b. tui, tumi, apni b’. to-ra, tom-ra, apna-ra you-[-HON], you, you-[+HON] you-pl The evidence that –ra is situated at the DP-level comes from the available interpretations of conjoined DPs with -ra. However, this also serves as evidence that –ra involves phrasal modification, and not functional projection. When co-occurring with two proper names, –ra appears structurally after the second proper noun. It can therefore be analyzed as appearing with the second proper nouns (e.g. [PN + PN-ra]), or with the highest conjoined DP (e.g. [PN + PN]- ra), which dominates the two DPs. The possible combinations, along with their interpretational differences are shown below. (190) meri ar mira(-ra) aSbe. Mary and Mira-ra will-come scope in the presence of a negative operator; whereas, -ra-marked nouns have wide scope, as in (12). Similar facts are seen in Japanese (Hosoi 2005) and Mandarin (Li 1999). Again, the definiteness or the specific indefinite reading of NP-ra, NP-men and NP-tachi might be at work here. (12) a. ekhane bacca nei b. ekhane bacca-ra nei here child be-neg here child-RA be-neg ‘There isn’t any child here.’ ‘The children aren’t here.’ Definiteness in Bangla is extensively argued relating to the syntax of NP-fronting across the numeral classifier (Dasgupta 1983, Chacón 2010, Simpson 2011, Biswas 2012, Dayal 2012, to appear). The literature follows Bhattacharya (1999 et seq) who first attributed the NP-fronting to specificity. Spec, DP is the landing site of the NP when the D 0 has a strong definite feature. Regardless of the status of -gulo as a classifier (Dayal to appear) or a number marker (Biswas 2013), NP-fronting is the only way to achieve definiteness. The NP-movement for the definite description chele-gulo ‘the boys’ in a layered DP structure is shown in (13). Apparently, similar NP-fronting also applies to -ra, given that it can be associated with a definite interpretation for a common noun. (13) [ DP chele [ NumP [-gulo][ ClP ∅ [ NP chele ]]]] (structure for -gulo, as in Biswas 2013) The fronting is also evident in Mandarin. In both languages, whenever the associative plural marked pronoun or proper noun co-occurs with a numeral classifier, the NP-ra/men must precede the numeral-classifier, as in (14). Since Japanese is a floating quantifier language fronting is not obvious. (14) {rito-ra/ o-ra} tin-jon maTh-e khelche [Bangla] Rito-RA/ he-RA three-cla hum field-loc playing ‘Rito & others, three in number, are playing in the field.’ ‘They, three in number, are playing in the field.’ On the basis of the uniformity of the -ra-marked common nouns with the proper nouns and pronouns with respect to the numeral classifier constructions and their interpretations regarding the associative and the definite readings, I argue that Bangla -ra, and essentially all associative plurals, are merged higher in the DP structure. Following Chacón (2011), I propose that -ra is situated above the DP. -ra takes a DP complement in a functional projection, namely the -raP. The DP moves to the Spec, -raP for phonological reasons. It has been claimed in the literature that Bangla -ra has clitic-like properties and requires a lexical item to its left (Ghosh 2010, Dayal 2012, to appear). The two structures, Figure 1and Figure 2, account for the fact that NP-ra could be definite or indefinite. For the definite reading of NP-ra, as has been discussed earlier, the NP moves to the Spec, DP in the presence of a strong definite feature [+def] in the D 0 . The DP then moves to the Spec, -raP to satisfy the phonological requirement. If the D 0 is not specified with a strong definite feature, the NP does not move to Spec, DP. The whole DP still moves to Spec, -raP. The indefinite interpretation comes from existential closure. Figure 2: Indefinite NP-ra Figure 1: Definite NP-ra 155 a. ‘Mary and Mira will come.’ [PN+PN] 26 b. ‘Mary, and also Mira and others will come.’ [PN + PN-ra] c. ‘Mary and Mira, and their associates will come.’ [PN+PN]-ra d. *[PN-ra + PN] At least three interpretations are available: (i) the conjoined DP consists of two singular proper nouns (e.g. Mary and Mira). (ii) –ra modifies only the second element of the conjunct yielding the referent of the first conjunct singular, and (iii) plurality on the basis of the conjoined DP when –ra attaches to the conjunct. The unavailable structure (d) shows that –ra cannot scope over the first conjunct. This suggests that –ra is an adjunct that modifies the DP to which it adjoins to. An important distinction between –ra and –gulo involve their compatibility in kind and generic predicates. In Biswas (2013), I have argued, following the standard accounts of genericity, the existence of an external generic operator (GEN) for the generic readings of NP-ra. An additional complication involves generic readings in nouns referring to the names of profession, religion, nationalities or other roles in society are defined as capacities (de Swart et al. 2007). –ra is mandatory when these nouns appear in characterizing or kind predicates. (191) shows that –ra is optional with kind terms, while it is mandatory with capacities (192) in kind and characterizing predicates. (191) a. meru-bhalluk-(ra) lupto-pray / Obolupto hoye gEche. (Kind) polar-bear endangered /extinct be went ‘Polar bears are endangered/extinct’ b. meru-bhalluk-(ra) stOnnopayi / SoktiSali hOy. (Characterizing) polar-bear-ra mammal / strong be ‘Polar bears are mammals/strong.’ (192) a. muci-*(ra) lupto-pray / Obolupto hoe gEche. (Kind) cobbler endangered /extinct be went 26 Some participants were skeptical about this interpretation, whereas, others accepted it without any concern. 156 ‘Cobblers are endangered/extinct’ b. muci-*(ra) konjuS / gorib hOe. (Characterizing) cobbler-ra miser / poor be ‘Cobblers are miser/poor.’ De Swart et al. (2007) argues that these nouns (i.e. capacity nouns) have a basic denotation as capacities. They are of type <e>, and are sortally different from kind terms and individuals. They have uniqueness presupposition, and are incompatible with quantificational adverbs like everywhere (e.g., #there is dog everywhere, Dayal 2004). I suggested that these atomic individuals are incompatible in characterizing predicates because the characterizing predicate expresses generalizations about a group. Hence, -ra is mandatory as it maps an atomic individual to a plural individual that can be quantified by the GEN operator to achieve a generic interpretation. The DP-modification explains why –ra is partially incompatible with quantifiers. Note that –ra cannot be string-adjacent to NPs. The higher syntactic position of -ra does not allow composition with a quantificational determiner. However, certain QPs (i.e. Q+NP) are compatible with –ra. NP–ra co-occurs only with degree quantifiers (193a). It co-occurs with a few weak determiners when it appears in a different word order (193b). But it does not co-occur with adnominal quantifiers (193b). (193) a.[{SOb, Onek, beSirbhag} chele] DP -ra 27 {all, many, most} boy-RA ‘all boys’, ‘many boys’, ‘most boys’ b. chele-ra i {SOb-ai, Onek-e, beSirbhag-i, prottek-e} t i boy-RA {all, many, most, each}-AGR 27 I find the combination acceptable only in generic sentences. However, other speakers and a google search attest such combinations in episodic sentences. 157 ‘all of the boys’, ‘many of the boys’, ‘most (of the) boys’, ‘each of the boys’ c. *{kichhu, prottek, kOto} chhele-ra {few, each, how many} boy-RA ‘all boys’, ‘many boys’, ‘most boys’ I argue that (193a) involves –ra attaching to a DP that contains a quantificational degree determiner (e.g. all, many, most) and the NP. (193b) is different from (193a) in two ways – (i) the determiners in (193b) have obligatory agreement morphemes, and (ii) they are partitive constructions. The distribution of Hebrew universal determiner (i.e. kol) provide evidence of a syntactic movement of the NP-ra in (193b). It is evidence for a higher position of –ra. Additionally, here I argue that the compatibility of –ra is also possible because –ra a DP-level object, and hence it is compatible with degree quantifiers. This observation leads to the fact that only NP-ra is compatible with numeral quantifiers in an inverted word order. While NP-gulo is incompatible with numeral constructions, numerals can co-occur with NP-ra only when NP-ra appears in the following word order. Also note, this word construction has a partitive interpretation. (194) a. thana-theke kichu peyada paThalo. police-stn-from few constable sent. ‘The police station sent few constables.’ b. peyeda-ra tin-jon elo constable-ra three-cla hum came ‘The constables, three in number, came.’ This combination is possible only when the NP-ra is anaphoric to an aforementioned entity in the previous sentence. Note that such constructions are interpreted as definite descriptions. 158 5.6.1 Section Summary I differ from previous accounts that –ra is a classifier (Dayal, 2012 2014). I also differ from the proposal that –ra projects a functional projection above DP (Biswas, 2013). I suggest that –ra is a DP-adjunct. The new proposal for –ra accounts for the plurality in pronouns, proper nouns, capacity nouns I also show that this proposal explains the incompatibility of -ra with quantifiers as –ra only adjoins to DPs and degree quantifiers. Consequences of two plural mechanisms 5.7 Although Bangla has an animacy-based split in number marking (i.e. -ra for humans, -gulo for non-human nouns), the predictions of the Animacy Hierarchy (i.e. the plural marker should affect the top section), however, only partially borne out in Bangla. The top segment of the hierarchy consists of the pronouns that are pluralized by the associative plural –ra, while the bottom section containing the common nouns are pluralized by –gulo, prototypical of a plural marker. However, -gulo is involved in pluralizing the third person pronoun, and can be infelicitous in certain human nouns. -ra is specified for plurality in animate nouns only. The Animacy Hierarchy asserts that if a language has number marking, the number distinction must affect the top segment of the hierarchy. The Animacy Hierarchy is represented below, marked with the restrictions of –ra and –gulo. –gulo appears with certain human nouns, but it is always interpreted as having a pejorative connotation. (195) Speaker (1P) > addressee (2P) > 3P > kin > human > animate > inanimate -ra -gulo -gulo Here I show that two-fold distinction actually tells us more about how individuation works in Bangla. As briefly mentioned in chapter 2 and earlier in this chapter, Grimm (2012) explores how number marking is represented cross-linguistically, and how it relates to individuation. Arguing against a binary count-mass distinction, Grimm proposes a scalar alternative that accounts for number marking in cases like granular aggregates (e.g. flour, rice, sand) and 159 collective aggregates (e.g. bee(s), grape(s) etc.), cross-linguistically show different selection of plural marking. The following is a schema of how plural marking is cross-linguistically represented (from Grimm 2013), and how Bangla fares cross-linguistically. (196) Language Liquids/substances Granular aggregates Collective aggregates Individual entities Kiowa I II III IV V VI Dagaare 0 0/singulative 0/singular 0/plural Bangla 0 -gulo -ra Welsh 0 0/singulative 0/plural English 0 0/plural Yudja 0 0/plural Table 17: Individuation across languages and in Bangla This comparison reveals that individuation in Bangla is largely dependent on substance and whole object properties of nouns, particularly human. The nouns represent three classes – substance, whole objects and human whole objects. Summary of the chapter 5.8 In this chapter I have presented a dual account of plural marking in Bangla. I proposed –gulo introduces a plural partition on the noun denotation. A definiteness-related movement of the NP is further required for –gulo satisfying the strong definite feature of the null D in noun phrases. It restricts the domain of the quantifier, and hence the resultant complex quantifier is interpreted as a strong quantifier. I propose –ra as an associative plural, adjoined to the NP. It results in a group like interpretation of the noun phrase. These two types of plural marking are independent of each other. This, ultimately reveals how speakers construe the world. It shows that Bangla nouns are divided on the basis of three properties – substance, whole objects and human objects. 160 Chapter 6: Processing difference in quantifiers involving -gulo This chapter presents an experimental report of different quantifier processing mechanisms in the presence and absence of –gulo. Recent research on the semantics of quantificational expressions like most has demonstrated that experimental research has been successful in revealing the most efficient semantic analysis and verification strategy among competitive theories. Such studies have confirmed the effect of the interface between semantics and cognitive systems, and shown that magnitude estimation serves as a diagnostic for such verification processes (Lidz et al. 2011, Tomaszewicz 2013, Kotek et al. 2015 among others). This chapter presents new experimental evidence in Bangla quantifier processing demonstrating eff fb hects of cardinality and ratio. The aim of this chapter is to verify if –gulo is associated with contextual restriction, which consequently results in different interpretations and if this difference is reflected in different processing mechanisms. In this chapter, I first present the data associated with two quantifier forms involving the presence and absence of –gulo, and the hypotheses associated with the interpretational differences (section 6.2). I then discuss processing implications for two modes of cardinality measurements (section 6.3). After a brief discussion of previous work on other quantifier understanding (section 6.4), I present the methods for this experiment (section 6.6), followed by the analysis and implications of the study. Introduction 6.1 I argue that –Ta and –gulo provide partitions on the denotation of the noun. –Ta provides a singular atomic partition over the denotation of the noun, –gulo on the other hand, does not restrict the denotation to atomic singularities. -gulo may co-occur with certain quantifiers and contributes to a strong, presuppositional count quantifier interpretation. These quantifiers without –gulo are not interpreted as strong presuppositional quantifiers. I argued that the strong, presuppositional interpretation is due to a lower DP projection. When a quantifier merges with the syntactic object containing -Ta and –gulo, the lower DP contributes to domain restriction, anchoring the quantifier domain to the context. Quantifier expressions without –Ta and –gulo do 161 not project a DP, and consequently do not have presuppositional interpretations. The semantic difference, although subtle in regular speech, is testable in certain quantifiers. Take Onek, for example. This cross-categorial degree quantifier may co-occur with nouns, verbs and adjectives. However, when co-occurring with –gulo, it is restricted to the nominal domain only. The following examples illustrate that Onek in its bare form appears with nouns, adjectives, verbs and adverbs (197), whereas, Onek-gulo is mandatorily adnominal (197d). (197) a. ram [Onek-(*gulo) lOmba] hoyeche. (Adjective, *-gulo) Ram ONEK-(GULO) tall became ‘Ram has become very tall.’ b. ajke ami [Onek-(*gulo) ghumiyechi] (Verb, *-gulo) today I ONEK-(GULO) slept ‘I slept a lot today.’ c. gaRi-Ta [Onek-(*gulo) jore] colche. (Adverb, *-gulo) car-TA ONEK-(GULO) speed-LOC running ‘The car is running very fast.’ d. [Onek-(gulo) baccha] khelche. (Noun, (-gulo)) ONEK-(GULO) kid playing ‘Many kids are playing.’ (Onek) ‘Many of the kids are playing.’ (Onek-gulo) Quantifiers co-occurring with –Ta are generally interpreted as mass quantifiers. As explained in previous chapters, such interpretation arises due to the semantics of –Ta restricting the noun denotation to singular portions. Due to this restriction, we consider only –gulo for experimentation in this chapter. In the next section, I provide details of the quantificational properties associated with the two forms Onek and Onek-gulo and set the stage for experimentation. 162 Onek-(gulo): Various contributing factors 6.2 6.2.1 Onek is Cardinal, Onek-gulo is proportional Quantifiers with and without –gulo represent different levels of strength of the DP. As mentioned before, [Onek–gulo NP] is interpreted as a strong presuppositional DP, while the lack of –gulo [Onek NP] renders a weak DP. The following shows that only Onek appears in existential sentences, as expected of a weak DP. (198) a. akaS-e Onek tara ache. sky-LOC ONEK star be ‘There are many starts in the sky.’ b. *akaS-e Onek-gulo tara ache. sky-LOC ONEK-GULO star be Although the closest interpretation of Onek is ‘many’, its cross-categorial distribution makes it similar to English ‘a lot’. Similar to English many, Onek and Onek-gulo represent cardinal and proportional meanings. Existential sentences allow only the cardinal reading, and only Onek is most natural in such sentences. Onek-gulo, in such sentences, must have a preceding context. For example, in the following example, Onek-gulo requires a context such that a speaker mentions extending the invitation to all the kids in the neighborhood, and only Onek-gulo is appropriate in the context. In contrast, Onek is not natural in such context and is appropriate in an out-of-the- blue context. (199) a. Onek bacca parTi-te eSechilo. ONEK child party-LOC came ‘Many children came to the party.’ = The children who came to the party were many. (Cardinal) b. Context: Speaker mentions to the hearer that she invited all the kids in the neighborhood, many of the kids came to the party. 163 Onek-gulo bacca parTi-te eSechilo. ONEK-GULO child party-LOC came ‘Many children came to the party.’ = Of the children, many came to the party. (Proportional) 6.2.2 Onek is high in absolute number, Onek-gulo is contextually restricted Congruent to the cardinal and proportional quantification associated with Onek and Onek-gulo, Onek is understood to be high in absolute number, while Onek-gulo is context-dependent. Onek- gulo has numerical interpretation that is relative to a given context. It often implies a precise number outcome. The numerical estimate of Onek-gulo is not mandatorily a high number, but it is expected to be ‘many’ within a normal range of counting. For example, in a context of a small archival workshop where only a small number of documents are sent to be restored, Onek-gulo may indicate a relatively low number (e.g., 15), yet considered a high number in the context. As shown in (200), in the context of such an archive where 20 old documents were sent for restoration, Onek is unacceptable due to its cardinal reading and expectation of a higher number. Onek-gulo, on the other hand, is possible as it is proportional and a relatively low number (e.g., 15) may still be considered ‘many’ in the context. (200) Context: A staff is reporting the number of document that returned from the workshop. kuRi-Ta dolil paThano hoyechilo. 20-TA document send-PPL were ‘20 documents were sent.’ a. #Onek (dolil) pherot eSeche. ONEK document return came ‘Many (documents) came back.’ b. Onek-gulo (dolil) pherot eSeche. ONEK-GULO document return came 164 ‘Many of the (documents) came back.’ The cardinality quantifier interpretation of Onek and the high number assumptions are related, so as the proportional and relatively low-number assumptions of Onek-gulo. In addition, the numerical construal is context-dependent. The context-dependent cardinality assumption of many is also an issue for the meaning of ‘many’. ‘many’ is also relative to the context and the interlocutors’ perspective. As Egré & Cova (2015) reports, judgment involving many in English varies in terms of the context of moral asymmetries. In a series of related experiments, they show that the numerical estimate varies according to the context. For example, in a context of an accident in a school where a number of children died, many has significantly different cardinal construal in the context of ‘dying’ and ‘surviving’. Similarly, a relatively low number for Onek in general (e.g., 15) may be considered high in a context-sensitive environment. 6.2.3 Size and quantifiability of objects In addition, size and quantifiability of the object also contribute to the interpretation. In the following example, the use of Onek-gulo is ungrammatical for ‘ants’ (201b), which are uncountable under a normal setting. The relative acceptability is a bit better, but still unacceptable for ‘mice’ (201c), which are relatively easy to count, and the acceptance is higher with ‘possums’. (201) Context: In a hotel, the speaker reports about the location. hoTel-Ta khub noNra. hotel-TA very dirty. ‘The hotel is really dirty.’ a. Sumon Onek(-gulo) arSola dekheche. Suman ONEK(-GULO) cockroach saw ‘Suman saw many cockroaches.’ b. Sumon Onek(*-gulo) piMpre dekheche. 165 Suman ONEK-GULO ant saw ‘Suman saw many (of the) ants.’ c. Sumon Onek-gulo pOsam dekheche. Suman ONEK-GULO possum saw ‘Suman saw many (of the) possums.’ 6.2.4 Contributions of -gulo –gulo returns a portion of the denotation of the noun. It also contributes to a DP projection that restricts the domain of the quantifier to the current context. The combination of the semantics of –gulo and contextual domain restriction results in a context-dependent small number interpretation for Onek-gulo. The miscellaneous interpretations of Onek and Onek-gulo create issues for testability. Unlike ‘most’, where a mechanism (i.e., more than half) is available for determining the cardinality, ‘many’ is influenced by context, size and the complement set. Since this dichotomy is similar to the availability of the cardinal and proportional readings of ‘many’, I utilize cardinality and ratio information for understanding available interpretations in experimental design. The current understanding is that the number is understood to be high in absolute terms in Onek, while the proportional reading looks for a relatively large proportion of a set. Although semantically the difference is testable, native speakers often use the forms interchangeably. Existing literature does not recognize any interpretational differences caused by the presence or absence of these particles. Thus, experimental evidence indicating different processes in cardinality expressions would be surprising in the context Bangla, as well as in the wider literature on cardinality measurement, which does not generally predict any important differences between counting and estimation. Since there has been little evidence suggesting the existence of such differences. Thus, it is worthwhile to investigate the semantic differences in these two forms along the lines of cardinality measurement. 166 Two modes of (cardinality) measurement: processing implications 6.3 There are two cognitive systems of numerosity. The parallel individuation or the object tracking system (OTS) is a precise number system for the enumeration of small numbers up to 4. However, mechanism similar to OTS avails in precise counting mechanism over numerals 4. The approximate number system (ANS) is for the estimation of numbers larger than 4. The ANS yields an inherently noisy and approximate number representation, while the OTS creates distinct representations for individual objects. This distinction is well represented in numerical cognition literature (Dehaene 1997; Feigenson et al. 2004 for reviews, Carey 2004; Le Corre & Carey 2007 for details). According to these accounts, the behavioral and neural signatures for these two processes are well defined and mutually exclusive. Counting and estimation of numerosities employ two different underlying mechanisms and they provide different numerical outputs. While the ANS detects differences in magnitude based on the ratio between two sets of objects, the OTS supports computations using the successor function, i.e., it counts. Previous research has shown that when an array of dots is flashed, the enumeration (or estimation) performance is sloppy with larger numbers, whereas it is fast and accurate for small numbers. The ANS is mastered across species (primates, infants, adults, even fish), and is accessed by the infants and adults alike (Whalen, Gallistel & Gelman 1999). Recent works exploring the contact between linguistic semantics and the number systems suggest that the ANS plays a deterministic role for verification of sentences involving enumeration and measure expressions. The numerical distance effect concerns the how two cardinalities are compared: the larger the numerical difference between two cardinalities, the shorter is the time to decide which number is larger. This performance is ratio dependent and is well modeled by the Weber’s law. In their report of a series of psycholinguistic experiments, Lidz et al. (2011 and subsequent work) shows adult participants rely on the ANS for the verification of sentences containing most in numerical estimation. The magnitude estimation, which is the characteristic feature of the ANS, is at play as the accuracy decreases with larger ratios. This work suggests that equivalent semantic representations may involve different verification strategies. These different verification strategies in turn show the involvement of different mental representations in verification processes. Not only mental models, but also localization studies identified brain areas specialized for quantity estimation, quantity 167 comparison and linguistic processing (Heim et al. 2012). These experiments provide extension to semantic theories that are otherwise unavailable for verification. In O’Connor & Biswas (under revision) we outline a possible role for the two modes of measurement, extending the measurement to the cardinality domain. A significant part of our proposal was based on these systems that are external to language, but nevertheless interact with it. The approximate number (ANS) detects numerosities by differences in magnitude or ratio between two sets of objects. The Object Tracking System (OTS) detects numerosities by accessing the atoms and applying a successor function like approach. Individual objects have mental representation for the OTS while the ANS has no representation for individual objects: it maps a plural object onto an inherently noisy number scale. It adheres to Weber’s law, which proposes that perceptual resolution decreases at higher magnitudes. The ANS can more accurately discriminate 4 from 8 dots (a 1:2 ratio) than 8 from 12 dots (a 3:4 ratio). As the ratio of two quantities approaches 1:1 from 1:2, the quantities become increasingly more difficult to discriminate. The OTS is amenable to counting and the capacity is limited to representing a maximum of three individuals in parallel and it is constrained by attention and working memory resources (Heim et al. 2012). Because this system supports computations using the successor function – i.e., for any cardinality n, the next numeral in the sequence represents n+1 – it is likely to underlie knowledge of counting (Carey 2004; Le Corre & Carey 2007). Quantifiers, if allowed both readings, can correspond to two different processing aspects in verification tasks. If the two systems are functionally different, the difference will be manifested in processing tasks. Shikhare et al. (2015) investigated the comprehension and processing of numerical and proportional quantifiers to explore the link between language and numbers. In their experiments they tested if the ANS creates different comprehension strategies for sentences with numerals and proportional quantifiers. In a visual display verification task they found that the ANS is responsible for the evaluation of both numerical and proportional quantifier sentences. Numerical reference information in numerical quantifier helps in selecting the magnitude comparison strategy when processing both types of quantifiers. 168 Previous work on quantifier understanding 6.4 Lidz et al. (2011) in their Interface Transparency Thesis (ITT), promotes the idea that “verification procedures are biased towards the algorithms that directly compute the relations and operations”. According to the ITT, it can then be expected that processing of Onek-gulo take the individual atoms into consideration, while Onek focuses on absolute number. The literature on visual cognition informs us that in counting tasks, visual selection of a target for enumeration “can be influenced by expectation and strategies” (Trick 2008). Thus, under time pressure, for a visual display of 200-500ms, precise counting is impossible. The ANS is engaged in this scenario. It generates a representation of magnitude, the classic signature of magnitude- dependency of the Weber’s law. This paradigm has been successfully implemented in experiments for the quantifier most in English, Polish and Bulgarian (Lidz et al. 2011, Tomaszewicz 2013). In these experiments, a fixed short time presentation of visual stimuli has engaged the ANS, resulting to magnitude estimation, which has shown one of the verification procedures (subtraction) to be the default strategy for most. Likewise, in the current study, we expect the use of ANS for creating different comprehension strategies to evaluate the quantifiers. We strategically block counting, albeit it may seem counter-intuitive as the semantics of –gulo assumes counting. We measure reaction times (RT) for assessing computational complexity of these two quantifiers. Measuring RT is a standard diagnostic in psycholinguistics and it has successfully demonstrated that longer RTs are indicators of complexity. Assuming that full quantifier comprehension engages the visual and linguistic systems with the ANS to evaluate the sets, we predict that the cardinal and proportional readings would vary in a truth value judgment task, and will involve different reaction time estimates. (202) a. Onek DoT holud. ONEK dot yellow. ‘Many dots are yellow.’ b. Onek-gulo DoT holud. ONEK-GULO dot yellow. ‘Many of the dots are yellow.’ 169 To evaluate the quantifiers (i.e., Onek and Onek-gulo) against a visual display consisting of yellow and blue dots, we predict different mechanisms and consequently, different representations in acceptance and reaction times. First, the participant needs to estimate the number of yellow dots, and hold the numerical estimate in short-term working memory. Then depending on the quantifier expression, distinct strategies will be implemented to evaluate the sentences in (202). Based on our understanding of cardinal and proportional quantifiers, Onek is similar to English cardinal many, and is expected to be sensitive to high cardinality. For the cardinal reading (‘Onek dots are yellow.’) in (202a), the numerical estimate of the yellow dots will be evaluated on the basis of an internal numerical criterion, and probably without any reference to the number of blue dots or total number of dots. Thus, we predict higher acceptance ratings for all higher cardinalities in ‘Onek’. For the proportional reading, Onek-gulo is similar to English many of the. For evaluating the sentence in (202b), the estimate of the number of yellow dots will be evaluated/compared against the blue dots. The comparison is sensitive to ratio. A 1:2 ratio is easier than a 1:1 ratio, according to the Weber’s law. The 2:1 ratio is expected to be the easiest. In terms of reaction times, Onek-gulo is expected to have higher reaction times altogether due to the shift to the ANS. –gulo explicitly argues for a counting mechanism, while limited time constraint allows only for estimation, hence a shift appears. This study involves more inter-participant variability due to the nature of the quantifiers, compared to previous studies. Among the other studies, Hackl (2009) compares the proportional quantifier most against more than half, which has a fixed reference point to compare the sets. For many or Onek-gulo, such reference point is missing and it is susceptible to subjective interpretation of the forms. Thus, it is possible that the participants adopt a strategy to focus on the reference set and respond accordingly (Shikhare et al. 2015). The results of this study can be used in future studies that could more accurately represent the semantic distinction. Current study 6.5 The current experiment used the similar experimental paradigm of Lidz et al. (2011). However, the goal of this experiment is to look for converging evidence through the verification procedures involving Onek and Onekgulo. In this probe-before visual verification experiment, participants 170 see these two forms and are asked to verify the truth-value of the sentence against a visual display. Onek predicts that they expect a high cardinality. Onek-gulo on the other hand, predicts availability of a counting structure. However, in a restricted time presentation, they are unable to count, and resort to the ANS in the process. This delays their response for Onek-gulo. If estimation follows Weber’s law, then they have more difficulty in higher ratios than the lower ones. Methods 6.6 6.6.1 Participants A group of 20 Bangla native speakers participated for this experiment. They grew up in West Bengal, where Bangla is used on a day-to-day basis. 12 males and 8 females within the age range 28-38 years, with no visual impediment participated voluntarily. 6.6.2 Materials and design In this online visual display verification task participants were asked to evaluate test sentences against pictures shown. They indicated their response by pressing ‘yes’ or ‘no’ on the computer keyboard while viewing displays of pictures containing blue and yellow dots. Each picture contained a range of blue and yellow dots on a black display. The experiment was designed using an open-source Python-based application (PsychoPy). It’s a 2 x 2 x 3 design with a total of 12 conditions. 2 forms (Onek vs. Onek-gulo) were crossed against 2 levels of cardinality (low cardinality vs. high cardinality) and 3 levels of ratio (1:1, 2:1, 1:2 of yellow:blue dots). The following is a schematic representation of the conditions and number of items in each condition. Ratio (yellow: blue) 1:1 2:1 1:2 Cardinality Low High Low High Low High Onek 6:6 16:16 6:3 16:8 6:12 16:32 Onek-gulo 6:6 16:16 6:3 16:8 6:12 16:32 Table 18: Conditions for the experiment and number of items per condition The pictures were displayed in an array. Each array consisted of 6 pictures representing 6 171 conditions in calculated random order (Latin Square design). 4 repetitions of the total of 12 trials for 2 quantifiers resulted in 288 displays (4 x 12 x 6). The quantifiers appeared alternatively through out the experiment. Due to the strenuous nature of the experiment, the participants were granted self-delimited breaks after each repetition. Each block starts with a prompt for the test sentence. The sentence in Bangla script is presented in the middle of the screen for 1000ms. Once the sentence disappears, a fixation cross appears for 500ms, immediately followed by the visual stimulus, displayed for 500ms. Participants are given 5000ms to indicate their ‘yes’ or ‘no’ response by a button press. They are instructed to respond immediately. After six such displays, a mandatory comprehension question appears that tests if the participant remembers the sentence they saw in that block. This is to ensure that they evaluate the correct sentence against each visual display. Figure 2: Visual representation of one block in the experiment 172 6.6.3 Data analysis 6.6.4 Results The acceptance ratings for each form against the conditions were calculated. The results are summarized in the following table and graph. Ratio (yellow: blue) 1:1 2:1 1:2 Cardinality Low High Low High Low High % of Acceptance Onek 55.63 80.42 66.88 90 32.08 63.92 Onek-gulo 43.75 68.13 68.54 87.08 28.75 56.46 Table 19: Percentage of acceptance for each quantifier against conditions Figure 3: Percentage of acceptance Table: average of RT for each quantifier against conditions Cardinality type Ratio type RT for Onek (ms) RT for Onekgulo (ms) High Ratio 1:1 847.689 892.603 Ratio 2:1 807.590 836.957 Ratio 1:2 921.890 942.008 Low Ratio 1:1 973.992 926.144 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Ra/o 1:1 Ra/o 2:1 Ra/o 1:2 Ra/o 1:1 Ra/o 2:1 Ra/o 1:2 High Low Onek Onekgulo 173 Ratio 2:1 956.701 841.656 Ratio 1:2 902.255 849.984 Table 20: Reaction time for each quantifier against conditions Figure 4: Reaction time for each quantifier Due to the complex design of this experiment, the data analyses were done in several parts. The acceptance rating and the reaction times were analyzed separately, and in combination with separate steps for each ratio. Mixed effects linear regression models were fitted to the data: generalized linear models for categorical acceptance rating task, while the RTs utilized the linear models. Due to convergence errors a fully specified random effects structure could not be used. Random effects were thus included to the model to account for variability across subjects and items. Model comparison was used to select the best-fitted model that optimally improved model fit. 6.6.5 Effects of acceptability The predictions were confirmed partially. Cardinality did not have significant effects on either word. Ratio, however, is partially significant in the processing. The 1:1 ratio was considered the most difficult ratio for interpreting the quantifiers. We predicted Onek would be selected for high cardinality. For the acceptance rating, a mixed effects generalized linear regression model was fitted to the data using the fixed effects for cardinality, quantifiers and their interaction. The 0 200 400 600 800 1000 1200 Ra*o 1:1 Ra*o 2:1 Ra*o 1:2 Ra*o 1:1 Ra*o 2:1 Ra*o 1:2 High Low Onek Onekgulo 174 converged model reveals a main effect of cardinality, although the effect size was descriptively much smaller (β = -0.64, z = -11.73, p < 0.001). A main effect of word was also significant (β = .31, z = 3.93, p < 0.001). No interaction of cardinality and word were found. The difference in cardinality was significant in both quantifiers. The ratio 2:1 was predicted to be favorable to Onek due to higher number of yellow dots. In the same setting of the generalized linear model as in Ratio 1:1, the model reveals a main effect of cardinality (β = -0.81, z = -11.747, p < .001) however, there was no main effect of word, neither were there any interaction. The 1:2 ratio was predicted to be favorable to Onek-gulo due to a relatively favorable proportional visual representation. In the same setting of the generalized linear model as in Ratio 1:1, the model reveals a main effect of cardinality (β = -0.85, z = -12.260, p < .001), and no main effect of word, neither were there any interaction. Both Onek and Onek-gulo in high cardinality shows marginal difference between ratio 1:1 and 2:1 (p < .05) and 1:1 and 1:2 (p < 0.5), while ratio 2:1 and 1:2 showed significant difference (p<.001). Below I present a summary of how significant the difference in acceptability judgment between the two quantifiers is across all conditions. High cardinality Low cardinality Ratio 1 - 1:1 *** *, p =.02 Ratio 2 - 2:1 *** **, p = .001 Ratio 3 - 1:2 Non-significant *** Ratio 1, 2 - (1:1 and 2:1) *** Marginal, p = .06 Ratio 2, 3 - (2:1 and 1:2) *** Non-significant Ratio 1, 3 - (1:1, and 2:1) ** * Table 21: Significance of distribution of different ratios across cardinality and ratio Applying generalized linear models for comparing acceptance ratings of the two quantifiers in various combinations; we arrive at the following generalizations: A. Different selection strategies for the quantifiers are more found in high cardinality, representing bias for higher cardinality in one of the quantifiers. B. The difference between the two quantifiers is relatively marginal in low cardinality, pointing to the acceptability of both forms in low cardinality. 175 C. Ratio 1:2, the best display for proportional reading, shows a significant difference in acceptance in the low cardinality, whereas it is non-significant in the high cardinality cases. D. Combining two exact opposite visual displays (ratio 2:1 and 1:2) shows significant difference in acceptance in higher cardinality, while the difference is non-significant in the low cardinality. E. Combining the two proportional context displays of relative difficulty (ratio 1:1 – difficult, 2:1 - easy) also shows significant difference in quantifier selection in the high cardinality, whereas the difference is marginal in low cardinality. Effects of reaction time Our general prediction regarding the reaction time was that Onek-gulo takes longer than Onek due to inhibition of counting. However, one major confounding factor of this design, and consequently of the prediction is that participants quickly adapt to the enumeration strategy (counting to estimation) and therefore it may not reflect in the RT. When they know that counting would be continually blocked through out the experiment, no expectation for counting is created, and thus, no violation effect for this shift from counting to estimation should be found in the RT. Therefore, the prediction that an overall longer RT for Onek-gulo is not feasible. For the RT, we fitted a regression model to the data using the fixed effects for cardinality, quantifiers and their interaction. The mixed effects linear regression model did not show any main effect of cardinality or word or their interaction in any of the ratios. 6.6.6 Discussion We predicted an overall higher acceptance rate for Onek in higher cardinalities. Onek-gulo was predicted to be ratio-dependent. Additionally, we predicted an overall higher reaction time for Onek-gulo, due to changes in enumeration strategy: resorting to the ANS in the absence of counting. The results indicate Onek is more sensitive to cardinality than Onek-gulo, and is prone to more acceptances in higher cardinality displays. Onek-gulo is more sensitive to ratio. Across the cardinalities, only Onek-gulo is statistically significant in all ratio types. No significant effect was found for the reaction time measurement. 176 The patterns of result in this experiment speak in favor of the hypothesis for contextual restriction. The ratio-dependency of Onek-gulo could be interpreted as a contextual restriction effect. The distinction between Onek and Onek-gulo involve the domain soecific restriction, supposedly supplied by the definite projection invoved with –gulo. The observation that Onek is prone to high cardinality, that the participants are not looking for context-specificty as they do for Onek-gulo. Thus, no numerical comparison is applied while evaluating displays for Onek. When participants see Onek-gulo, they consider the numerical estimate of the whole set or the blue dots. This is due to the semantics of –gulo that restricts the domain of the quantifier to the current visual display. A major drawback of this study is the expectation for the reaction times. However, from this study we may come up with a threshold for the internal numerical reference point for comparing. From the graph we see that the acceptance rating for both of the forms is higher in 2:1 ratio with high cardinality cases. Conclusions 6.7 The results from the experiment provide general assumption regarding the semantic estimate of the two quantifiers. As well, it provides evidence for the domain restriction properties associated with the definiteness in –gulo. 177 Chapter 7: Conclusions and Implications 7 Introduction As I conclude, I briefly summarize the main ideas put forth in this dissertation (section 7.1). I discuss their implications for the number system in general, and in processing, particularly related to numerical processing. This study presents an alternative approach to explain the complex nominal architecture in Bangla, which extends to two similar ‘classifier’ languages (section 7.2). Summary of the dissertation 7.1 This dissertation investigates the classifier particles in Bangla. So far, Bangla has been categorized as a classifier language. In this dissertation, I point out several inconsistencies in the implications of such proposal, and with the help of new observations, I propose an alternative account of number marking associated with these particles. I attempt to convince the reader that the widely acknowledged assumptions about number marking in classifier languages need to be updated. Such idea has been suggested in recent literature (Nomoto 2013). I carry this idea forward by giving a fuller representation of Bangla DP, and additionally I discuss interpretations associated with definiteness within the DP. Three assumptions - (i) numeral classifiers enable counting by individuating an unindividuated mass, (ii) classifier languages lack genuine number morphology, and (iii) classifier languages lack definite projection or definite articles (cf. Chierchia 1998) have been challenged in this dissertation. First, in chapter 2, I show that Bangla bare nouns have several interpretations, ranging from kinds, arguments, indefinite, pseudo-incorporated nominals, and definite descriptions. I present a layered DP structure that syntactically distinguishes pseudo- incorporated, indefinite and definite structures, and consequently explain their semantic interpretations. Additionally, I provide an alternative explanation for the incompatibility of number marking in classifier languages. I argue the availability of number neutrality in bare nouns in classifier languages restrict the expression of morphological plural marking in syntax. 178 The number marking languages, on the other hand, do not display number neutrality in the level of the bare nouns, and hence a morphological plural marker expresses the plural marking. In extension to number marking and representation of the bare nouns, I further investigate the distributions and interpretations of –Ta with numerals, nouns, and quantifiers. The data show that –Ta is crucial for counting interpretations, regardless of the mass count distinction in the nouns. It contributes towards an exact measure in quantifier constructions, including numerical and non-numerical quantifiers. This leads to the proposal that –Ta provides a singular atomic partition to count nouns, while the partition represents a single portion in mass nouns. -Ta restricts the domain of the referent to singularities. To account for the exact cardinality interpretation of –Ta-marked constructions (nominal and quantificational), I argue that it depends on a standard phonologically null functional item unit is merged with –Ta (following O’Connor & Biswas under revision). This account enables us to merge counting and estimation readings in numerical and non-numerical constructions alike. This mechanism of partitioning extends to –gulo, the plural counterpart of –Ta. Arguing against the proposal that –gulo is a plural classifier, I show that the current account explain the data more accurately. –gulo has the same licensing conditions as English –s. Additional interpretations or semantic configurations are a result of additional syntactic components, such as definiteness. I dissect the available interpretations under several categories, and show that (i) – gulo is not a semantic plural marker, and returns both atoms and non-atoms in the denotation of the noun, (ii) the apparent semantic plurality is due to the maximality presupposition that is associated with the definiteness projection that is obligatorily available with –gulo. I argue that – ra is an associative plural, and adjoined to the DP. As –Ta and –gulo are integrated with the definite meanings, I discuss Bangla definite constructions and propose two DP projections for Bangla. I provide a compete repository of definite constructions available in Bangla ranging from bare nouns, nouns with classifiers, marked word orders and their combination with demonstratives. I show that a unified syntactic account for all definite constructions is possible in Bangla. Semantically, I show the effect of the two DPs result in uniqueness presupposition and maximality presuppositions. This explains the binary division of morpho-syntacitc forms available in different contexts. While the maximality requires the NP or QP to be anaphoric, and the uniqueness presupposition satisfies definite interpreations of the bare nouns. I present report of experimental investigation in chapter 6. 179 Cardinality effect of quantifiers goes away when it combines with –gulo, and is interpreted as a proportional quantifier. The change of interpretation in the quantifier is dependent on the contextual restriction introduced by the lower DP projection associated with -gulo. This study has some typological implications – if Bangla is ultimately considered a classifier language, then classifier languages show more fine-grained number system than usually expected. Bangla shows existence of singular and plural number marking properties, and associative plural suffixes, leading to the basis of how individuation takes place in this language. It also shows that the difference between the classifier and non-classifier languages is not semantic. It is a morpho-syntactic difference that can be accounted for by syntactic projections. In addition, it also shows that number and definiteness marking is classifier languages have a more nuanced system. Other closely-related languages in India 7.2 Bangla, Oriya (Odia) and Assamese form a group of languages that share many morpho- syntactic features, and are often referred to as “classifier languages of India”. These languages are spoken in geographically adjacent states in India, and high mutual intelligibility is often found. They are different from rest of the Indic languages in terms of their morpho-syntactic realizations. Among other features, numerically quantified DPs generally have classifier- particles with numerals (Bhattacharya 1999, Ghosh 2001, Chacon 2011). Below I briefly describe how they are different from Bangla in terms of number marking and definiteness, and how the current account provides insights to analyze them. A detailed cross-linguistic comparison must be done before arriving at any generalization. 7.2.1 Oriya (Odia) Between Oriya and Assamese, Oriya is more similar to Bangla in terms of the morphological forms of the classifiers and how they function. For examples, the classifier -Ta/-Ti is required for a numeral when it combines with a noun, and is interpreted as indefinite. There are several ways to represent definiteness in this language, however, the post-nominal –Ta is the default singular definite representation, just as Bangla. Additionally, there are markers for indefinite (-e) and definiteness (-ka and -jaa) (Sahoo1999). NP-fronting is also associated with definiteness. 180 Combination of these particles is more sensitive to the human/nonhuman distinction than that of Bangla (203). The following shows that the particles –Ta combines with numerals in indefinite interpretation (203). It also appears post-nominally without a numeral, and is interpreted as definite (204). The plural markers are not necessarily interpreted as definite, in contrary to Bangla (205). The human-nonhuman distinction is prominent in the case of plural markers too. (203) a. tini-Ta {kukurO/Daktar} b. dui jOnO {*kukurO/Daktar} three-Ta dog/doctor two cl hum dog/doctor ‘three dogs’ ‘two doctors’ ‘three doctors’ (204) a. bOhi-Ta (Oriya) b. boi-Ta (Bangla) book-Ta book-Ta ‘a book’ ‘the book’ c. pila-jOne d. bacca-Ta child-cl hum child-Ta ‘one child’ ‘the child’ (205) a. bOdOmaS {kukur/Daktar}-guDaka b. bOdOmaS {*kukur/Daktar}-mane wicked dog/doctor-pl wicked dog/doctor-pl hum ‘wicked dogs’ ‘wicked doctors’ ‘wicked doctors’ Sahoo (1999) considers –Ta as a classifier, assuming a default singular number and strong definite features associated with a covert article, makes NP–Ta definite and singular. Similarly to Bangla, these markers also combine with quantifiers, however, if there are different interpretations along the lines of Bangla is not known yet, and future work is encouraged. 181 7.2.2 Assamese (Asamiya) Assamese (Asamiya) has a rich set of numeral markers (‘classifier) in comparison to Bangla and Oriya. Similar to Bangla and Oriya, these particles attach to numerals, nominals and demonstratives. They are divided on the basis of animacy restriction. Minute morpho- phonological differences are found with reference to gender, dimensionality and size (Ghosh 2001). The pejorative interpretation with human nouns is strictly maintained in Assamese. Politeness phenomenon is also more prominent in Assamese. Below I show different types of classifiers in Assamese (examples from Goswami & Tamuli 2003). (206) a. dOh-Ta bOlOdh b. dOh-Ti gai ten-TA bull 10-TA cow ‘ten bulls’ ‘ten cow’ c. dOh-zOn manuh d. dOh-zoni manuh ten-cl hum.m. man ten-cl hum.fem. man ‘ten men’ ‘ten women’ Similarly to Bangla, and Oriya, when the classifiers appear after the nouns, the marked order is interpreted as definite description, as shown below. (207)a. gai-zani b. lOra-Ti c. musi-to cow-cl non-human-fem boy-cl hum-dim. cobbler-cl hum-m-[-hon] ‘the cow’ ‘the boy’ ‘the cobbler’ There are many assigned plural markers in Assamese than in Bangla or Oriya, and they are defined along the [±human] distinction. For example, bor/bilak is used with animate and inanimate nouns, but haMt is particularly used for human plurals. It can also have a pejorative connotation when used with specific human nouns. xOkol is used instead when the human nouns require honorific mention. –lok is used for pluralizing pronouns, and it is similar to Bangla –ra. (208) a. manuh-bor b. kitap-bilak 182 man-cl pl book-cl pl ‘the men’ ‘the books’ c. lOra-haMt d. xi-haMt boy-cl pl he-cl pl ‘the boys’ ‘those (people)’ e. xikhyak-xOkol f. tekhet-xOkol teacher-cl pl s/he-cl pl ‘the teachers’ ‘those (ladies/gentlemen)’ g. toma-lok h. eoM-lok you fam -cl pl she-cl pl ‘you’ ‘these (people)’ Ghosh (2001/10) considers these particles classifiers. The definiteness is attributed to a strong feature in the D-domain. The basic difference between the two languages has been suggested to be the availability of formal matching features in the nouns and classifiers. The reason for more nuanced variation in Assamese is because it allows formal features both in the nominal and the classifier. Ghosh also suggests a relative grammaticalization status connecting Hindi with Bangla and Assamese. In that, the grammaticalization process of Bangla is “stunted” at the DP level, whereas Assamese is operative on the semi-lexical frame. Summarizing, these three languages share particles, traditionally defined as ‘classifiers,’ that occur both with and without numerals. The numerically quantified NPs are interpreted as indefinite noun phrases, whereas, without the numeral, they are interpreted as definite descriptions. In their occurrence in the ‘bare classifier’ form (i.e. without the numeral), the complex noun phrase is generally interpreted as singular and definite, with exceptions of additional indefinite markers that results in an indefinite noun phrase (e.g. Oriya). In contrary to the expectations of classifier languages, these languages have one or multiple markers that are particularly assigned for plurality. Additionally, definiteness can be expressed via a combination 183 of markers and word orders. Most interestingly, the numeral-less occurrence with the nominal is generally interpreted as definite. A detailed cross-linguistic investigation is required before deriving any generalization to connect these languages. However, the surface similarities nevertheless suggest that a unified account for these languages is possible. Additionally, the complexities and availability of more nuanced particles in these languages represent a relative status. It represents a scale where Assamese has more nuanced features while Bangla has the least. Following the line of thought in Simpson (2005), I suggest that the nominal repository of available features represent the status of the language in a grammaticalization process. Simpson argues that this grammaticalization shows the use of classifiers without numeral as one of its characteristic feature. Similarly, Ghosh (2001/10) also suggests a relative status of grammaticalization for Bangla and Assamese, in comparison to Hindi, where Bangla is in a relatively advanced stage. Thus, I conjecture that the cross-linguistically a grammaticalization process is available for classifier languages, and this process promotes changes towards less complex morpho-syntactic features. Conclusions 7.3 To explain the varied data patterns in Bangla, I have provided an alternative account where I have claimed the classifiers as number markers. This idea is not radical, as the connection between the two has been asserted over and over again (See Borer 2005, Nomoto 2013 for two opposite accounts). I have provided yet another piece of evidence suggesting the connection between the two. While this loses the cross-linguistic dichotomy regarding the classifier-number marking strategies, it gains a more streamlined account. This ultimately suggests that within the nominal domain, number marking is essential, either by means of classifier selection or number markers, and must be represented overtly by either means. The consequences of this account and cross-linguistic implications remain open for future research. 184 REFERENCES Aikhenvald, Alexandra. 2000. Classifiers. A Typology of Noun Categorization Devices. OUP. Bale, Alan & David Barner. 2009. The interpretation of functional heads: Using comparatives to explore the mass/count distinction. Journal of Semantics 26:3, 217 -252. Barner, David & Jesse Snedeker. 2005. Quantity judgments and individuation: evidence that mass nouns count. Cognition 97(1). 41–66. doi:10.1016/j.cognition.2004.06.009 Bhattacharya, Tanmoy. 1999a. The Structure of the DP in Bangla. London: University of London dissertation. Bhattacharya Tanmoy (1999b). DP internal specificity in Bangla. Yearbook of South Asian Languages and Linguistics 2. 71-99. Bhattacharya, Tanmoy. 1999b. DP internal specificity in Bangla. The Yearbook of South Asian Languages and Linguistics 1999, ed. by Rajendra Singh, 71–99. Thousand Oaks & London: SAGE. Bhattacharya Tanmoy (2000). In search of the vague ‘One’. Cambier-Langeveld, Tina, M. Reford, E. van der Torre (eds.) Proceedings of ConSOLE 7., 33-48. Leiden: SOLE. Biswas, Priyanka. 2011. Reanalyzing the default classifier in Bangla. Talk presented at Formal Approaches to South Asian Languages 2. MIT, Cambridge. Biswas, Priyanka. 2013. Bangla associative plural -ra: a cross-linguistic comparison with Chinese men and Japanese -tachi. Proceedings of WCCFL 31. Biswas, Priyanka. 2014a. Reanalyzing Definiteness in Bangla. In Carpenter, K., O. David, F. Lionnet, C. Sheil, T. Stark and V. Wauters (eds.) Proceedings of the 38th Annual Meeting of the Berkeley Linguistics Society, pp. 19-30. Biswas, Priyanka. 2014b. Plurality in Bangla: Nominal structure and interpretation with emphasis on the interaction of plurality with definiteness. Qualifying paper, University of Southern California. Borer, Hagit. 2005. In Name Only. Oxford: Oxford University Press. Brisson, C. M. (1998). Distributivity, maximality, and floating quantifiers. (Unpublished Citeseer, Rutgers University. Carey, Susan (2004), Bootstrapping & the origin of concepts. Daedalus 133(1): 59-68. 185 Chacón, Dustin A. 2012. Head movement in the Bangla DP. Journal of South Asian Linguistics, 4(1). Chatterjee, Suniti. 1926. The Origin and Development of Bengali Language. Rupa & Co. Champollion, Lucas & Uli Sauerland. 2009. Move and accommodate: A solution to Haddock’s puzzle. Handout CSSP 2009. Cheng, Lisa. L., & Sybesma, Rint. 2005. Classifiers in four varieties of chinese. Cheng, Lisa. L., & Sybesma, Rint. 1999. Bare and not-so-bare nouns and the structure of NP. Linguistic Inquiry, 30(4), 509-542. Chierchia, Gennaro. 1997. Partitives, reference to kinds and semantic variation. Lawson, Aaron (ed.), Proceedings of SALT VII. 73-98. Chierchia, Gennaro. 1998. Reference to kinds across languages. Natural Language Semantics 6, 339-405. Chung, Sandra. 2000. On Reference to Kinds in Indonesian. Natural Language Semantics 8: 157-171. Coppock, Elizabeth & David Beaver. 2012. Exclusivity, Uniqueness and definiteness. Empirical Issues in Syntax and Semantics 9, ed. Christopher Piñón, pp. 1–17 Dasgupta, Probal. 1983. On the Bangla classifier /Ta/, its penumbra and definiteness. Indian Linguistics 44:11–26. Dasgupta, Probal. 1985. On Bangla nouns. Indian Linguistics 46:37–65. Dasgupta, Probal, & Tanmoy Bhattacharya. 1993. Classifiers and the Bangla DP. Papers from the Fifteenth South Asian Language Analysis Roundtable Conference, ed. by Alice Davison & Frederick M Smith, 59–69. Iowa City: South Asian Studies Program, University of Iowa. Dayal, Veneeta. 2012. Bangla classifiers: mediating between kinds and objects. Rivista di Linguistica/ Italian Journal of Linguistics 24.2:195–226. Dayal, Veneeta. (2014). Bangla Plural Classifiers. Language and Linguistics, 15(1), 47-87. Dobrovie-Sorin & Beyssade 2012 (ISBN ISBN: 978-94-007-3001-4m Chapter 4, pg. 136 Dasgupta 1988. Quantification LI paper? Dehaene, Stanislass (1997), The Number Sense: How the Mind Creates Mathematics. New York: Oxford. 186 Elbourne, Paul. 2016. Incomplete descriptions and indistinguishable participants. Natural Language Semantics. 1-43. Etxeberria, U. (2008). On basque quantification and on how some languages restrict their quantificational domain overtly. Matthewson (2008), , 225-277. Feigenson, Lisa, Stanislas Dehaene, & Elizabeth S. Spelke (2004), Core systems of number. Trends in Cognitive Sciences 8(7). 307-314. Ghosh, Rajat. 2001. Some Aspects of Determiner Phrase in Bangla and Asamiya. Tezpur: Tezpur University dissertation. Giannakidou, Anastasia. 2004. Domain restriction and the arguments of quantificational determiners. Paper presented at the Proceedings of SALT 14 110-126. Gillon, Carrie. 2006. n, C. 2006. DP structure and semantic composition in Skwxwœ7mesh (Squamish). Proceedings of the North East Linguistic Society 35, eds. L. Bateman & C. Ussery, 231-244. Gillon, Carrie. 2009. Gillon, C. 2009. The semantic core of determiners: Evidence from Skwxwœ7mesh. In Determiners: Variation and Universals, eds. J. Ghomeshi, I. Paul and M. Wiltschko, 177-213. Linguistik Aktuell. Amsterdam: John Benjamins. Grimm, Scott. 2012. Degrees of Countability: A Mereotopological Approach to the Mass/Count Distinction. SALT 22, 1-15. Hackl, Martin. 2009. On the grammar and processing of proportional quan- tifiers: most versus more than half. Natural Language Semantics 17(1). 63–98. Heim, S., Amunts, K., Drai, D., Eickhoff, S. B., Hautvast, S., & Grodzinsky, Y. (2012). The language- number interface in the brain: A complex parametric study of quantifiers and quantities. Frontiers in Evolutionary Neuroscience, 4. doi: 10.3389/fnevo.2012.00004. Hyde, Daniel (2011), Two Systems of Non-Symbolic Numerical Cognition. In Frontiers in Human Neuroscience 5: 150. Inagaki, Shunji & David Barner. 2009. Countability in absence of count syntax: Evidence from Japanese Quantity Judgments. In Shunji Inagaki, Makiko Hirakawa, Setsuko Arita, Yahiro Hirakawa, Hiromi Morikawa, Mineharu Nakayama, Hidetosi Sirai & Jessika Tsubakita (eds.), Studies in Language Sciences 8: Papers from the 8th Annual Conference of the Japanese Society of Language Sciences, 111–125. Tokyo: Kurosio. 187 Kotek, Hadas, Yasutada Sudo & Martin Hackl. 2015. Experimental investigations of ambiguity: the case of most. Natural Language Semantics 23. 119-156. Krifka, Manfred. 2004. Bare Plurals Kind, indefinite, both neither.Salt13.pdf Le Corre, Mathieu, and Susan Carey (2007), One, two, three, four, nothing more: An investigation of the conceptual sources of the verbal counting principles. Cognition 105(2): 395-438. Li, Xu-Ping. 2011. On the Semantics of Classifiers in Chinese: Bar-Ilan University dissertation. Li, Peggy, YarrowDunham&SusanCarey.2009. Of substance: The nature of language effects on entity construal. Cognitive Psychology 58(4). 487–524. doi:10.1016/j.cogpsych.2008.12.001. Lidz, Jeffrey, Paul Pietroski, Tim Hunter & Justin Halberda (2011), Interface Transparency and the Psychosemantics of most. Natural Language Semantics 19. Matthewson, Lisa 1996, Determiner systems and quantificational strategies Evidence from Salish. University of British Columbia PhD thesis. Matthewson, Lisa. 1999. Wide scope indefinites. Natural Language Semantics 7: 79-134. Milsark, Gary L. 1997. Toward an explanation of certain peculiarities in the existential construction in English. Linguistic Analysis 3: 1-30. Nomoto, Hiroki. 2013. Number in Classifier Languages. University of Minnesota. Obenauer, Hans-Georg (1992), L’interpretation des structures WH et l’accord du participe passé. In: Hans George Obenauer and Anne Zribi-Hertz (eds.), Structure de la phrase et théorie du liage. Paris: Presses Universitaires de France. Pupa, Francesco. 2010. On the russellian reformation. Pholosophical Studies 147: 2, 247 Rett, Jessica. (2014). The polysemy of measurement. Lingua, 143, 242-266. Roberts, Craige (2003). Uniqueness in definite noun phrases. Linguistics and Philosophy 26: 287-350. Schwarz, Florian. 2009. Two Types of Definites in Natural Language, PhD thesis, University of Massachusetts Amherst. Schwarz, Florian. 2013. 2013. Different Types of Definite Crosslinguistically. Language and Linguistics Compass Volume 7, Issue 10, pages 534–559. Schwarzschild, Roger (2002), The Grammar of Measurement. In Brendan Jackson (ed.) Proceedings of SALT XII. NY CLC Publications. 188 Schwarzschild, Roger (2005), Measure phrases as modifiers of adjectives. In Recherches Linguistiques de Vincennes 34: 207–228 Solt, Stephanie. 2016. The case of most and more than half: On measurement and Quantification. Language 92.1: 65-100. Simpson, Andrew. 2005. Classifiers and DP structure in southeast Asian languages. The Oxford Handbook of Comparative Syntax, eds R. Kayne and G. Cinque (Oxford University Press: Oxford), 806-38. Simpson, Andrew, (2011) Hooi Ling Soh and Hiroki Nomoto, 2001. Bare classifiers and definiteness: a cross-linguistic investigation. Studies in Language 35:168-93. Simpson, Andrew & Priyanka Biswas. 2016. Bare nominals, classifiers and the representation of definiteness in Bangla. Linguistic Analysis 40(3-4): 167-198. Stanley Jason & Zoltan G. Szabo. 2000. On quantifier domain restriction. Mind and Language 15 (2&3):219 - 61. Syed, Saurov. 2015. Focus movement within Bangla DP. Proceedings of WCCFL 32. Syed, Saurov 2016. Decomposing Definiteness: Arguments for a Split D-domain in Bangla. Proceedings of WCCFL 33. Simpson, Andrew and Saurov Syed. In press. Blocking effects of higher numerals in Bangla: a phase-based analysis (draft) Tomaszewicz, Barbara, 2013. Linguistic and Visual Cognition: verifying proportional and superlative most in Bulgarian and Polish. In Szymanik, J., & Verbrugge, R., (eds.), Journal of Logic, Language and Information. Springer. von Fintel, Kai. 1999. Quantifier domain selection and pseudo-scope. Cornell Context dependence conference. [http://web.mit.edu/fintel/fintel-1999-cornell-context.pdf] Whalen, John, Charles. R. Gallistel, & Rochel Gelman (1999). Nonverbal counting in humans: the psychophysics of number representation. Psychological Science, 10, 130 – 137. Zamparelli, Roberto. 2000. Layers in the determiner phrase. New York: Garland.

Abstract (if available)

Abstract The study of Bangla nominal structure and the implications thereof have largely been based on the assumptions that Bangla is a classifier language. The main focus of this dissertation is to explore the structure and interpretations of Bangla classifier particles in various noun constructions. Not all of these particles have common properties. I argue that –Ta and –gulo contribute to portioning out portions from the denotation of the noun, modified by the individual semantic functions of these particles, whereas the function of –ra is more of a modifier. This proposal accounts for new observations regarding measure expressions concerning the presence and absence of –Ta, and explains empirical findings concerning the availability of number marking in classifier languages cross-linguistically. The implications of this proposal extend to how nouns combine with numerals and other counting quantifiers alike, and account for the presupposition interpretations in quantifiers and anaphoric definite reference. The questions addressed in this dissertation involve unresolved puzzles regarding count-mass distinction, number marking, quantificational properties and definiteness.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Functional categories: the syntax of DP and DegP

PDF

Where number lies: plural marking, numerals, and the collective-distributive distinction

PDF

A unified syntactic account of Mandarin subject nominals

PDF

Factors affecting relative clause processing in Mandarin

PDF

Direct objects in Persian

PDF

Superlative ambiguities: a comparative perspective

PDF

Towards the unity of movement: implications from verb movement in Cantonese

PDF

Syntactic and non-syntactic factors in reflexive pronoun resolution in Mandarin Chinese

PDF

Comparative iIlusions at the syntax-semantics interface

PDF

Syntactic derivation and the theory of matching contextual features

PDF

Interaction between prosody and information structure: experimental evidence from Hindi and Bangla

PDF

Subjectivity, commitments and degrees: on Mandarin hen

PDF

The grammar of individuation, number and measurement

PDF

Exploring the effects of Korean subject marking and action verbs’ repetition frequency: how they influence the discourse and the memory representations of entities and events

PDF

Binding and scope dependencies with 'floating quantifiers' in Japanese

PDF

Silence in answers: a study of ellipsis in Hindi

PDF

A minimalist analysis of participial constructions: towards a phase account of non-finite structures

PDF

The morphosyntax of states: deriving aspect and event roles from argument structure

PDF

The case of a person: The person case constraint in German

PDF

Perspective in Turkish complementation

Asset Metadata

Creator Biswas, Priyanka (author)

Core Title Number marking and definiteness in Bangla

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Linguistics

Publication Date 12/01/2016

Defense Date 09/28/2016

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Bangla,Bengali,classifiers,definiteness,measure expressions,number marking,OAI-PMH Harvest,South Asian languages

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Pancheva, Roumyana (committee chair), Kaiser, Elsi (committee member), Li, Audrey Yen-Hui (committee member), Simpson, Andrew (committee member), Uzquiano-Cruz, Gabriel (committee member)

Creator Email biswas.priyanka@gmail.com,pbiswas@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-674720

Unique identifier UC11336450

Identifier etd-BiswasPriy-4953.pdf (filename),usctheses-c16-674720 (legacy record id)

Legacy Identifier etd-BiswasPriy-4953.pdf

Dmrecord 674720

Document Type Dissertation

Rights Biswas, Priyanka

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA