Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
The phonology and phonetics of Turkish intonation
(USC Thesis Other)
The phonology and phonetics of Turkish intonation
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
The Phonology and Phonetics of
Turkish Intonation
by
Canan Ipek
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
May 2015
Copyright 2015 Canan Ipek
1
Contents
Acknowledgements 11
Dedication 12
Chapter 1: Introduction 13
1.1 Introduction to the study .............................................................................................13
1.2 Introduction to the AM model of intonational phonology...........................................14
1.3 Phrasal prominence and prosodic typology .................................................................17
1.3.1 Head prominence languages ...............................................................................17
1.3.2 Edge prominence languages ...............................................................................18
1.3.3 Head/Edge prominence languages......................................................................19
1.4 Focus and prosodic prominence...................................................................................20
1.5 Previous work on Turkish prosody..............................................................................23
1.5.1 Word stress..........................................................................................................23
1.5.2 Phrase stress........................................................................................................27
1.5.3 Sentence stress ....................................................................................................29
1.6 Previous intonational models of Turkish .....................................................................34
1.6.1 Kan (2009) ..........................................................................................................35
1.6.2 Kamalı (2011) .....................................................................................................41
Chapter 2: Intonational phonology of Turkish 47
2.1 Tones............................................................................................................................47
2.1.1 Pitch accents: H*, !H*, ^H*................................................................................48
2.1.2 Boundary tones ...................................................................................................72
Chapter 3: Phonetics of Turkish intonation 109
3.1 General introduction ..................................................................................................109
3.2 Experiment 1: Distinguishing H* and H*H- .............................................................110
3.2.1 Method ..............................................................................................................116
3.2.2 Results...............................................................................................................118
2
3.2.3 Experiment 1 discussion ...................................................................................122
3.3 Experiment 2: Distinguishing H*H- and H*Hn.........................................................124
3.3.1 Method ..............................................................................................................124
3.3.2 Results...............................................................................................................127
3.3.3 Experiment 2 discussion ...................................................................................130
3.4 Experiment 3: A perception experiment....................................................................131
3.4.1 Method ..............................................................................................................132
3.4.2 Results...............................................................................................................137
3.5 Local general discussion............................................................................................138
Chapter 4: Summary, discussions, and conclusion 140
4.1 Summary of the intonational phonology of Turkish..................................................140
4.2 Discussions ................................................................................................................141
4.2.1 New findings.....................................................................................................142
4.2.2 Turkish as a head/edge marking language........................................................146
4.2.3 On the relation between nuclear pitch accent and nuclear stress......................149
4.2.4 Concluding remarks and future directions.......................................................156
References 159
3
List of Tables
1 Results for the mean f0 difference under neutral and narrow focus conditions......62
2 Comparison of pitch accents in the three intonational models................................72
3 Results for the mean peak f0 values of high tone on finally stressed words under
neutral and immediately prefocus positions............................................................96
4 List of pitch accents and boundary tones proposed in the current model and their
respective associations ..........................................................................................108
5 Realization of sentence stress in SV intransitives based on eventive versus generic
nature of the sentence ............................................................................................125
6 Peak f0 values for each high tone under three conditions.....................................134
7 Condition of each target sentence in each set........................................................135
4
List of Figures
1 Metrical representation of the sentence ‘Marianna made the marmalade’ .............17
2 Pitch track example displaying pitch range expansion due to focus in English for
the sentence ‘Susan bought a new car’ produced in neutral context (black pitch
contour) and when the adjective ‘new’ is narrowly focused (red pitch contour)....20
3 Pitch accent placement in Turkish according to Kan (2009) ..................................28
4 Sample pitch track from Kan (2009). Pitch contour is divided into three
phonological phrases (PPh) and each of them is hosting one pitch accent .............29
5 Sample pitch track of a three-word sentence from Kan (2009:89) .........................30
6 Derivation of sentence stress culminatively. From Kan (2009:97).........................30
7 Schematic representation of stress domains determined based on phases, CP and
vP. TP and AspP being complements of CP and vP, respectively creates SDs ......31
8 Tonal analysis of ‘Alanyalılar maymunu arıyomuş’ (People of Alanya have been
looking for the monkey). From Kan (2009:84).......................................................36
9 Tonal analysis of ’Ayla Yalova’yı zor bulmuş’ (Ayla had difficulty in finding
Yalova). From Kan (2009:161) ...............................................................................37
10 Tonal analysis of ‘Alanyalılar yaralı maymunu arıyomuş’ (The people of Alanya
have been looking for the monkey, which is injured). From Kan (2009:178) ........38
11 Comparing PPh boundary tones on finally and non-finally stressed words. From
Kan (2009:178)........................................................................................................40
12 Tonal analysis of ‘numaraları limonluya yönlendirmeli’ (He should forward the
numbers to the one with lemons). The prenuclear word, ‘numaraları’ is non-finally
stressed and carries H*L lexical pitch accent. From Kamalı (2011:70) .................42
13 Tonal analysis of ‘bunalanları limonluya yönlendirmeli’ (He should direct those
who get overheated to Limonlu). The nuclear word, limonluya, is non-finally
stressed and carries H*L lexical pitch accent. From Kamalı (2009:73) .................42
14 Analysis of tonal event on finally-stressed prenuclear words. From Kamalı
(2011:68) .................................................................................................................43
15 Analysis of tonal event on finally stressed nuclear word.From Kamalı 2011: 84) .44
5
16 Tonal realization of nuclear pitch accent on a non-finally stressed word. From Kan
(2009:84) .................................................................................................................46
17 Pitch track of a four-word declarative sentence in all-new context ........................50
18 Schematic representation of an ideal L+H* bitonal pitch accent............................51
19 Gradual rise of f0 on a three-syllable finally stressed word....................................52
20 Gradual rise of f0 on a three-syllable finally stressed word....................................52
21 Sample pitch track of a Japanese phrase that shows the realization of an HL tone
unit. From Beckman & Pierrehumbert (1986:264) .................................................53
22 Schematic representation of H*L pitch accent........................................................54
23 Target of L tone.......................................................................................................55
24 Tonal realization of nuclear pitch accent in a five-word declarative sentence
produced in neutral focus context. Fourth word is the nuclear word ......................57
25 Tonal realization of nuclear pitch accent in a three-word declarative sentence
produced in neutral focus context. Second word is the nuclear word.....................58
26 Tonal realization of nuclear pitch accent in a four-word declarative sentence
produced in narrow focus context. Second word is narrowly focused....................59
27 Superimposed pitch tracks for the sentence ‘Emine çayını mendile damlatmış’
(Emine spilled her tea on the handkerchief) under two focus conditions: Neutral
(black pitch track), and when the first word is narrowly focused (red pitch track) 60
28 Superimposed pitch tracks for the sentence, ‘Emine çayını mendile damlatmış’
(Emine spilled her tea on the handkerchief), produced under neutral context (black
pitch track), and when the verb is narrowly focused (red pitch track)....................61
29 Mean f0 values of the stressed syllable (final syllable) of each word under neutral
and narrow focus conditions....................................................................................62
30 Tonal analysis of ‘Ayla yirmi mandalina yedi’ (Ayla ate twenty tangerines).
Nuclear pitch accent, !H*, does not display any rise. Peak f0 on the nuclear word is
realized lower than that of the previous f0-peak. From Kan (2009:184) ................64
31 Tonal analysis of ‘Ayla Yalova’yı zor bulmuş’ (Ayla had difficulty in finding
Yalova). Pitch contour on the nuclear word, zor, displays no rise. Peak f0 on the
nuclear word is almost the same as previous f0 peak. From Kan (2009:161) ........65
6
32 Tonal analysis of nuclear pitch accent when the nuclear word is non-finally
stressed. From Kamalı (2011:75) ............................................................................66
33 Tonal analysis of a finally stressed nuclear word. From Kamalı (2011:78) ...........67
34 Non-finally stressed nuclear word with no f0-rise ..................................................68
35 Finally stressed nuclear word with slight f0-rise.....................................................68
36 Pitch track of a five-word interrogative sentence. Nuclear pitch accent is realized
with expanded pitch range.......................................................................................70
37 Pitch track of a yes/no question. The word the question particle is attached to hosts
the high tone with the highest peak in the entire sentence ......................................71
38 The subject NP, fakülteler, and the object NP, lokantalarını, each contain one non-
finally stressed word. The right edge of each NP is marked by a rising tone, LH- 74
39 The subject NP, lokantanın sandalyeleri, contains two words with non-final stress.
The end of the second word, which is the last word of the subject NP, is marked by
a rising tone, LH-.....................................................................................................74
40 The subject NP contains three words. All words in subject NP have non-final stress
and only the last syllable of the NP is marked by a rising tone, LH-......................75
41 Non-finally stressed postposition, aracılığıyla, hosts to rising tones, one on its
stressed syllable marked by H* and the other on its final syllable marked by LH-
which demarcates the right edge of PP. ‘I will send the money to the customer
through Leman on Friday’.......................................................................................76
42 The LH- boundary tone marks the end of the subject NP ‘bembeyaz sandalyeler’,
which conrresponds to an ip ...................................................................................77
43 The subject NP includes an adjective phrase, tamamıyla bembeyaz ‘completely
white’ before a head noun, but the adjective phrase-finally syllable is not marked
by any tone. LH- tone marks only the en of the subject NP, forming an ip............77
44 Right edge of noun clause is marked by a rising tone, LH-. Pitch track of a subject
NP clause is extracted from the sentence Alanyalıların misafirleri ağırlamamaları
(dün bazılarını endişelendirdi) (That people of Alanya did not host the guests
(upset some yesterday)). Also note that the end of the first word, NP, is marked by
an ip bounary, LH-. .................................................................................................78
7
45 Sample tonal realization of an if clause. Extracted from the sentence Almanyalılar
Hollandalıları yenerse, Brezilya gruptan çıkar (If Germans beat the Dutch, Brazil
will move to the next round) ...................................................................................79
46 Pitch track of the phrase ‘The man who went to Alanya’s village’ in (19a) shows
lack of tonal marking at the right edge of a subject relative clause. .......................80
47 Pitch track of the phrase “The goal that Germany scored against Holland” in (19b)
shows lack of tonal marking at the right edge of an object relative clause.............81
48 The subject NP contains two finally stressed words both of which host a high tone.
The phrase medial high tone is marked by H* pitch accent indicating its phrase-
medial status, while the phrase final one is marked by H*H- tone .........................83
49 Sentence initial adjective yedi (‘seven’) is phrase medial and hosts L+H* pitch
accent. From Kan (2009:182)..................................................................................84
50 Sentence initial adjective yedi (‘seven’) forms its own PPh and its right edge is still
marked by a L+H* pitch accent. From Kan (2009:182) .........................................85
51 Two prenuclear high tones on a finally stressed word are marked as H-. From
Kamalı (2011:68) ....................................................................................................86
52 Using non-finally stressed words to display declination within an ip/NP ..............87
53 Each word within subject NP is finally stressed. The phrase medial nouns within
subject NP host a H* pitch accent and they undergo declination. The phrase final
high tone, which is marked by H*H- boundary tone, reaches a slightly higher level
than the preceding high tone ...................................................................................88
54 Schematic representation of prominence marking in declaratives..........................89
55 The non-finally stressed word, tamamıyla, does not immediately precede the
nuclear word and hosts a single High tone realized on its stressed syllable only ...90
56 The non-finally stressed word, tamamıyla, immediately precedes the nuclear word
and shows double f0 peaks, one on its stressed syllable and the other on its final
syllable.....................................................................................................................91
57 In the two-word subject-NP, alanyalı gemiciler, the non-finally stressed firstword,
alanyalı, hosts a post-lexical pitch accent realized on its stressed syllable.............92
8
58 In the two-word subject-NP, alanyalı gemiciler, the non-finally stressed first word,
alanyalı, carries a high tone on its final syllable when it immediately precedes the
narrowly focused word............................................................................................93
59 Superimposed pitch tracks of the sentence, Neriman Eylülü limana göndermiş,
produced under four focus conditions .....................................................................95
60 Mean peak f0 values of high tone on finally stressed words under neutral and
immediately prefocus conditions.............................................................................95
61 Sample pitch track of an SV intransitive sentence preceded by a PP. “Babies sleep
towards the evening”. Nuclear pitch accent is realized on the verb........................97
62 Sample pitch of a SV intransitive sentence preceded by a PP. “Babies sleep
towards evening”. Nuclear pitch accent is realized on the Subject.........................97
63 Pitch track of a four word sentence from Kamalı (2011:68). Third word is the
nuclear word and immediately prenuclear high tone is higher than the initial high
tone, however both prenuclear high tones are marked by H-..................................99
64 Pitch track of a four word sentence where the third word is the nuclear word. High
tone on the immediately prenuclear word is higher than that of the first prenuclear
high tone. from Kan (2009: 184) ..........................................................................100
65 Pitch track of a wh-question. Left edge of the wh-word is marked by Ln tone ....101
66 Sample pitch track of a five-word declarative sentence........................................102
67 Pitch track of a yes/no question and a low IP-boundary tone (L%) marks its right
edge .......................................................................................................................103
68 Pitch track of a tag question, “Leman’s nephew is going to Germany by car, isn’t
he?”, and a low IP boundary tone (L%) marks its right edge ...............................104
69 Pitch track of an imperative sentence, ‘Call Alanya!’, and a low IP-boundary tone
(L%) marks its right edge......................................................................................105
70 Pitch track of a wh-question and a high IP-boundary tone (H%) .........................106
71 Pitch track of the first part of a coordinate clause, “Leman took Kemal to the
Restaurant and (went to her friend from there)”,. A high IP-boundary tone (H%)
marks its right edge. ..............................................................................................107
72 Tonal analysis of a three word declarative sentence. The first word, Ayla, is
lexically stressed on its final syllable, and forms one PPh. Rising contour on the
9
first word is marked by bitonal L+H* which also marks the right edge of PPh.
Taken from Kan (2009:155)..................................................................................112
73 Tonal analysis of a four word declarative sentence The first word, yaralı, is
lexically stressed on its final syllable and is phrase medial within a PPh (yaralı
yalovalılar). The rising tone on this PPh phrase medial tone is marked by bitonal
pitch accent L+H*. From Kan (2009:176) ............................................................113
74 Tonal analysis of finally stressed words according to Kamalı (2011) ..................114
75 Distinction of phrase medial high tone vs phrase final high tone on finally stressed
words according to current model .........................................................................115
76 Magnitude of mean f0 rise at Prosodic Word and Intermediate Phrase boundaries ...
...............................................................................................................................119
77 Sample pitch track of the sentence (22a) illustrating H* at the end of 2
nd
word
(tuzak), i.e., Prosodic Word boundary...................................................................120
78 Sample pitch track of the sentence (22b) illustrating H*H- at the end of 2
nd
word
(tuzak), i.e., Intermediate Phrase boundary...........................................................120
79 Rhyme duration (ms) of PW-final syllable (H*) vs. ip-final syllable (H*H-) ......121
80 Euclidean distance (Hz) between vowel across a PW boundary (H*) vs. an ip
boundary (H*H-) ...................................................................................................122
81 Pitch track of the sentence “Restaurant’s chairs will be painted in blue on Friday.”
The second word (sandalyeleri) is ip-final and its stressed on its second syllable.
The second f0 rise on this word, labeled LH-, marks the right edge of ip ............123
82 Comparison of magnitude of f0 rise in the H*H- vs H*Hn contexts ....................128
83 Pitch track of the sentence “Towards the evening, babies sleep”. The sentence has
a generic reading of the verb, uyur, and the verb is nuclear accented...................129
84 Pitch track of the sentence “Towards the evening, fish burnt”. The sentence has an
eventive reading of the verb, and the subject, balıklar, is nuclear accented .........129
85 Rhyme duration in the H*H- versus H*Hn contexts.............................................130
86 Pitch contour of a PP + S + V sentence with an eventive reading, and its tonal
analysis according to the current model ................................................................132
10
87 Pitch track of a target sentence where f0 peak of H*Hn in the original sentence is
lowered. It is lower than the first f0 peak, H*, but still higher than the nuclear pitch
accent, !H*.............................................................................................................133
88 Pitch track of a target sentence where the nuclear pitch accent !H* is ‘deleted’ ..134
89 Frequency of correct identification of the most prominent word under three
conditions ..............................................................................................................137
90 Schematic representation of the marking of sentence prominence in Turkish
declarative sentences compared to other languages ..............................................148
91 Pitch track example displaying pitch range expansion on the narrowly focused
word in English .....................................................................................................151
92 Mean vowel duration of the stressed syllable when it carries the nuclear pitch
accent (+NPA) versus when it is not (-NPA)........................................................155
11
Acknowledgements
This PhD has been such an enriching experience in many ways and a lot of people have
contributed to make it an amazing journey. I am not going to give a long list of those people; you
know who you are. However, I must especially thank to Maria Luisa Zubizarreta, Sun-Ah Jun,
and Hagit Borer; this journey could definitely not have been completed without them and their
constant support. My most special thanks to Leyla Ilgın; this journey wouldn’t have even started
if she didn’t introduce me to the field of linguistics.
Let’s see where this journey is going to take me next.
12
Dedication
To my mom, my sister, and my dad
I love you more than any words can express.
13
Chapter 1: Introduction
1.1 Introduction to the study
The prosodic structure of Turkish and its intonational properties is a highly understudied
research field in the literature of Turkish linguistics. There are only a handful of studies that
touched upon this issue, and still majority of them are not concerned with proposing a
comprehensive phonological model (e.g. Nash 1973, Ozge 2003, Levi 2005, Kan 2009, Kamalı
2011). Rather, the focus has been mostly on global aspects of intonation of particular clause
types or of informational structural units without delving into the details of internal structure of
tunes. Among those that deal with the relation between intonation and clause types, some
recurrent observations are that a declarative sentence is realized by a falling pitch at the end,
while rising pitch contour marks the right-edge of a wh-question or an incomplete clause, and a
polar question is marked by a rise-fall pitch pattern (Demircan 1983, Özsoy 2004). The relation
between informational structural units like theme and rheme and their tonal properties were the
main focus of Özge (2003, Özge & Bozsahin 2010). Although speech melodies of particular
informational units were described under the title of phonology of Turkish intonation in Özge
(2003), a detailed phonological analysis was not the main focus of the study as was also stated by
the author. Instead, the focus was on the mapping between a particular informational structural
unit and its melody. The first study that directly attempts to provide an extensive analysis and
description of Turkish intonation and the prosodic structure above the word was proposed by
Kan (2009). However, the inventory of pitch accents and boundary tones proposed by Kan
(2009) are not distinctive as expected from a phonological model. For example, pitch accents
such as H* or L+H*, and phrase boundary tones such as L+H- or L+!H- are linguistically not
meaningful, nor have distinctive function. Kamalı (2011) provides a uniform analysis of tonal
contours and tonal inventory proposed in Kan (2009) and proposes a phonological model of
Turkish intonation. The major limitation of the intonational model proposed by Kamalı is that
the corpus of data used to propose a phonological model were composed of sentences that
contain single word prosodic phrases, as a result the model falls short in predicting the tonal
organization of multi-word prosodic phrases within a sentence. In short, although studies directly
focusing on Turkish intonation and prosodic structure began to emerge recently, there is still a
14
lack of a full-scale phonological model. This current study aims to fill this gap in Turkish
intonation, and aims to provide a comprehensive model of intonational phonology of Turkish.
The research questions addressed are:
(1) What is the tonal inventory (i.e., types of pitch accents and boundary tones)?
(2) Which prosodic units are tonally marked?
(3) How is sentence prominence tonally realized in declaratives, including focus
declaratives and questions.
These research questions will be addressed using the framework of Autosegmental-
metrical model of intonational phonology (Pierrehumbert 1980, Beckman & Pierrehumbert 1986,
Pierrehumbert & Beckman 1988, Ladd 1996/2008). The unique and theoretically relevant
findings will be discussed in order to better understand the typology of intonational systems
developed in the same framework, and the relationship between pitch accent and sentence
prominence.
The prosodic structure and the intonational phonology of Turkish proposed in this current
study is introduced in Chapter Two. The experiments carried out to support the current model’s
analysis of Turkish intonation and prosodic structure is described in Chapter Three. Finally,
Chapter Four summarizes the findings of the current study, and discusses the findings in the
context of prosodic typology on one hand and the relation between nuclear pitch accent and
nuclear stress on the other, and ends with concluding remarks.
1.2 Introduction to the Autosegmental-Metrical model of intonational phonology
The framework adopted in this current study and the recent studies on Turkish intonation
is the Autosegmental-metrical (AM) model of Intonational Phonology (Pierrehumbert 1980,
Beckman & Pierrehumbert 1986, Pierrehumbert & Beckman 1988, Ladd 1996/2008), which
analyzes a continuous pitch contour into a string of discrete pitch targets. The model identifies
two pitch targets: High (H) and Low (L). In addition, complex tones can be described as a
combination of the two: HL or LH. These targets are associated with designated parts of the
segmental string. Tonal events in between two tonal targets are derived by phonetic
15
interpolation. These transitions are not phonologically relevant as suggested by other approaches
(e.g. British Tradition (Crystal 1969)).
Two different kinds of tones are distinguished in the AM theory: (1) head-marking tones,
a.k.a pitch accents, and (2) edge-marking tones, a.k.a boundary tones. Head-marking tones are
associated with metrically prominent syllables while boundary tones mark the edge of prosodic
groups. A pitch contour, or tune or melody, is represented as a combination of these pitch
accents and boundary tones.
In the AM approach to intonation, tune is autonomous from the segmental string it gets to
coupled with, hence autosegmental. The alignment of pitch-accents and boundary tones with a
particular segmental string is derived post-lexically based on the metrical representation of the
segmental string, hence metrical. Therefore, metrical structure of a text lies at the heart of tune-
text association.
Metrical structure of a text is a syntagmatic representation of the relative prominence of
syllables, words, and phrases. Levels of prominence is represented on a metrical grid which
depicts the temporal structure of the tune (Liberman 1975, Liberman and Prince 1977). Pitch
accents can only be aligned with metrically strong, in other words lexically prominent syllables.
However, type of lexical prominence differs among languages. In general, languages are
categorized under one of the three categories with respect to word prosody based on the phonetic
realization of lexical prominence: These are tone languages, pitch-accent languages, and stress-
accent languages (Beckman 1986, Ladd 1996, Cruttenden 1997, Fox 2000). A language is
categorized as a tone language, e.g. Mandarin and Hausa, if each syllable or morpheme is
associated with a prespecified pitch contour at the lexical level. In a pitch-accent language, like
Japanese and Basque, certain, and not all, syllables have predetermined pitch accents at the
lexical level. Therefore, in a lexical pitch-accent language, the pitch-contour of particular
syllable is independent of intonation. A language is categorized to have stress-accent, e.g.
English or German, if a syllable in a word is more prominent than others by duration and/or
amplitude, and this syllable is eligible to get a pitch-accent at the post-lexical level depending on
intonation. Pitch-accents, whether determined by intonation post-lexically or at the lexical level
independent of intonation, are attracted to these lexically specified locations, and are labeled
with a “star-tone” (*), e.g. H* or L*. Complex tones include a trailing or leading tones in
addition to a starred tone, and is separated by the starred tone by a plus sign (+), e.g. L+H* or
16
H*+L. The leading tone determines the pitch of the syllable that precede the starred tone. So, for
an L+H* complex tone, H* is aligned with the lexically stressed syllable while L characterizes
the tone of the immediately preceding syllable. A trailing tone, on the other hand, determines the
pitch property of the syllable that follows the starred tone. In a H*+L tone, the lexically stressed
syllable would carry the starred tone, H*, while the following syllable would bear the L tone.
In addition to stressed or head syllables, tones are also associated with the edges of
prosodic phrases demarcating prosodic groups of different sizes at and above the word level.
There are two approaches on how to define prosodic units above the word. According to the first
approach which is known as Prosodic Phonology, prosodic constituents are defined based on the
syntactic structure of a sentence (Selkirk 1984, 1986; Nespor and Vogel 1986, Hayes 1989;
Hayes and Lahiri 1991, among others). In this approach, two prosodic phrases are distinguished
above the word level: (1) Intonation Phrase (IP) which roughly corresponds to a clause, and (2)
Phonological Phrase (PhP), which roughly corresponds to an XP (e.g., Noun Phrase or Verb
Phrase). The second approach, known as Intonational Phonology, defines prosodic constituents
based on the intonation of an utterance. (Pierrehumbert 1980; Beckman and Pierrehumber 1986,
Pierrehumbert and Beckman 1988; Jun 1993/1998/2005/2014, among others). Phrases above the
word, in this approach, include IP, Intermediate Phrase (ip), and Accentual Phrase (AP) which
roughly corresponds to a single content word and adjacent clitics and function words (Jun 1993,
1998). Whether syntactically or intonationally defined, the organization of prosodic units follow
the Strict Layer Hypothesis (Selkirk 1984, 1986; Nespor and Vogel 1986). That is, a prosodic
unit of one level is composed of one or more prosodic units of the immediately lower level, and
each prosodic unit is exhaustively contained in the superordinate unit. Not all prosodic units are
realized in every language, though, and not all prosodic units are tonally marked. Some tonally
mark all three, e.g. IP, ip, & AP in Georgian (Jun et al. 2007, Vicenik & Jun 2014), while others
mark one, e.g. IP in Portugese (Viana & Frota 2007) or two, e.g. IP & ip in English (Beckman &
Pierrehumbert 1986), or IP & AP in Persian (Tehrani 2007). In addition, a language might
tonally mark the edge of a prosodic word, as well, e.g. Serbo-Croatian (Godjevac 2000), West
Greenlandic (Arnhold 2014).
17
1.3 Phrasal prominence and prosodic typology
The advent of the AM model of intonational phonology paved the way for describing
prosodic systems of different languages within the same framework using the same prosodic
categories. This way, it has become possible to compare similarities and differences among
languages. By now, there are quite a few number of languages whose prosody and intonation
have been described using the AM model of intonational phonology. Recent work on prosodic
typology (Jun 2005b, 2014) classifies languages into three groups based on how prominence is
realized at phrasal level. The first group is called head-prominence languages. In such languages,
phrasal prominence is marked culminatively when the head of a prosodic unit such as a stressed
syllable receives a pitch-accent. The second group is called edge-prominence languages where
the prosodic phrasing of an utterance is manipulated to align the prominent word with the
right/left edge of some prosodic unit. Finally, the third group is called head/edge prominence
languages, and as is implied by its name, prominence in such languages is marked culminatively
as well as demarcatively, i.e., the prominent word hosts a pitch accent and occurs at the edge of a
higher prosodic unit.
1.3.1 Head prominence languages
A language is categorized as head-prominence under this typology if the phrase level
prominence is marked by the head of a phrase whether the head is lexically marked by pitch
(e.g., Swedish, Mandarin) or stress (e.g., English). Phrase head might be derived culminatively
from the head of a word, and determined at the phrasal level. Figure 1 displays the derivation of
a phrase head culminatively.
Figure 1: Metrical representation of the sentence ‘Marianna
made the marmalade.’
x sentence-level prominence
x x phrase-level prominence
x x x word-level prominence
x x x x x x x x x syllable nucleus
Marianna made the marmalade
T* T*
*
Pitch accents
18
A phrase-level prominence in a head-marking language can be marked by a post-lexical pitch
accent, T*. In such case, the post-lexical pitch accent can come either from a lexical stressed
syllable as in English or from a postlexical stressed syllable as in French.
1.3.2 Edge prominence languages
In an edge-prominence language prominence is marked by phrase edges (e.g. Seoul
Korean (Jun 1993), Mongolian (Karlsson 2014), West Greelandic (Arnhold 2014) ) . In such
languages, there is no lexically specified head, i.e., no stressed syllable, or a lexically specified
pitch accent, or tone. Likewise, there is no post-lexically marked head either, hence no post-
lexical pitch accent. In these languages, prominence of a word is marked by aligning the word by
the edge of a prosodic phrase, usually an Accentual Phrase (AP). A typical AP is defined as a
tonally marked prosodic unit that contains one content word including function words (Beckman
& Pierrehumbert 1986, Jun 1993). Korean, for instance, has neither lexical pitch accent nor
lexical stress (Jun 1993, 2005), i.e. no syllable in a word stands out as prominent in isolation.
However, Jun (2011) analyzed prosodic marking of complex-NP focus in Korean, and found that
when a complex-NP, such as a three-word NP, was focused, the last word in the NP was the
most prominent and it often started a new intermediate phrase, i.e. the most prominent word was
aligned with the left-edge of an intermediate phrase. Change in prosodic grouping due to
prominence in Korean is displayed in 4. Under broad focus, 4a, the complex NP forms one ip.
However, when the complex NP is narrowly focused, the most prominent word, NP3, starts a
new ip as in 4b, and is therefore aligned with the left-edge of ip.
(4) a. [NP1 NP2 NP3]
ip
[NP V]
ip
(Broad Focus)
b. [NP1 NP2]
ip
[NP3 NP V]
ip
(Narrow Focus)
In other words, what is accomplished by a pitch accent in head-prominence languages is
done by phrasing in edge-prominence languages (Jun 1993, Ladd 1996/2008).
19
1.3.3 Head/Edge prominence languages
In head/edge prominence languages, phrasal prominence is marked by both the head and the
edge of a phrase. That is, the head of a prominent word receives a pitch accent and the word is
aligned with the edge of a prosodic phrase. Similar to head-prominence languages, pitch-accent
might be from lexical-pitch accent or might be derived post-lexically. In this type of languages,
the location of lexical stress or pitch accent is often the edge of a word -- e.g., initial syllable of a
word (e.g. Bengali (Hayes & Lahiri 1991, Khan 2008); Georgian (Jun et. al 2007); Tamil (Keane
2007)), or the final syllable a word (French (Jun & Fougeron 2002); Kiche (Nielsen 2005)). And
the edge tone of a prosodic unit where a word is aligned with tends to be the opposite of the pitch
accent tone type (Jun 2014). For instance, in Bengali, a narrowly focused constituent is marked
by L* H
P
tonal sequence (Hayes & Lahiri 1991) where L* pitch accent marks the head of the
word that is aligned with the beginning of the focused phrase (corresponding to an ip) and a High
tone of the phrase (i.e., H
P
) marks its right edge. As the domain of the focused constituent
changes, right edge marking tone, H
P
, shifts accordingly demarcating the right edge of the
focused constituent. Shift in edge-tone, H
P
, in Bengali as a consequence of change in focus
domain is schematized in 5.
(5) a. […] [ WORD ]
FOCUS
[…]
b. […] [WORD WORD]
FOCUS
[…]
c. […] [WORD WORD WORD ]
FOCUS
[…]
One of the questions that the current study seeks to answer is where Turkish falls in the
prosodic typology of languages based on phrasal prominence: Is Turkish a head-prominence,
edge-prominence, or head/edge prominence language?
L*
H
P
L*
H
P
L*
H
P
20
1.4 Focus and prosodic prominence
The relation between focus (informational prominence) and prosodic prominence
(acoustic prominence) is well-established but the exact nature of this relation is still under
debate. In many languages marking of focus, which I define here as the part of a sentence (that is
under assertion) in a particular discourse context, is identifiable by particular prosodic cues,
especially tonal cues; pitch accents or boundary tones or both. In other words, focus is marked by
some kind of adjustments in the f0-contour
1
.
As mentioned in the previous section, in languages where phrasal prominence is marked
by a lexical or post-lexical pitch accent, the focused constituent of the sentence is associated with
the intonational nucleus of the sentence, a.k.a nuclear pitch accent (NPA). The following
example in 6 indicates use of nuclear pitch accent to identify focus. The presence of NPA is
indicated by small capitals.
(6) Susan bought a NEW car.
Figure 2. Pitch track example displaying pitch range expansion due to focus in
English for the sentence in 6 produced in neutral context (black pitch contour)
and when the adjective ‘new’ is narrowly focused (red pitch contour).
1
In some languages, primary means of marking focus is morphological (e.g. Navajo (Vallduvi and Engdahl (1996b))) or
syntactic (e.g. Hindi (Kidwai (2000))) and not prosodic.
Time (s)
3.438 4.938
Pitch (Hz)
70
400
21
The expansion in pitch range on the focused word, in particular rise in fundamental
frequency (f0), is what we hear as the most prominent in the sentence, i.e., nuclear pitch accent.
In 6, nuclear pitch accent is associated with the word ‘new’, and as can be seen in Figure 2, f0-
rise is aligned with the adjective ‘new’. In this example, focused constituent is the word bearing
the nuclear pitch accent. Cases as such where the nuclear pitch accent renders the word it is
aligned with as focused is referred to as narrow focus. The focus is narrow because the domain
of pitch accent is confined to the single word it is aligned with.
However, the relation between focus and pitch-accent is not always one-to-one. Consider
the example in 7.
(7) John bought a CAR.
The nuclear pitch accent in 7 is on ‘car’ as is demonstrated by small capitals, and this
sentence is compatible as an answer to any of the three questions in 8.
(8) a. What did John buy? [a car] NP focus
b. What did John do? [bought a car] VP focus
c. What happened? [John bought a car] wide-focus
The identification of focus in 8 can be done in context, and out of context the sentence is
ambiguous. This non-isomorphic relation between focus marker — pitch accent — and focus
domain is explained by focus projection in the literature. Put another way, constant location of
pitch accent in 7 is compatible with multiple focus domains such as NP in 8a, VP in 8b, and the
entire sentence in 8c. When the focus domain is the entire sentence, this type of focus is referred
to as ‘wide focus’, unlike narrow focus where the focus domain is restricted to a subconstituent
in a sentence. These terms used to refer to the size of focus domain were first introduced by Ladd
(1980). The reference domain in determining focus domain is usually a sentence. Therefore,
when the focus marker signals a focus domain smaller than a sentence, we are talking about
narrow focus, and when the focus domain is the sentence itself, we are talking about broad focus.
Nuclear accent can signal narrow focus only in certain positions. For example, if the
nuclear pitch accent is on new in 9, focus domain can only be narrow.
22
(9) Susan bought a NEW car.
On the other hand, a nuclear pitch accent on ‘car’ can assign four different focus domains
as displayed in 10.
(10) Susan bought a new CAR
The example in 10 displays that the relation between prosodic prominence and focus is
not isomorphic. If this relation was isomorphic, i.e. if the nuclear pitch accent could only render
the unit it is realized on as focused, then we could only account for one of the four possible
interpretations in 10 and that would be the case where car is focused because it is the word that
carries the nuclear pitch accent.
The concept of focus projection goes back, at least, to Chomsky (1971) and Jackendoff
(1972). By now it is widely agreed that the relation between prominence and focus cannot be
reduced to one-to-one mapping. However there are two distinct views on what plays the
fundamental role in the identification of focus. One view directly links focus identification with
pitch accent; known as pitch-accent first theories in the literature (Selkirk 1984, Gussenhoven
1995). According to this view focus identification is determined by the distribution of pitch
accents and the last pitch accented word, NPA, is perceived as most prominent, i.e. focused. The
second approach, known as stress-first theories (Hale & Vergnaud 1987, Ladd 1996, Zubizarreta
and Vergnaud 2005, Zubizarreta 2014), does not assume a direct connection between the two.
According to the stress-first theories, the location of Nuclear Stress (NS) of the sentence, i.e. the
rhythmically most prominent word, plays a fundamental role in the identification of focus. This
constraint is stated as the following: “the focused constituent must contain the rhythmically most
prominent word, i.e. the word that bears NS.” (Zubizarreta 2014). NS is then aligned with the
NPA at the intonational level. Under this view, NS is an abstract rhythmic notion which is
23
acoustically identified possibly in different ways across languages. In this present study, I will
argue that findings from Turkish support the latter view.
1.5 Previous work on Turkish prosody
1.5.1 Word stress
This section is on word level prominence in Turkish. Typologically, Turkish is grouped
under the stress-accent languages. Regular word stress falls on the final syllable in Turkish, and
this pattern is insensitive to any rhythmic considerations (Lees 1961, Lewis 1967, Sezer 1981
among others).
(6) ki.táp
2
‘book’
mah.ke.mé ‘court’
ad.li.yé ‘courthouse’
nos.tal.jí ‘nostalgia’
nok.ta.la.má ‘punctuation’
yol.suz.lúk ‘corruption’
A large class of suffixes do not interfere with the regular stress pattern, i.e., word stress
shifts to the rightmost syllable as more suffixes are added.
(7) mahkeme-ler-imiz-dén ‘from our courts’
court.plu.poss1pl.abl
Although final lexical stress is the pervasive word stress pattern and majority of the
words are stressed on their final syllables, Turkish also has number of instances of non-final,
irregular stress. Non-final stress might be due to exceptional root stress, lexically stressed
suffixes or pre-stressing suffixes.
2
Dots indicate syllable boundary.
24
Exceptional root stress is observed in certain place names, unfamiliar personal names,
and certain borrowed words. Stress on such words is quantity-sensitive: it falls on the
antepenultimate syllable if the ante-penult is heavy and the penult is light, and falls on the
penultimate syllable otherwise (Sezer 1981).
(8) Án.ka.ra vs. Al.mán.ya
Although the weight sensitive rule as described by Sezer correctly predicts word stress in
majority of the exceptionally stressed roots, there are still exceptions to this rule, e.g. Bel.ÇÍ.ka
(“Belgium”). There are different approaches to be able to capture exceptions to exceptional word
stress (Inkelas & Orgun (1998), Kabak & Vogel (2001)), Details of what mechanism better
captures the exceptional root stress phenomena in Turkish is not the focus of this present study.
The reader may refer to the works cited for the details of different proposals, and to Inkelas &
Orgun (2003) for a detailed comparison of the two approaches.
Further suffixation does not interfere with exceptional root stress, i.e., lexical stress does
not shift to the final syllable as suffixes are added to an exceptionally stressed word. Compare 9a
with 9b.
(9) a. Ánkara-lı-lar-dan
b. kitap-lık-lar-dán
In addition to roots, certain suffixes are also lexically marked for irregular stress (Inkelas
1999, Kabak and Vogel 2001). Suffıxes, -Íyor and -Íver, are two examples of lexically stressed
suffixes. When two lexically marked suffixes co-occur, word stress falls on the first lexically
specified suffix (Van der Hulst & van de Weijer 1991).
(10) a. konuş-úyor
b. konuş-úver.iyor
25
Yet another source of non-final stress is pre-stressing suffixes. These suffixes cause the
word stress fall on the preceding stressable syllable. Negative morpheme, -mA, is one of the
prestressing suffixes.
(11) a. konuş-úyor
b. konúş-mu-yor
In short, regular stress pattern assigns word stress to word final syllables, yet exceptions
to this pattern occur; non-final stress might come from exceptional root stress, from lexically
stressed syllables, or pre-stressing syllables.
Even though the main body of literature on Turkish word prosody groups Turkish as a
stress-accent language and uses the term ‘stress’ to refer to word-level prominence (e.g. Lees
1961; Sezer 1983; Kaisse 1985; Barker 1989; Inkelas 1999), Levi (2005) proposes that Turkish
is a pitch-accent language like Basque or Japanese. Although the purpose of this current study is
not to discuss the place of Turkish in the typology of word-level prominence, I would like to
open a new sub-section and comment on Levi’s proposal at this point because one of the
intonational models of Turkish, Kamalı (2011), I will discuss later is built upon this proposal of
Levi’s.
Turkish as a pitch-accent language? Levi 2005 is on the acoustic correlates of lexical
accent
3
in Turkish. This phonetic study compares three suprasegmental features, duration,
amplitude, and f0, of finally and non-finally stressed words using near minimal pairs. The results
showed significant differences for duration, amplitude, and f0 between finally and non-finally
stressed words; especially f0 peaks were found to be dramatically different between finally and
non-finally stressed words. As a next step, the author ran a discriminant analysis (DA) to
determine which of these acoustic cues was most relevant as a perceptual cue. The results from
DA analysis predicted f0 to be the most important acoustic cue. Based on these results, which the
author equated with the results for Japanese from Beckman (1986), Levi proposed that Turkish
should be categorized as a pitch-accent language.
Before discussing on Levi’s interpretation of the experiment results, I would first like to
comment on the method used to measure the acoustic correlates of word level prominence in the
3
Levi (2005) avoids using the term ‘stress’ to refer to word level prominence.
26
study and earlier phonetic studies conducted with the same purpose, i.e. to determine the acoustic
correlates of word stress.
The seminal work on the acoustic correlates of stress in English conducted by Fry (1955,
1958) paved the way for similar studies done for different languages later. Fry (1955) identified
increased duration and intensity, and changes in f0 as main correlates of word stress. Fry (1958)
found that of these features associated with stress, higher f0 was the most reliable cue to
perceived prominence at the word level. Since Fry’s study, similar acoustic studies for many
languages have been conducted, e.g. Polish (Jassem et al. 1968), Tagalog (Gonzalez 1970),
Chickasaw (Gordon 2004), and Turkish (Levi 2005), which showed some combination of
increased duration, greater intensity, and/or higher f0 as main correlates of word stress.
However one major limitation of Fry’s study, and many following it including Levi
(2005), was that the target words occured in focal positions in utterances, a position that hosts
both word-level stress and phrasal-level prominence which are shown to be realized by different
phonetic means. For example, Sluijter and van Heuven (1996) in their study that aims to tease
apart acoustic cues to word stress and phrasal prominence showed that in Dutch most reliable
acoustic correlates of word stress were increased duration and high frequency emphasis, and f0
movement and intensity differed significantly for accented and non-accented syllables rather
than between stressed and non-stressed syllables. One very important implication of this finding
is that f0 characteristics found to be contributing to perceived prominence is due to syllables
carrying phrasal prominence marked by a pitch accent rather than word stress. As is well known
by now, speakers use pitch accents to convey semantic or pragmatic information such as focus.
In most recent works, pitch changes are shown to be a major cue to phrasal prominence and not
much to word stress. For instance, Sluijter and van Heuven (1996a) show that in English word
level stress is cued by spectral tilt and vowel reduction, while pitch features become relevant at
phrasal level. Similarly, Gordon (2004) found duration and intensity as more robust markers of
word level stress in Chickasaw, and f0 of phrasal prominence.
With these findings in mind, let’s now look at Levi’s (2005) experiment materials used to
determine acoustic correlates of word stress. In her study, Levi created two groups of finally and
non-finally stressed words, and crucially embedded each target word in the carrier phrase 12, and
participants produced the carrier phrase that contains the target word.
27
(12) Ahmet ‘x’ dedi.
Ahmet X said
‘Ahmet said X.’
As can be seen in 12, the target word immediately precedes the verb, which is known to
be the position for sentence stress in Turkish (Göksel & Özsoy 2000), i.e., in general the
informationally most prominent word in Turkish immediately precedes the verb. Therefore, the
acoustic measurements taken from the stressed syllable of target words reflect phrasal
prominence, sentence stress in this case, rather than word stress, hence the significant f0 change
the author found is due to post-lexical pitch-accent to mark phrasal prominence. I strongly
believe that the covariation of stress and accent in Levi’s work is the major source that gave rise
f0 to stand out as the most reliable cue, and duration and intensity much less so, which
eventually led Levi to suggest Turkish to be a pitch-accent language, a proposal that is also
criticized by Ladd: “[...] the traditional interpretation of Fry’s 1958 findings still informs
published work (e.g. Levi 2005 on Turkish, which reaches the rather misleading conclusion that
Turkish is a ‘pitch accent language’)” (Ladd 2008:58).
1.5.2 Phrase Stress
Kabak and Vogel (2001) proposes that phrase stress in Turkish falls on the leftmost unit
of a phonological phrase (PPh), and the leftmost word within a PPh is marked by a pitch accent
at the intonational level (Kan 2009). A schematic representation of this association is given in
Figure 3 (from Kan 2009).
28
Figure 3. Pitch accent placement in Turkish according to Kan (2009).
As can be seen in Figure 3, in a PPh that contains more than one prosodic word (PWd),
3a and 3c, each PWd carries word stress and the stress of the first PWd is promoted to PPh level
and a pitch accent (PA) is docked onto the stressed syllable of the leftmost word in the PPh.
When a PPh is composed of one PWd, 3b, since it is the only word within the PPh, it is also the
head of the phrase. Therefore, its stressed syllable is promoted to the higher PPh level and is
associated with a PA.
The relation between pitch accents and phonological phrases in Kan (2009) is displayed
in Figure 4. The sentence exemplified in Figure 3 is composed of three PPhs. The first PPh
(PPh1), alanyalıların ağırladığı, and the third PPh (PPh3), okulları gezdi, each contain two
prosodic words, and in each of them only the leftmost prosodic word hosts a pitch accent (PA).
The second PPh (PPh2), bayülgen, contains only one prosodic word, therefore it is also the head
of the PPh and it is stressed syllable is marked by a pitch accent (PA).
29
Figure 4. Sample pitch track from Kan (2009). Pitch contour is divided
into three phonological phrases (PPh) and each of them is hosting one
pitch accent.
According to Kan (2009), PPh is the domain of pitch accent assignment and therefore
each PPh hosts only one pitch accented prosodic word.
1.5.3 Sentence stress
Kan (2009) is the first study to provide an extensive analysis of Turkish prosody at
sentence level. The details of her analysis and the phonological model she proposes, I will
summarize in Section 1.6.1. In the current subsection, I will review her analysis of sentence
prosody with respect to pitch accent distribution and eventually the projection of sentence stress
culminatively.
As mentioned earlier, domain of pitch accent distribution in Kan(2009) is Phonological
Phrases (PPh). A PPh is left-headed (Kabak & Vogel 2001), and the head of a PPh requires an
intonational pitch accent (Kan 2009). A second property of PPh in Kan is that the right of a PPh
is tonally marked, distinguished from pitch accents by a “-“ diacritic in the tonal tier. The sample
pitch track in Figure 5, provides prosodic phrasing of a four-word sentence. The two prosodic
PPh1 PPh2
PPh3
PA
PA
PA
30
phrases, PPhs, are demarcated by L+!H- and L- boundary tones, respectively, and the first word
in each PPh hosts a pitch accent, H*.
Figure 5. Sample pitch track of a three-word sentence from Kan (2009:89)
As can be seen in Figure 5, the head of each PPh carries a pitch accent. Kan (2009)
ascribes sentence stress to the strongest syllable of the rightmost PPh, i.e., the lexically stressed
syllable of the head of the rightmost PPh projects to the IP (Intonation Phrase) level, creating
sentence stress. This bottom-up derivation is exemplified in Figure 6.
Figure 6. Derivation of sentence stress culminatively.
From Kan (2009:97).
PPh1
PA
PPh2
PA
31
In short, the head of a phonological phrase, which is the leftmost word within a PPh,
receives a pitch accent and the head of the rightmost phonological phrase is promoted to
sentence stress, i.e. the stressed syllable of the prosodic word that is aligned with the left edge of
the rightmost prosodic word carries sentence stress. Then the next question is how to determine
the PPh boundaries. This is not a topic addressed by Kan (2009). One such attempt was made by
Üntak-Tarhan (2006) where the location of sentence stress is derived syntactically. Below, I will
review this proposal.
Üntak-Tarhan (2006) A Syntactic Approach to Sentence Stress in Turkish: The theoretical
framework behind the analysis of Üntak-Tarhan (2006) takes its roots from Chomsky’s (2000,
2001) recent development of Minimalism which is known as the theory of phases and multiple
spell-out. Chomsky suggests that CPs and transitive vPs constitute phases, while TP and
unaccusative vP do not. Üntak-Tarhan (2006) defines what she calls Stress Domains (SD) and
the highest element in the SD is assigned sentence stress. An SD is determined by phases such
that the complement of a phase creates a SD. This is schematized in Figure 7.
Figure 7. Schematic representation of stress domains determined based on
phases, CP and vP. TP and AspP being complements of CP and vP, respectively
creates SDs.
This mechanism, the author suggests, accounts for the unusual stress behaviour of
certain structures such as structures containing manner adverbials, intransitives, as well as the
CP
C
SD
vP
v
SD
VP
Asp
TP
AspP
32
transitive sentences with bare objects. In what follows, I will briefly describe how the
mechanism predicts sentence stress.
A group of adverbs, which Erguvanlı-Taylan (1984) classified as ‘non-derived adverbs’,
have unique morphological and syntactic properties. Morphologically, they have the form of an
adjective, therefore in isolation they are ambiguous between being an adjective or an adverb.
Syntactically, they behave similar to regular arguments and their position is fixed unlike other
adverbs, and prosodically they require sentence stress (Kamalı 2008). A sample sentence with a
non-derived adverb is provided in 13. The bracketing indicates syntactic as well as prosodic
grouping where the non-derived adverb, hızlı (“fast”), carries sentence stress.
(13) Ali (hızlí kitap okur).
Ali fast book read-AOR
“Ali reads book(s) fast.”
Syntactically, non-derived adverbs behave like arguments such that they cannot be
deleted (e.g., ‘bad’ in 14a) or cannot be stranded (e.g., ‘well’ in 14b).
(14) (a) Ali Ahmet-e *(kötü) davran-dı.
Ali Ahmet-DAT bad behave-PAST
“Ali behaved Ahmet *(badly).”
Üntak-Tarhan (2006:71)
(b) *[Elbise dik-er]
i
Ayşe güzel t
i
dress sew-AOR Ayşe well
“Ayşe sews dresses well.”
Üntak-Tarhan (2006:63)
Another property of ‘non-derived adverbs’ is that they cannot precede an accusative case
marked with an adverbial reading. This is exemplified in 15.
33
(15) *Ayşe güzel elbise-yi dik-ti.
Ayşe well dress-ACC sew-PAST
Intended meaning: “Ayşe sewed the dress well.”
Üntak-Tarhan (2006:65)
Such syntactic restrictions do not hold for what Üntak-Tarhan (2006) refers to as
Complex Manner Adverbs (CMA) as is exemplified in 16.
(16) Ayşe güzelce elbise-yi dik-ti.
Ayşe nicely dress-ACC sew-PAST
“Ayşe nicely sewed the dress.”
Üntak-Tarhan (2006:65)
What is more CMAs do not carry sentence stress unless they immediately precede the
verb. Following these observations, Üntak-Tarhan (2006) proposes that non-derived adverbs,
which she refers to as simple manner adverbs (SMA) like hızlı (“fast”) or güzel (“beautiful”) are
in the specifier position of VPs, and therefore they receive sentence stress because they are the
highest element in the stress domain, or the spell-out. On the other hand, CMAs attached to vPs,
hence are outside the SD and do not carry sentence stress.
With the assumptions on what constitutes a phase and what not, Üntak-Tarhan (2006)
proposes an account on how to derive two separate stress patterns in intransitive sentences. In
Üntak-Tarhan’s classification, sentence stress falls on the verb when the verb is unergative.
Since verb itself is the only element in the SD created by AspP, the verb receives sentence stress.
When the verb is unaccusative, the sentence stress falls on the subject
4
. Recall that unaccusative
vP are assumed not to constitute a phase, therefore AspP does not create a SD. As a result the
only SD in the sentence is TP and since subject is the highest element under TP, the SD, it
receives sentence stress.
4
Although unaccusative/unergative is the distinction Üntak-Tarhan (2006) uses to predict the location of sentence stress, Kan
(2009) showed this distinction does not account for the variability observed on where sentence stress falls in intransitive
sentences, and yet does not bring an explanation on what causes shift in sentence in intransitives. In Chapter 3, I will show that
the two stress pattern on such sentences depends on the eventive/generic nature of the verb.
34
Sentence stress on the accusative objects is derived in a different fashion. Separate from
bare objects, which are base generated within VP, Üntak-Tarhan (2006) assumes that accusative
objects are generated as specifiers of AspPs, therefore get sentence stress by being the highest
unit in the first spell-out domain which is created by AspP.
To summarize, sentence stress mostly falls on the immediately preverbal unit but that’s
not always the case. When, for instance, a SMA precedes a bare object, the sentence stress
obligatorily falls on the SMA rather than on the immediately preverbal bare object. Another type
of structures where sentence stress falls on a unit other than the immediately preverbal unit is
intransitives. In such structures, the sentence stress can be realized on the subject or on the verb.
Interim Summary of Turkish Prosody: In this subsection, I reviewed the relevant work on
Turkish prosody at word, phrase and sentence levels. Turkish is considered as a stress-accent
language where the regular stress pattern assigns lexical stress to word final syllables. So-called
irregular stress pattern is also present in Turkish. This might be due to exceptional root stress, or
might result from lexically stressed or pre-stressing syllables. In addition, I also presented the
proposal by Levi (2005) where Turkish is categorized under pitch-accent languages similar to
Japanese or Basque, and discussed the problems with this proposal.
When we move to prosodic studies above the word level, we see that there is limited
work at this level. Kabak & Vogel (2001) is one study that looks at the phrase stress
phenomenon in Turkish. They suggested that main phrasal stress falls on the PPh initial word
whose stressed syllable is associated with a pitch accent in Kan (2009). Kan (2009) is the first
study that is directly concerned with sentence level prosody. Kan proposed that in a sentence,
main stress falls on the left-most word of the right-most PPh. Therefore, where sentence stress
falls depends on where the boundaries of PPh lie, a point that Kan (2009) does not attempt to
address. In that context, I presented one such attempt by Üntak-Tarhan (2006), which is a
syntactic approach to predict sentence stress based on the notion of phases by Chomsky (2000).
In the next section, I will present in detail the two previous phonological intonational
models developed in the AM framework.
35
1. 6 Previous intonational models of Turkish
There are two previous intonational models that propose Intonational Phonological model
of Turkish prosody: Kan (2009) and Kamalı (2011). In this section, I will provide a review of
these two models.
1.6.1 Kan (2009)
Kan (2009) is the first study to provide an extensive account of Turkish intonation and
prosody using the AM framework of intonational phonology. The model is developed based on a
large corpus of data extracted from spoken dialogues acted out by nine native speakers. The
target sentences were produced in focus-neutral contexts. In the light of the data, an inventory of
pitch accents and boundary tones were proposed.
As mentioned earlier, Kan (2009) proposes that the prosodic level governing pitch-accent
distribution is the Phonological Phrase (PPh) in Turkish. The pitch accent placement rule states
that the head of a Phonological Phrase requires an intonational pitch accent. The leftmost unit in
the PPh is identified as the head following Kabak & Vogel (2001).
The second proposal is that the end of a PPh is tonally marked. The type of the tone at the
right-edge of a PPh depends on the lexical properties of the PPh-final word as described in 17.
(17) a. When a syllable that bears phrase stress (+ a corresponding pitch accent) is non-
PPh-final, a non-accentual pitch event controls the melody between the pitch
accent and the end of the phrase, which we refer to as a phrase accent adopting
the terminology of the AM model.
b. When a phrase-stress-bearing syllable is located at the end of a PPh, a
corresponding pitch accent marks the end of the PPh, where a phrase accent is not
instantiated.
36
Now, let us look at the intonational model developed by Kan (2009) based on these
assumptions. I will first present the pitch accent inventory and then move onto boundary tones.
The complete set of pitch accents proposed by Kan involves L+H*, L+!H*, H*, and !H*. Of
these pitch accents, the bitonal ones, L+H* and L+!H*, are used to mark a tonal event that
involves f0 rise while the monotonal ones, H* and !H*, are used to mark a tonal event that is
rather flat. Both type of pitch accents, i.e., rising and flat, are displayed in Figure 8.
Figure 8. Tonal analysis of ‘Alanyalılar maymunu arıyomuş’ (People of
Alanya have been looking for the monkey). From Kan (2009:84).
Alanyalı-PLU monkey-ACC look for-PROG-EVıD
“People of Alanya have been looking for the monkey.”
In Figure 8, the prenuclear pitch accent displays a rising nature and is marked by bitonal
pitch accent, L+H*, while the tonal event on the nuclear word is rather flat and is marked by a
monotonal, !H*, pitch accent.
Crucially, all the pitch accents proposed by Kan are in free variation, i.e. they are
linguistically not meaningful nor have any distinctive function as expected from a phonological
model. Those various pitch accents rather reflect the phonetic shape of the pitch contour instead
of depicting the underlying phonological pitch accent/s. This can be observed in the tonal
analysis of the pitch contour given in Figure 9 and Figure 10.
37
Figure 9. Tonal analysis of ’Ayla Yalova’yı zor bulmuş’ (Ayla had difficulty in
finding Yalova). From Kan (2009:161)
‘Ayla had difficulty in finding Alanya’
There are three pitch accented words in Figure 9, realized on the first three words of the
four-word sentence. The first pitch accent realized on the finally stressed word, ayla, is marked
by a L+H* because it is rising. On the other hand, the second and third pitch accents are marked
by H* pitch accent because the pitch contour on the words they are realized is not rising. The
question at this point is whether these two different pitch accents should be included in the pitch
accent inventory. When we look at the details of the sentence in Figure 9, we can see that the
first H* pitch accent realized on the non-finally stressed word, yalovayı, is only one syllable
away from the preceding high tone which is part of a bitonal pitch accent, L+H*. The second H*
pitch accent (nuclear pitch accent) which is realized on the monosyllabic word, zor, immediately
follows the preceding high tone, L+H-; i.e. there are no syllables in between two high tones.
Therefore, there is not enough time between two high tones in neither of the cases for a potential
low pitch target to be realized. With this in mind, now let’s compare this tonal analysis with
another pitch track given in Figure 10.
38
Figure 10. Tonal analysis of ‘Alanyalılar yaralı maymunu arıyomuş’ (The
people of Alanya have been looking for the monkey, which is injured). From
Kan (2009:178).
Pitch track in Figure 10 is taken from Kan (2009: 178). Similar to pitch contour in Figure
9, first three words in this sentence host a pitch accent. Different from the sentence in Figure 9,
here high tone targets are separated from each other by at least two syllables, i.e. there is enough
time for any underlying low tone to be realized between the two high pitch targets. As can be
seen in Figure 10, this time both of the prenuclear pitch accents are rising and are marked by
L+H* pitch. The nuclear pitch accent, on the other hand, does not display any rising contour, but
is rather realized with a flat contour maintained throughout the finally stressed nuclear word, but
crucially the level of this tone is lower than the previous high tone.
First let’s compare prenuclear pitch accents in Figure 9 and Figure 10. What these two
pitch tracks show is that when there is enough phonetic space, prenuclear pitch accents are
realized with a rising pitch accent. Therefore, it is highly likely that the lack of rising for one of
the prenuclear pitch accents in Figure 9, the second pitch accent which is marked by monotonal
pitch accent, H*, is due to undershooting. This possibility is mentioned by Kan as well.
However, both L+H* and H* are included as possible prenuclear pitch accents in the tonal
inventory.
39
Now let us look at the nuclear pitch accents in Figure 9 and Figure 10. In both of the
figures, the nuclear pitch accent is realized with a flat tone rather than a rising one. The
difference between the two nuclear pitch accents is that in Figure 9, it sustains the same height as
the previous high tone while in Figure 10, f0 following the last prenuclear pitch accent drops to a
lower level and the level tone is sustained until the end of the stressed syllable of the nuclear
word. If we look at the details of the two sentences, we can see that in Figure 9, the nuclear pitch
accented word is one syllable and more importantly the final syllable of the word that
immediately precedes the nuclear word carries a bitonal boundary tone, LH-. That means, there
is no syllable in between two high targets, i.e., prenuclear LH- boundary tone and nuclear pitch
accent H*, for a possible low target to be realized. As can be seen in Figure 8, when there is
enough space, the nuclear pitch accent is still realized with a flat tone but at a lower level than
the previous high tone. Therefore, the distinction between H* and !H* in Kan (2009) as nuclear
pitch accents is also purely phonetic rather than phonological.
Now, let us look at the boundary tone inventory in Kan (2009). The next prosodic level in
Kan’s model is the Phonological Phrase (PPh). Kan suggested that the right edge of a PPh is
tonally marked. The PPh final tones proposed are L+H-, L+!H-, H-, and L-. Similar to pitch
accents, the first three boundary tones are in free variation. Crucially, the type of the right edge
marking tone depends on the lexical properties of the PPh final word, i.e. if the last word in the
PPh is finally stressed, a pitch accent marks the right edge of a PPh . On the other hand, if the
last word in a PPh is non-finally stressed, a boundary tone demarcates the right edge of a PPh.
Tonal analysis of the sentence in Figure 11 display two different treatment of the PPh edge
depending on the lexical properties of the PPh final words. The first word in the example,
alanyalılar, is non-finally stressed and forms one PPh, and its right edge is marked by a
boundary tone L+H-. The second word, yaralı, in the same sentence is finally stressed.
According to Kan’s analysis, it also forms one PPh, but this time the right edge of the PPh is
marked by a pitch accent L+H* which also marks lexical prominence of the stressed syllable.
40
Figure 11. Comparing PPh boundary tones on finally and non-finally
stressed words. From Kan (2009:178)
As a consequence, the tonal event realized on the stressed syllable of a finally stressed
word is always marked as a pitch accent no matter whether it is phrase medial or phrase final.
This creates a mismatch between the prosodic structure of a sentence and its intonational
structure such that while the prosodic structure indicates a bigger prosodic grouping, the
intonational analysis of the same sentence will not reflect this prosodic grouping if the right edge
of a PPh coincides with a finally stressed word. This mismatch can be clearly seen the tonal
analysis of the sentence in Figure 11. The second word in that sentence, yaralı, is marked by a
L+H* pitch accent on its finally stressed syllable although the prosodic structure of the sentence
suggests that the word itself is a PPh, i.e., a bigger prosodic grouping than the pitch accent
implies. Therefore, according to the model proposed by Kan (2009), tonal analysis of a sentence
does not correctly reflect its prosodic structure when the PPh edges coincide with a finally
stressed word. Considering the fact that majority of the words in Turkish are lexically stressed on
their final syllables, it becomes even more important to develop an intonational model that
distinguishes the function of a tonal event on a phrase medial and phrase final words that are
lexically stressed on their final syllables.
41
To recap Kan’s (2009) model, it is the first study that proposes an extensive analysis of
Turkish prosodic structure and its intonation using the framework of AM model of intonational
phonology. The model is developed based on an analysis of large corpus of data. Type of
sentences analyzed are declarative sentences produced in the neutral focus context. Two prosodic
structures are distinguished above the word: (1) Phonological Phrase (PPh), (2) Intonation
Phrase(IP). Each of these higher prosodic units are tonally marked at their right edges. However,
tonal marking of a PPh is not without problems, especially when the last word in a PPh is finally
stressed. This is because Kan (2009) assumes that the right edge of a PPh is marked by a pitch
accent when the phrase final word is lexically stressed on its final syllable. As I have shown in
Figure 9, this creates a mismatch between the tonal structure of a sentence and its prosodic
structure.
Another problematic point in Kan’s model is the status of different pitch accents and
boundary tones. Kan provides a rather rich inventory of pitch accents and boundary tones. These
different pitch accents and boundary tones are linguistically not meaningful and their alternation
do not have any semantic effect. Therefore the tonal categories proposed are not distinctive as
expected from a phonological model.
The next intonational model developed by Kamalı (2011) addresses both of these issues;
i.e. multiplicity of the tones in free variation and the status of the tone on finally stressed words,
that emerge in Kan’s model. However, Kamalı’s model has other weaknesses. In the next
section, I will review the intonational model proposed by Kamalı (2011).
1. 6. 2 Kamalı (2011)
Kamalı (2011) provides a unified account of different pitch accents and edge tones
proposed by Kan (2009) and builds a phonological model of Turkish intonation. However, this
model is built upon the assumptions that Turkish is a lexical pitch-accent language, a proposal
first made in Levi (2005). Following this assumption, the words with non-final stress are claimed
to have a lexical pitch accent, H*L, as in Tokyo Japanese (Beckman & Pierrehumbert 1986,
Pierrehumbert & Beckman 1988), and the words with a final stress are claimed to have no lexical
specification. Now, let us look at the tonal analysis proposed by Kamalı developed based on
these assumptions.
42
The data used to develop the intonational model of Kamalı is collected from four Turkish
native speakers who produced four-word sentences under neutral focus condition. The elicitation
was done in question-answer pairs.
In this model, a non-finally stressed word is marked by H*L lexical pitch accent. This is
true for both prenuclear words as well as the nuclear word. Pitch tracks in Figure 12 and Figure
13 display tonal marking of a prenuclear non-finally stressed word and a nuclear non-finally
stressed word, respectively.
Figure 12. Tonal analysis of ‘numaraları limonluya yönlendirmeli’ (He
should forward the numbers to the one with lemons). The prenuclear
word, ‘numaraları’ is non-finally stressed and carries H*L lexical pitch
accent. From Kamalı (2011:70).
Figure 13. Tonal analysis of ‘bunalanları limonluya yönlendirmeli’ (He
should direct those who get overheated to Limonlu). The nuclear word,
limonluya, is non-finally stressed and carries H*L lexical pitch accent.
From Kamalı (2009:73).
43
Tonal event on finally stressed words is treated differently depending on whether the
finally stressed word is prenuclear or nuclear. A note on the stimulus design is necessary at this
point. Each prenuclear word in the stimuli forms a Phonological Phrase (PPh)/Major Phrase
(MaP)/Accentual Phrase (AP), therefore the stressed syllable of a prenuclear finally stressed
word is always at the right edge of a PPh/MaP/AP. Therefore, the tonal event on a prenuclear
finally stressed word is marked as a boundary tone, H-. This is exemplified in Figure 14.
Figure 14. Analysis of tonal event on finally-stressed prenuclear words.
From Kamalı (2011:68).
Now, let us look at the analysis of tonal event on a finally stressed nuclear word in this
model. Kamalı observes that the nuclear word when finally stressed displays a miniscule rise or
no rise at all, and due to the lack of any substantial rise, she does not associate any pitch accent
with a finally stressed nuclear word but rather analyzes this tonal event as rightward L- tone
spreading, a boundary tone associated with the left edge of all constituents except for the
utterance initial position. In short, according to Kamalı’s analysis, a prenuclear finally stressed
word hosts a H- boundary tone marking the right edge of a PPh/MaP/AP while the finally
stressed syllable of the nuclear word is not associated with any tone. This distinction between a
prenuclear and nuclear finally stressed words is displayed in Figure 15.
H-
H-
44
Figure 15. Analysis of tonal event on finally stressed nuclear word.From
Kamalı 2011: 84).
The intonational model by Kamalı is quite neat compared to the one proposed by Kan
(2009) in the sense that it provides a uniform analysis of pitch accents and boundary tones: the
number of pitch accents is reduced to one, H*L, and the right edge of a PPh/MaP/AP is
systematically marked by a H- boundary tone. However, the model has one major limitation: it is
completely silent about the status of the phrase medial tones realized on finally stressed words.
Recall that in Kamalı’s model, non-finally stressed words carry H*L lexical pitch accent while
finally stressed words have no lexical specification. A high tone on a prenuclear finally stressed
word is analyzed as a boundary tone, H- and since no pitch accent is associated with finally
stressed words, lexical or post-lexical, we are left with no proposal on how to analyze the
possible tonal event on a phrase medial finally stressed word. (As shown in Chapter 2, phrase
medial finally stressed words do show a High tone.)
To summarize, while Kamalı’s model improves Kan’s intonational model in certain
aspects, especially by providing a more unified analysis of pitch accents and boundary tones, it is
incomplete in the sense that it falls short in providing a tonal analysis for multi-word phrases,
especially when these phrases are composed of finally stressed words. The bottomline here is
that both of the intonational models run into the problem of coupling prosodic structure with the
hierarchical organization of the intonational contour in the presence of finally stressed words:
While in Kan’s model the problem is a mismatch between the prosodic structure and the tonal
tier, Kamalı’s model has no saying on the status of phrase medial high tones on finally stressed
words. Considering the fact that majority of the words in Turkish is finally stressed, it becomes
even more important to have an intonational model that correctly predicts the hierarchical
organization of the intonational contour of a sentence in the presence of finally stressed words as
No T*
45
well as non-finally stressed words. Therefore, in this current study, I aim to provide an
intonational model where the prosodic structure of a sentence is correctly reflected in the
intonational organization of the sentence, and likewise tonal analysis of the sentence correctly
predicts its prosodic structure regardless of the type of the words, i.e., whether they are finally or
non-finally stressed, or regardless of the phrasal status of the words, i.e., whether they are phrase
medial or phrase final.
Another issue with these two intonational models is the analysis of tonal event on the
nuclear word. In both of the previous intonational models, the tonal event on the nuclear word
was treated somehow differently from those of prenuclear words. The common point that is
exemplified in both of the models is that the tonal event on the nuclear word might be realized
with a flat contour displaying no rise at all which is different from prenuclear pitch accents
which has a rising tonal nature. In Kan, lack of rising on the nuclear word was analyzed using
two separate nuclear pitch accents, H* and !H*, while the rising nature of the nuclear word is
captured by bitonal L+H* or L+!H* pitch accents. As mentioned earlier, these tones are in free
variation and therefore it is not clear what the true underlying nature of nuclear pitch accent in
Turkish is. Kamalı’s approach is completely different from Kan’s. Kamalı suggests that when the
nuclear word is non-finally stressed, it displays a rise-fall tonal event which she marks by the
lexical H*L pitch accent. On the other hand, when the nuclear word is finally stressed, it displays
either a miniscule rise or no rise at all. Following this, no pitch accent is associated with the
nuclear word when it is finally stressed. Therefore, according to Kamalı’s proposal, the tonal
event on a nuclear word depends on the type of the word, i.e., whether it is finally or non-finally
stressed. However, the distinctive property of the nuclear pitch accent does not depend on the
lexical properties of the nuclear word as suggested by Kamalı. For example, Kan (2009) displays
a pitch contour where the non-finally stressed nuclear word is realized with no f0 rise unlike
what Kamalı claims. This is exemplified in Figure 16. In this example, the second word,
almanyayı, is the nuclear word which is lexically stressed on the non-final syllable, man. As can
be seen from the figure, the pitch contour displays no rise on the nuclear word itself and is rather
flat. Kamalı treats this example from Kan (2009) as an exception. However, I will also show that
whether the tonal event on a nuclear word displays any rise or not does not necessarily depend
on whether it is finally or non-finally stressed. The bottomline here is that an analysis of the
nuclear pitch accent based on the type of the word does not correctly predict the tonal nature of
46
the nuclear word. In the intonational model I propose, I aim to improve the analysis of nuclear
pitch accent in a way that it correctly describes the true nature of it.
Figure 16 .Tonal realization of nuclear pitch accent on a non-finally
stressed word. From Kan (2009:84).
L+H* !H* L-
[Ayla]
PPh
[Almanyayı özlemiş]
PPh
‘Ayla missed Germany’
Another limitation of the previous two models is the lack of tonal analysis of different
sentence types. The pitch accents and boundary tones proposed in both studies are based on
declarative sentences produced in neutral contexts. Therefore, the tonal inventory provided in
both models is limited to such structures only. In this present study, I also aim to provide tonal
analysis of different sentence types, aiming to provide a more complete inventory of boundary
tones and pitch accents of Turkish intonation.
47
Chapter 2: Intonational phonology of Turkish
In this chapter I will describe the intonational phonological model of the prosody of
standard Turkish as spoken by the participants in this study. The intonational model and prosodic
structure proposed in the current study are based on data collected in a series of experiments
between the years 2011-2014. A total of 16 Turkish subjects who are living in Los Angeles less
than six years participated in various parts of the data collection process though not all subjects
participated in all tasks. All subjects were born in Turkey and speaking Turkish as their native
language. Turkish is defined here as standard Turkish based on the dialect of Istanbul.
Each subject first read the randomized list of sentences to himself/herself. After
familiarizing themselves with the sentences, subjects read each sentence to a USB microphone
attached to a laptop computer. The sentences were presented on a powerpoint slide one sentence
at a time. The subjects were allowed to re-read a sentence if it was produced disfluently. The
recordings were made and analyzed in Praat (Boersma and Weenik 2015).
The intonational model I propose for Standard Turkish is developed using the theoretical
framework of the Autosegmental-Metrical theory of Intonational Phonology, as described in
Chapter One. The model includes a full inventory of pitch accents and boundary tones.
2.1 Tones
The current intonational model recognizes multiple levels of hierarchical prosodic units
that are demarcated by some edge tone. Based on the production data analyzed, I propose three
tonally marked prosodic units for Turkish: (i) Prosodic Word (PW), (ii) Intermediate Phrase (ip),
and (iii) Intonation Phrase (IP), all of which are marked by an edge tone. A prosodic word is
marked on its left edge, while an intermediate phrase and Intonation Phrases are marked by a
right edge tone. In addition to these edge tones, I will also propose an additional edge tone, Tn,
that marks the left edge of the most prominent word, i.e., nuclear word, in the sentence (cf. ‘n’ in
Tn refers to ‘nuclear’).
Pitch accents are realized on prominent syllables within a phrase. Data suggests that the
type of pitch accent depends on whether it marks post-lexical prominence on a nuclear word or
48
on a prenuclear word. A prenuclear pitch accent diplays a regular rising nature. The nature of
nuclear pitch accent depends on the sentence type. In declarative sentences, including focus
declaratives, nuclear pitch accent has a highly compressed pitch range, sometimes not showing
any f0 rise at all. On the other hand, in questions, nuclear pitch accent is realized with the highest
peak in the sentence. Following the three-way distinction, I proposed three different pitch
accents in the phonological model of Turkish intonation. Section 2.1.1 presents the realization of
each pitch accent using pitch track examples. Section 2.1.2 is on the analysis of intermediate
phrase boundary tones and the prominence marking left-edge tone. Finally, Section 2.1.3
displays pitch tracks examples from different sentence types to present the IP boundary tones.
Section 2.3 is a summary of this chapter.
2.1.1 Pitch accents: H*, !H*, ^H*
In this section, I will present the phonetic and distributional characteristics of the pitch
accents proposed in the current phonological model of Turkish intonation. While doing so, I will
also compare the analysis proposed here with those in the two previous models; Kan (2009) and
Kamalı (2011).
I propose three pitch accents in the inventory of Turkish pitch accents: H*, !H*
(downstepped High pitch accent), and ^H* (upstepped High pitch accent). A H* pitch accent is
associated with the stressed syllable of prenuclear prosodic words. A !H* tone marks the nuclear
word in declarative sentences, both in neutral and focus contexts, and an ^H* pitch accent is
realized on the stressed syllable of the most prominent word in questions. Unlike Kan (2009) or
Kamalı (2011), who proposed bitonal pitch accents such as L+H* (Kan 2009) or H*L (Kamalı
2011), in this present model the low L tone is analyzed as marking the left edge of a prosodic
word while a monotonal pitch accent; H*, !H*, or ^H*, is associated with the stressed syllable of
a phrase medial prosodic word. Thus according to the current model, a prosodic word that hosts a
pitch accent can have one of the three phonological representations given in 18.
(18) i.
L
[ …H*]
PW
: pre-nuclear words
ii.
L
[ …!H*]
PW
: nuclear word in declaratives
iii.
L
[ …^H*]
PW
: nuclear word in questions
49
In short, the rising pitch contour on a prosodic word is decomposed into discrete L and H
tones where the L tone is docked onto the left edge of a prosodic word while the High tone is
associated with the stressed syllable, regardless of the location within a word, marking post-
lexical pitch accent. In the next two subsections, I will provide pitch track examples to display
each pitch accent.
Prenuclear pitch accent:
L
[ …H*]
PW
This subsection is on the nature of prenuclear pitch accents. I propose that post-lexical
prominence for prenuclear prosodic words is marked by a monotonal pitch accent, H*, while a L
tone is associated with the left edge of every prosodic word. This tonal analysis is exemplified in
Figure 17. The pitch track in Figure 17 displays tonal realization of a four-word declarative
sentence produced in neutral focus. To display the location of lexical stress, a stressed syllable is
written in capitals on the words tier (2
nd
tier). The other tiers are for tones (1
st
tier) and English
glosses (3
rd
tier). The same format is used for all pitch track examples. For now, the reader may
ignore the exact labeling of each tone and just focus on the nature of the rising tone on
prenuclear prosodic words and its tonal analysis.
50
Figure 17. Pitch track of a four-word declarative sentence in all-new context.
In Figure 17, there are two prenuclear non-finally stressed words and both of them are
hosting a rising tone; starting low from the beginning of the prosodic word rising gradually
towards the lexically stressed syllable. In the current phonological model, this gradual rising
pitch contour towards the stressed syllable of a prosodic word is represented by an L tone at the
left edge of a prosodic word and a monotonal H* pitch that is associated with the stressed
syllable of the prosodic word. As mentioned above, this monotonal pitch-accent analysis is
different from the analysis by Kan (2009) and Kamalı (2011) both of which propose a bitonal
pitch-accent. I do not propose a bitonal pitch accent for Turkish, and in what follows I will show
the reasons for this. First, I will show why not L+H* as is proposed by Kan (2009).
In the intonational model developed using the AM model of intonational phonology, Kan
(2009) does not motivate the use of a bitonal pitch accent L+H* (or L+!H*). The choice of this
particular bitonal pitch accent rather seems to capture the rising nature of the pitch contour only,
but not really reflecting the phonetic nature of the rising tone itself. A bitonal pitch accent like
L+H* in the AM tradition means a tonal movement from a preceding syllable such that the
starred tone, H*, is aligned with the stressed syllable and the unstarred (leading) tone, L in this
L
H*
L
H*
NUCLEAR
WORD
PRE-NUCLEAR W O R D S
51
case, determines the pitch of the preceding syllable, i.e., under ideal conditions, an L+H* pitch
accent means an L tone target on the syllable immediately preceding the lexically stressed
syllable, and rising to the High tone target on the stressed syllable. A schematic representation of
an ideal L+H* tone is given in Figure 18:
Figure 18. Schematic representation of an ideal L+H* bitonal pitch accent.
σ σ σ ˈσ (σ)
However, I did not necessarily observe this tonal shape to be the prevailing contour on
phrase medial prosodic words. An ideal scenario to exemplify the nature of this rising tone, i.e.,
whether it is gradually rising towards the stressed syllable from the beginning of a prosodic word
or whether there is an L tone target on the syllable that immediately precedes the lexically
stressed syllable, would be to increase the number of syllables in a prosodic word where the
lexically stressed syllable is farther away from the beginning of the prosodic word. This phonetic
distance would provide enough time for the low (L) target to be fully realized and not get
undershot before the f0 rise begins. This is exemplified in pitch tracks in Figure 19 and Figure
20. The target prosodic word in both figures is the first word, lemanın and lemanların,
respectively both of which are lexically stressed on their final syllables. In Fig. 19, the target
word is composed of three syllables and in Fig. 20 the number of syllables in the first word is
increased to four. If the underlying tone is a bitonal pitch accent like L+H* as is suggested by
Kan (2009), the expectation would be to see the low tone to be realized on the penultimate
syllable of the first word before rising to the final syllable, rather than a gradual rise towards the
stressed syllable from the beginning of the word. As can be seen in both Figure 19 and Figure 20,
the tonal contour on the target word starts rising towards the High target right after the beginning
of the prosodic word. This remains to hold true even when the number of syllables is increased
as exemplified in Figure 20.
H*
L
52
Figure 19. Gradual rise of f0 on a three-syllable finally stressed word.
Figure 20. Gradual rise of f0 on a three-syllable finally stressed word.
53
As can be seen in both figures, unlike what would be expected from an underlying L+H*
tone, the tonal contour on both of the target words display a gradual rise. This observation
suggests that L tone does not target the syllable that precedes the lexically stressed syllable.
Therefore I do not suggest L+H* as the underlying pitch accent.
Now let’s look at a different analysis of the same phenomenon proposed by Kamalı
(2011). In her model of Turkish intonation, Kamalı (2011) proposes H*L as the underlying pitch
accent. This H*L tone is first suggested by Levi (2005) who argues Turkish to be a lexical pitch
accent language, and following the generality among pitch accent languages, she suggested H*L
as the underlying tone. Following Levi (2005), Kamalı (2011) assumes Turkish to be a pitch-
accent language as well and adopts H*L as the underlying pitch accent. What is more she
suggests that this labeling is “more desirable for a more uniform treatment of the tonal inventory
of Turkish.” (Kamalı 2011: 88). While it is desirable to provide a uniform analysis of the
phonology of Turkish intonation, the underlying phonological form needs to be derived from the
patterns reflected in the phonetic nature of the pitch contours. Phonetically an H*L pitch accent
means a sharp fall from High to Low tone, a property of the tonal contour observed in Japanese
for instance. The following pitch track in Figure 21 is taken from Beckman & Pierrehumbert
(1986:264).
Figure 21. Sample pitch track of a Japanese phrase that shows the realization of an
HL tone unit. From Beckman & Pierrehumbert (1986:264).
The sharp fall to the low target can be seen in both HL tone unit in Figure 21. A
schematic representation of the ideal realization of an H*L pitch accent, then would be like the
one given in Figure 22 where high target is aligned with the stressed syllable and the trailing low
target with the following syllable.
54
Figure 22. Schematic representation of H*L pitch accent.
σ σ ˈσ σ σ
An underlying tone like this suggests that in an ideal situation, i.e., when there are
enough syllables for both H and L tones to be fully realized and not undershot due to lack of
enough phonetic space, we would expect to see f0 minimum, i.e. the Low target, to be reached
right after the f0 maximum, i.e. High target, has been accomplished. In other words, we would
not expect a gradual fall towards the Low target following the High tone. The true nature of the
fall can be observed by increasing the phonetic space between two High targets. This would
provide enough time for the L target following the first H tone to be fully realized before the f0
rise towards the second H target begins, and not undershot due to tonal crowding. This is
exemplified in Figure 23, where the relevant domain is the L target following the first H* pitch
accent. In this pitch track example, the first word on which the first H* pitch accent is realized is
non-finally stressed. The reason for choosing the first word as non-finally stressed and the
second one as finally stressed is to be able to see whether the L tone in [H L] sequence is
targetting the immediately following syllable as is implied by H*L pitch accent and as suggested
by Kamalı(2011) or the beginning of the next prosodic word which is what I claim. If the L tone
targets the beginning of the next prosodic word, then we would expect the phonetic realization of
f0 fall to be gradual. The pitch track in Figure 23 displays that this is actually the case; the L tone
after the H* pitch accent does not display a sharp fall right after High target is reached but rather
falls gradually throughout the rest of the word and L target is reached at the beginning of the next
word, limanı.
H*
L
55
Figure 23. Target of L tone.
As is exemplified in the above figures, phonetic realization of [L H] or [H L]
sequences does not support a bitonal pitch accent analysis such as L+H* where the H* would be
preceded by an f0 valley or H*L accent where we would observe a sharp f0 fall following the H
target. Instead the L tone targets the left edge of a prosodic word. Therefore in the current model
of Turkish intonation, L target is phonologically associated with the left edge of a prosodic word
and post-lexical pitch-accent is marked by H* monotonal pitch accent realized on the stressed
syllable of prenuclear prosodic words. The difference between the previous two models and the
current model regarding prenuclear pitch accents can be summarized as follows:
Kan (2009) Kamalı (2011) Current model
L+(!)H* H*L
L
[ …H*]
PW
In this subsection, I have provided examples to argue for the alignment of phrase-medial
L and H tonal targets realized on prenuclear prosodic words, and compared my proposal with the
two previous intonational models developed in the AM framework.
H*
L
H*H-
56
In the next section, I will present the tonal characteristics of the nuclear pitch accent,
again comparing current analysis with Kan and Kamalı’s proposals.
Nuclear pitch accents:
L
[ …!H*],
L
[ …^H*]
In this section, I will present pitch track examples to establish the tonal realization of
nuclear pitch accent, i.e., the pitch accent associated with informationally most prominent word
in the sentence. I will look at this phenomenon in declarative sentences produced in neutral and
narrow focus contexts, and in questions.
I will first start with the tonal characteristics of sentence prominence in neutral
declarative sentences. In general, immediately preverbal position is associated with sentence
prominence in Turkish. Therefore in a simple SOV sentence, object is the nuclear word. In the
experimental literature, a recurrent finding on the acoustic correlates of sentence stress is that the
most prominent word is realized with higher fundamental frequency (e.g. Cooper et al. 1985,
Fear et al. 1995, de Jong 1995). Data suggests that this generalization does not apply to Turkish.
Unlike existing literature on the acoustic properties of sentence stress, in Turkish fundamental
frequency on the nuclear word in declarative sentences displays a highly compressed pitch range,
sometimes with no visible rise at all. This is exemplified in Figure 24. The pitch track in this
figure represents tonal realization of a five-word sentence where the fourth word, limanda, is the
nuclear word. Lack of any clear f0 rise on the nuclear word can be clearly seen in this figure. In
the current model of Turkish intonation, I mark this tonal property of the nuclear pitch accent
using !H* pitch accent. The Nuclear word is tonally marked by a L tone at the left edge of the
prosodic word and a !H* pitch accent on the stressed syllable of the nuclear word.
57
Figure 24. Tonal realization of nuclear pitch accent in a five-word declarative
sentence produced in neutral focus context. Fourth word is the nuclear word.
One might think that the compressed nature of pitch range on the nuclear word in Figure
24 is due to phonetic declination as nuclear word is the fourth word in the sentence. However the
lack of substantial f0 rise on the nuclear word is still observed even when the nuclear word is the
second word in the sentence. This is exemplified in Figure 25.
NUCLEAR WORD
58
Figure 25. Tonal realization of nuclear pitch accent in a three-word declarative
sentence produced in neutral focus context. Second word is the nuclear word.
Lack of substantial tonal event on the nuclear word is not confined to sentences produced
under neutral context. The same fact holds true even in sentences with narrow focus (Ipek 2011).
The following pitch track in Figure 26 displays sample pitch track of a four word sentence where
the second word is narrowly focused. Similar to the neutral focus context, the narrowly focused
word is realized with a highly compressed pitch event.
!H*
NUCLEAR WORD
WORZZZZZZZZD
59
Figure 26. Tonal realization of nuclear pitch accent in a four-word declarative
sentence produced in narrow focus context. Second word is narrowly focused.
Even more interestingly, compression of pitch range in marking sentence prominence is
observed for the first word in a sentence, as well. This, I exemplified using superimposed pitch
tracks in Figure 27. The pitch tracks in this figure display tonal contour of the same four word
sentence produced in neutral context (black pitch track) and when the first word is narrowly
focused (red pitch track). It can be seen in this comparison that the f0 rise on the first word when
it is narrowly focused is much smaller (33Hz higher than the preceding L) compared to its
neutral counterpart (54 Hz higher) and f0 peak is realized at 272 Hz for the first word under
neutral condition and at 238 Hz when it is narrowly focused.
!H*
NARROWLY
FOCUSED
60
Figure 27. Superimposed pitch tracks for the sentence ‘Emine çayını mendile
damlatmış’ (Emine spilled her tea on the handkerchief) under two focus
conditions: Neutral (black pitch track), and when the first word is narrowly
focused (red pitch track).
Further evidence on the lack of substantial f0 event on the nuclear word can be seen in
contexts where the verb, the last word in a sentence in the canonical order, is narrowly focused.
In neutral contexts, the verb is usually the postnuclear unit as the sentence prominence falls on
the immediately preverbal unit in such contexts, and f0 reaches its minimum at this point. The
superimposed pitch tracks in Figure 28 displays sample tonal event where the pitch contours
represent the production of four-word sentence produced in neutral context (black pitch contour)
and when the fourth word, the verb, is focused (red pitch contour). It can be seen in this figure
that the tonal event on word four when it is narrowly focused is not any different from its neutral
counterpart.
NEUTRAL FOCUS
WORD 1 FOCUS
WORD 1
272 Hz
238 Hz
61
Figure 28. Superimposed pitch tracks for the sentence, ‘Emine çayını mendile
damlatmış’ (Emine spilled her tea on the handkerchief), produced under neutral
context (black pitch track), and when the verb is narrowly focused (red pitch
track).
These pitch track examples suggest that sentence prominence in declarative sentences,
both in neutral context and narrow focus contexts, does not trigger any pitch range expansion on
the nuclear word itself unlike what would be expected based on the experimental literature on
acoustic correlates of sentence prominence. Even more strikingly, if there is any change on
fundamental frequency due to prominence, it seems to be in the direction of lowering f0 and not
increasing it, e.g. maximum f0 on a narrowly focused first word in a sentence is lower than its
neutral counterpart as shown in Figure 27.
Pitch track examples given in Figures 26-28 display sample tonal contours of one
sentence produced under neutral and narrow focus conditions. I tested current model’s proposal
that increase in fundamental frequency in encoding sentence prominence is systematically
missing in Turkish intonation. In order for this, I created five four-word sentences, and five
Turkish native speakers produced these sentences under four narrow focus conditions, narrow
focus on each word, and in neutral focus condition. The bar graph in Figure 29 shows mean
maximum f0 value (semitone) on the stressed syllable (final syllable) of each four word in
neutral condition and under narrow focus condition. A significant difference is displayed by a
star (*) on the relevant bar and statistical results are given in Table 1.
NEUTRAL FOCUS
WORD 4 FOCUS
WORD 4
62
Figure 29. Mean f0 values of the stressed syllable (final syllable) of each word
under neutral and narrow focus conditions.
It can be seen from the chart that when the first word or the second word is narrowly
focused, maximum f0 on the narrowly focused word is lower than that of its neutral counterpart
and this difference in both cases is statistically significant (see Table 1 for the statistical results).
For word 3 (nuclear word in neutral condition), and word 4 (post-nuclear word in neutral
condition) maximum f0 is not any different under neutral or narrow focus conditions.
Table 1. Results for the mean f0 difference under neutral and narrow focus
conditions.
Intercept z p
word1 1.9497 8.535 <0.001 *
word2 1.8688 4.676 <0.001 *
word3 -0.1837 -0.703 0.891
word4 -0.2013 -0.602 0.982
These findings from narrow focus condition replicate the results from Ipek (2011) and
provide further evidence to support the claim that increase in fundamental frequency due to
prominence is not a property of Turkish intonation in declarative sentences. Instead, the change
in fundamental frequency, if any at all, is in the direction of lowering. This property of tonal
4
6
8
10
12
14
16
18
20
word1 word2 word3 word4
Max f0 (semitone)
NEUTRAL
FOCUS
*
*
63
event on the nuclear word, I mark by using !H* pitch accent in the tonal inventory of Turkish.
However, when the most prominent word is sentence initial, a regular H* pitch accent is used
because the downstep diacritic, “!”, implies lowering in f0 range with respect to the preceding
high tone, and this is not the case when the most prominent word is sentence initial. However,
the fact that even a narrowly focused sentence initial word has lower f0 peak compared to its
neutral counterpart suggests that narrowing pitch range is part of the phonlogical specification of
nuclear word in Turkish.
Nuclear pitch accent has been analyzed different from the prenuclear pitch accents in Kan
(2009) and Kamalı (2011) as well. However, both are quite different from the proposal I make
here. In these two models, only those sentences with neutral focus context are analyzed.
Therefore, I will keep the comparison in the neutral context.
Nuclear pitch accent in Kan (2009) and Kamalı (2011)
Kan (2009) proposed four nuclear pitch accents. These are L+H*, L+!H*, H*, !H*. It is
worthwhile to remember at this point that Kan (2009) is not concerned with the phonological
analysis of Turkish intonation; the edge tones and pitch accents rather reflect the surface shape of
the tonal contour. In line with this, the alternation of these four nuclear pitch accents do not
contribute to any semantic change but rather reflect the different phonetic realizations of the
nuclear pitch accent. Of these four nuclear pitch accents, monotonal ones, H* and !H*, were
used when the pitch contour on the nuclear word did not display any rise but was rather flat, an
observation that is in line with the current data I have presented. The choice of H* or !H* is
purely phonetic: a lower level nuclear accent was marked by !H*, otherwise H* was used to
mark flat nuclear tone which maintained the same level tone as the previous high tone.
Realization of the two monotonal nuclear pitch accents is exemplified in Figure 30 and Figure
31, respectively.
64
Figure 30. Tonal analysis of ‘Ayla yirmi mandalina yedi’ (Ayla ate twenty
tangerines). Nuclear pitch accent, !H*, does not display any rise. Peak f0 on the
nuclear word is realized lower than that of the previous f0-peak. From Kan
(2009:184).
‘Ayla ate twenty tangeries.’
!H*
65
Figure 31. Tonal analysis of ‘Ayla Yalova’yı zor bulmuş’ (Ayla had difficulty in
finding Yalova). Pitch contour on the nuclear word, zor, displays no rise. Peak
f0 on the nuclear word is almost the same as previous f0 peak. From Kan
(2009:161).
Pitch tracks in Figure 30 and Figure 31 display two different tonal analysis of nuclear
pitch accent when it displays no f0-rise. In Figure 30, the nuclear pitch accent is marked by !H*
pitch accent because the level tone throughout the nuclear word is realized at a lower level than
the previous f0 peak, while it is marked by H* pitch accent, as exemplified in Figure 31, when
the level tone on the nuclear word is not realized lower than the previous f0 peak. However, as
mentioned earlier, lack of f0 lowering on the nuclear word in Figure 31 is mostly likely to be due
to lack of enough space in between the preceding f0 peak and the nuclear pitch accent because,
in this example in Figure 31, there are no syllables between the preceding high tone, L+H-, and
the nuclear pitch accent, H*.
The bitonal nuclear pitch accents are proposed for those that display substantial f0 rise.
However a substantial rise on the nuclear word did not come out in the corpus of data I collected.
Kamalı (2011: 92) also reports a similar observation, though she restricts the lack of rise to
finally stressed words only: “ […] L+H* nuclear tone […] means that there is a distinct rise that
is not part of the plateau. Curiously, we did not document any such cases.” Kamalı (2011).
H*
66
Now, let’s look at how the tonal phenomenon on the nuclear word is analyzed in Kamalı
(2011). First of all, tonal event on the nuclear word in this model is treated differently depending
on whether the nuclear word is lexically accented or not. Remember that Kamalı (2011) assumes
Turkish to be a lexical pitch accent language. Following this assumption, in her model lexically
accented words, those that are non-finally stressed in the traditional sense, carry H*L lexical
pitch accent. When the nuclear word is non-finally stressed, i.e., lexically accented, it carries the
H*L pitch accent as it is the case for lexically accented prenuclear words. Therefore, no
phonological distinction is made between the nuclear pitch accent and the prenuclear pitch
accents. In sum, according to Kamalı, both the prenuclear words and the nuclear word, when
lexically accented, display a rise and this is marked by H*L pitch accent. Pitch track in Figure
32. display one example from the analysis by Kamalı. The second word, which is non-finally
stressed, is the nuclear word and carries the lexical pitch-accent H*L.
Figure 32. Tonal analysis of nuclear pitch accent when the nucleear word is non-finally
stressed. From Kamalı (2011: 75).
Tonal event on a regularly (i.e., finally) stressed word, on the other hand, is analyzed
completely differently. First of all, a distinction is made between the surface realization of a
lexically accented word and regularly stressed words such that while the first one is suggested to
display f0 rise, the latter one is suggested to retain a plateau throught the nuclear word with no f0
rise. Following this observation no pitch accent is associated with a finally stressed nuclear word.
The plateau-like nature of the pitch contour on the nuclear word is sustained by rightward L-
H*L
NUCLEAR WORD
67
tone spreading. A sample pitch track with a finally stressed nuclear word and its tonal analysis
according to Kamalı (2011) is given in Figure 33.
Figure 33. Tonal analysis of a finally stressed nuclear word. From Kamalı
(2011: 78).
While it is true that pitch contour on the nuclear word often displays a plateau-like
realization, I do not agree with Kamalı (2011)’s proposal that this plateau-like f0 realization is
the property of words that are only finally stressed. Finally stressed nuclear words are sometimes
realized with a f0 rise, though the degree of rise is minor (e.g., Figures 24, 25, 35) and non-
finally stressed nuclear words can also be realized with no f0-rise, as well. The following pitch
track in Figure 34 displays one such example where the nuclear word is non-finally stressed and
does not show any f0 rise.
No T*
NUCLEAR WORD
L-
68
Figure 34. Non-finally stressed nuclear word with no f0-rise.
Figure 35. Finally stressed nuclear word with slight f0-rise.
NUCLEAR WORD
NUCLEAR WORD
69
These two pitch track examples in Figure 34 and Figure 35 show that whether the tonal
event on the nuclear word displays f0 rise or whether it is realized with a level tone is not a
matter of whether the nuclear word is finally or non-finally stressed. Examples of a flat tonal
contour on a non-finally stressed nuclear word is also present in Kan (2009) (see Figure 16 for
one such example). As a consequence, the distinctive tonal property of the nuclear pitch accent
does not lie in different tonal realization of finally stressed versus non-finally stressed words on
the nuclear word as is suggested by Kamalı (2011). What ties finally and non-finally stressed
nuclear words is the smallness of the f0-rise, if there is any at all, compared to prenuclear pitch
accents. Therefore, the distinction need to be made between the nuclear pitch accent and the
prenuclear pitch accents: The main distinctive property of the nuclear pitch accent is its highly
compressed f0 range compared to prenuclear pitch accents. As I have also shown in the focus
data, this fact holds true even for sentence initial nuclear words.
Based on the observation of this difference between nuclear pitch accent and the
prenuclear pitch accents in the production data, I have distinguished the two in the phonological
model I propose: the H* pitch accent is associated with prenuclear pitch accents while the NPA
is marked by !H* to reflect the compressed nature of its pitch range.
So far, I have described the tonal nature of the nuclear pitch accent in declarative
sentences produced in both neutral and narrow focus contexts. I have shown that while a
prenuclear pitch accent is realized by substantial f0 rise, the nuclear pitch accent is marked by a
tonal event that has a highly compressed pitch range. I have provided quantitative data from
sentences produced in the narrow focus context by comparing them with their neutral counterpart
to further support this observation. In addition, I also compared my analysis with two previous
intonational models by Kan (2009) and Kamalı (2011). I have shown that while in both previous
models pitch range compression on the nuclear word is observed, the way this difference of the
nuclear word is analyzed differed from the analysis I provide here. Below I summarize the tones
associated with the nuclear word in decarative sentences in three intonational models:
Kan (2009) Kamalı (2011) Current Model
L+(!)H* H*L (non-finally stressed)
L
[…!H*]
(!)H* No T* (regularly stressed)
70
Now, I will propose another nuclear pitch accent in the tonal inventory of Turkish,
upstepped H* pitch accent, ^H*, which marks sentence prominence in questions. Data suggests
that unlike declaratives, sentence prominence in questions is realized by expanded pitch range on
the prominent word. This is exemplified in Figure 36 where the wh-word, which is the most
prominent word in the sentence, hosts the highest f0 peak (before the H% boundary?).
Figure 36. Pitch track of a five-word interrogative sentence. Nuclear pitch
accent is realized with expanded pitch range.
Pitch track in Figure 37. displays the same phenomenon in yes/no questions. This type of
questions are formed by adding the unstressable clitic –mI after a sentence or a phrase, and the
sentence stress falls on the word to which the clitic –mI is attached to.
^H*
71
Figure 37. Pitch track of a yes/no question. The word the question particle is
attached to hosts the high tone with the highest peak in the entire sentence.
Pitch tracks in Figure 36 and Figure 37 show that sentence prominence in questions is
realized by expanded pitch range on the nuclear word such that the nuclear word hosts the
highest f0 peak in the sentence. This is the exact opposite of the realization of nuclear pitch
accent in declaratives where nuclear pitch accent is realized with compressed pitch range. This
difference is reflected in the phonological model by two separate pitch accents: !H* for
declaratives and ^H* for questions.
Interim summary: In this section, I described pitch accents I propose in the phonological
model of Turkish intonation while comparing the current model with two previous intonational
models; Kan (2009) and Kamalı (2011). The comparison was made for declarative sentences
produced in neutral focus contexts as the two previous models were developed based on this
sentence type only. Based on the corpus of data analyzed, I proposed three monotonal pitch
accents in the tonal inventory of Turkish: H*, !H*, and ^H*. The H* pitch accent is associated
with the stressed syllable of prenuclear words. The downstepped high pitch accent (!H*) is
proposed to mark the nuclear pitch accent in declarative sentences including narrow focus
sentences. The choice of !H* as NPA in the current model is due to the highly compressed nature
L H* L H*H- L H* LH- L H* H*Hn L ^H* L%
lemaNIN ye!eN" alMANyaya araBAYla giD"yomu
leman’s nephew to germany by car going.Qpart
70
200
100
150
Pitch (Hz)
Time (s)
0 2.249
^H*
72
of the nuclear pitch accent. The third pitch accent, ^H*, is proposed for marking sentence
prominence in questions because pitch accents on the most prominent word in such sentences is
realized with the highest f0 peak in the entire sentence. Unlike Kan (2009) and Kamalı (2011), I
did not propose a bitonal pitch accent such as L+H* or H*L. Instead, the L tone which is
analyzed as the leading or the trailing tone in Kan (2009) and Kamalı (2011), respectively, is
associated with the left edge of prosodic words.
Table 2. Comparison of pitch accents in the three intonational models
Kan (2009) Kamalı (2011) Current model
Prenuclear
pitch accents
{L+H*, H*} H*L (lexically accented) H*
Nuclear pitch
accents
{L+H*, L+!H*, H*,
!H*}
H*L (lexically accented)
!H* (declaratives)
^H* (questions)
Pitch accents proposed in Kan’s model were in free variation depending on the phonetic
shape of the contour, therefore I used curly brackets to represent this variability. In Kamalı
(2011), only words that are non-finally stressed in the traditional sense are associated with a
pitch accent, which is a lexical pitch accent. Words that are traditionally finally stressed are
assumed to be not accented in Kamalı’s model. Since all prenuclear words in Kamalı’s stimulus
were a one word phrase, a finally stressed word was always phrase final, and a high tone realized
on its final syllable is analyzed as a phrase final tone, H-. The status of phrase-medial finally
stressed words is not examined in this model, and a finally stressed nuclear word has no tone
associated with the final syllable.
I will now move onto the description of boundary tones in the next section.
2.1.2 Boundary tones
In this section, I will provide pitch track examples to display the boundary tones in the
intonational phonology of Turkish. In the current model, the next prosodic domain above the
prosodic word is intermediate phrase (ip) which is demarcated by a boundary tone (T-) at its
right edge. In addition, I will also introduce another boundary tone (Tn) that marks the left edge
73
of nuclear word in declarative sentences, including focus declaratives. Finally, I will present the
boundary tones for the highest prosodic domain, Intonational Phrase (IP), identified in the
prosodic structure. I will first start with the intermediate phrase boundary tones (T-) and then
move onto boundary tone marking the left edge of nuclear word (Tn). Finally, I will present pitch
contours to exemplify the intonation phrase (IP) boundary tones.
Boundary tones marking the right edge of an intermediate phrase (ip), T-
One or more prosodic words form an intermediate phrase (ip) demarcated by a boundary
tone, T-, at its right edge. An ip usually forms a syntactic unit which often corresponds to a
phrase or a clause, and yet the relation is not necessarily one-to-one; i.e., not all syntactic phrases
or clauses form an ip, and likewise not all ips correspond to a syntactic unit. In what follows I
will first display the relation between ips and syntactic units while presenting pitch tracks to
exemplify the ip-boundary tones.
One systematic observation is that the right edge of NPs in Turkish is tonally marked.
This is exemplified in the following three pitch track examples where I varied the number of
words within each subject NP. All the words within subject NP are chosen to be non-finally
stressed so that we can see if there is any tonal event happening at the end of the phrase. That is,
when a word is non-finally stressed, a high tone realized at the end of an NP-final word is not a
pitch accent associated with the stressed syllable of the word but a boundary tone marking the
right edge of the NP/ip.
Figure 38 shows an example pitch track of a sentence where both the subject NP and the
object NP include only one word. Here, the single word phrase shows two rises, one over the
stressed syllable, marked by H* pitch-accent, and the other at the end of the phrase, marked by
LH- intermediate phrase boundary tone.
74
Figure 38. The subject NP, fakülteler, and the object NP, lokantalarını, each contain one
non-finally stressed word. The right edge of each NP is marked by a rising tone, LH-.
Figure 39. The subject NP, lokantanın sandalyeleri, contains two words with non-final stress.
The end of the second word, which is the last word of the subject NP, is marked by a rising tone,
LH-.
H*
LH-
H* LH-
H*
LH-
H*
75
Figure 40. The subject NP contains three words. All words in subject NP have
non-final stress and only the last syllable of the NP is marked by a rising tone,
LH-.
In addition to NPs, Postpositional phrases (PPs) are also tonally marked at their right
edges. Figure 41 displays a sample pitch track of a sentence with a PP, leman aracılığıyla, whose
right edge is marked by LH- tone. In this example, PP-final word is chosen to be non-finally
stressed so that we can be certain that any tonal event realized on its final syllable is not a pitch
accent but a boundary tone demarcating its right edge.
H*
H*
H*
LH-
76
Figure 41. Non-finally stressed postposition, aracılığıyla, hosts two rising tones,
one on its stressed syllable marked by H* and the other on its final syllable
marked by LH-, which demarcates the right edge of PP. ‘I will send the money to
the customer through Leman on Friday’.
On the other hand, not every syntactic constituent forms an intermediate phrase. For
instance, an adjective phrase does not constitute its own prosodic grouping. The following two
figures, Figures 42 and Figure 43, illustrate this lack of prosodic grouping for adjective phrases.
Figure 42 shows a pitch track of a sentence where the subject NP includes an adjective and a
noun, while the subject NP in Figure 43 includes an adjective phrase and a noun. But in both
figures only the end of the subject NP is marked by a rising tone, LH-.
LH-
POST-POSITIONAL
PHRASE
H*
77
Figure 42. The LH- boundary tone marks the end of the subject NP ‘bembeyaz sandalyeler’,
which conrresponds to an ip.
Figure 43. The subject NP includes an adjective phrase, tamamıyla bembeyaz ‘completely
white’ before a head noun, but the adjective phrase-finaly syllable is not marked by any tone.
LH- tone marks only the en of the subject NP, forming an ip.
H*
H*
LH-
ADJECTIVE N O U N
H*
H*
H* LH-
ADVERB ADJECTIVE N O U N
78
Similar to phrases, certain clauses also form one ip demarcated by a right edge tone. For
example, a noun clause is marked by a rising tone at its right edge. This is exemplified in Figure
44 where the right edge of the noun clause is marked by LH- ip-boundary tone.
Figure 44. Right edge of noun clause is marked by a rising tone, LH-. Pitch
track of a subject NP clause is extracted from the sentence Alanyalıların
misafirleri ağırlamamaları (dün bazılarını endişelendirdi) (That people of
Alanya did not host the guests (upset some yesterday)). Also note that the end of
the first word, NP, is marked by an ip bounary, LH-.
Likewise, the right edge of an if clause, for example, is also marked by a rising tone. This is
exemplified in Figure 45.
LH-
79
Figure 45. Sample tonal realization of an if clause. Extracted from the sentence
Almanyalılar Hollandalıları yenerse, Brezilya gruptan çıkar (If Germans beat the
Dutch, Brazil will move to the next round).
On the other hand, not every clause is tonally marked. One such clause that does not form a
prosodic unit is a relative clause (RC) that precedes the nouns they modify, which I will refer to
as pre-relatives
5
. A subject relative clause is formed by the suffix –(y)An. On the other hand,
when relativizing a non-subject, the suffix –DIK is used. Sample sentences for each type of
relative clause is given in 19.
(19) a. Subject Relative Clause
Alanya-nın köy-ü-ne giden adamlar yarın muhtarı görecek.
Alanya-GEN village-3POSS-DAT go-AN man.PLU tomorrow chief-ACC see-FUT
“The men who went to Alanya’s village will see the chief tomorrow.”
5
There is also another type of relative clause in Turkish which is known as -ki relatives. In sentences with -ki
relative clauses the modifier -ki relative clause follows the head-noun. Tonal properties of such sentences is not
analyzed in this study. Interested readers can refer to Kan (2009) for a detailed tonal analysis of -ki relative clauses.
80
b. Non-subject Relative Clause
Almanya-nın Hollanda-ya at-tığ-ı gol-ler yarın televizyon-da göster-il-ecek.
Germany-GEN Holland-DAT score-DıK-3POSS goal-PLU tomorrow televison-LOC
show-PASS-FUT
“The goals that Germany scored against Holland will be shown on TV
tomorrow.”
Data suggests that intonationally, a pre-relative clause, subject relative clause or non-
subject relative clause, does not form a prosodic unit of its own but is realized as part of the
NP/ip it appears in. This tonal property of pre-relative clauses, i.e., they do not form a prosodic
unit marked by some boundary tone is also reported by Kan (2009). Sample tonal realization of
each type of relative clause is displayed in Figure 46 and Figure 47.
Figure 46. Pitch track of the phrase ‘The man who went to Alanya’s village’ in (19a)
shows lack of tonal marking at the right edge of a subject relative clause.
RIGHT EDGE
OF RC
81
Figure 47. Pitch track of the phrase “The goal that Germany scored against
Holland” in (19b) shows lack of tonal marking at the right edge of an object
relative clause.
In this subsection, I have presented the tonal realization of an intermediate phrase as well
as the relation between an intermediate phrase and syntactic units such that certain syntactic
phrases/clauses form one prosodic unit and their right edge is marked by a rising tone. I have
also shown that not every syntactic phrase/clause form one prosodic unit. For example, an
adjective phrase does not form its own prosodic grouping. Likewise, a relative clause does not
form a prosodic unit of its own. In order to see if there is a tonal event happening at the end of a
phrase/clause, I used non-finally stressed words and marked the tonal event on the final syllable
of a non-finally stressed word as LH- to demarcate an ip-phrase boundary.
The next question at this point is how to analyze the tonal event on a finally stressed
word when it is also marking the right edge of an intermediate phrase. The next subsection
addresses this question.
RIGHT EDGE
OF RC
82
Analysis of ip-boundary tone when ip-final word is finally stressed
As I mentioned earlier, majority of the words in Turkish are lexically stressed on their
final syllables. This being the case, the right edge of an intermediate phrase often coincides with
a lexically stressed syllable, and therefore when a finally stressed word appears as the last word
in an intermediate phrase, the tone realized on its final syllable has two functions: (1) pitch
accent, and (2) an ip boundary tone. In order to specify this dual function of the ip final high tone
realized on the stressed syllable, I use H*H- notation in the intonational phonological model of
Turkish. This way, I distinguish a high tone realized on the phrase medial finally stressed word,
which I mark using H* pitch accent, from that of a phrase final word that is lexically stressed on
its final syllable. Pitch track example given in Figure 48 displays one such example with a two
word subject NP which forms one intermediate phrase. In this two-word intermediate phrase,
lemanın oğlu, both words are lexically stressed on their final syllables and both host a high tone
on their finally stressed syllables. The first high tone is realized on the final syllable of a phrase
medial finally stressed word, lemanın, and therefore is marked by a regular prenuclear pitch
accent, H*. The high tone which is realized on the stressed syllable of the second word which is
also the intermediate phrase final word, oğlu is marked by H*H- indicating its dual function:
both as a pitch accent (“*”) and as an ip-boundary tone (“-”).
83
Figure 48. The subject NP contains two finally stressed words both of which
host a high tone. The phrase medial high tone is marked by H* pitch accent
indicating its phrase-medial status, while the phrase final one is marked by
H*H- tone.
Distinguishing phrase medial and phrase final high tone on a finally stressed word is
another major distinction of the current model from that of Kan’s (2009) and Kamalı’s (2011)
where no such distinction is made in the tonal tier on such words.
In Kan’s model, the next prosodic unit above the prosodic word is a Phonological Phrase
(PPh), and the end of a PPh is tonally marked. The nature of the tone at the end of a PPh is
determined by the phonological context as described in 20.
(20) a. When a syllable that bears phrase stress (+ a corresponding pitch accent) is non-
PhP-final, a non-accentual pitch event controls the melody between the pitch
accent and the end of the phrase, which we refer to as a phrase accent adopting
the terminology of the AM model.
H*H- H*
84
b. When a phrase-stress-bearing syllable is located at the end of a PhP, a
corresponding pitch accent marks the end of the PhP, where a phrase accent is
not instantiated. (Italics are mine)
(Kan 2009:83)
Following 20b, a high tone on a finally stressed word is analyzed as a pitch accent for
both phrase medial and phrase final words in Kan’s model. Pitch track examples in Figure 49
and Figure 50 exemplify this analysis by Kan (2009). In Figure 49, the words, yedi and alanyalı,
form one PPh in which the finally stressed yedi is phrase medial and hosts L+H* pitch accent on
its lexically stressed syllable. In Figure 50 the same two words, according to Kan’s analysis,
form their own PPhs in which case the adjective yedi is phrase final and the High tone on its final
syllable is still marked by L+H* pitch accent.
Figure 49. Sentence initial adjective yedi (‘seven’) is phrase medial and hosts
L+H* pitch accent. From Kan (2009:182).
L+H* L+!H- !H* L- L%
[[Yedi Alanyalı]PPh [ev-lerin-i arı-yo-muş]PPh]IP
seven Alanyalı house -3Pl.Poss-Acc look for-Prog-Evid
‘Seven people from Alanya have been looking for their house’
L+H*
85
Figure 50. Sentence initial adjective yedi (‘seven’) forms its own PPh and its
right edge is still marked by a L+H* pitch accent. From Kan (2009:182).
L+H* L+H* L+!H- L+!H* L- L%
[[Yedi]
PPh
[Alanyalı]PPh [ev-lerin-i arı-yo-muş]PPh]IP
seven Alanyalı house -3Pl.Poss-Acc look for-Prog-Evid
‘Seven people from Alanya have been looking for their house’
Therefore, in Kan (2009), regardless of whether a high tone on a finally stressed word is
phrase-medial or phrase-final, it is marked by a pitch accent. This being the case, the prosodic
structure of a sentence is not always reflected in its intonation. This can clearly be seen in the
mismatch between the prosodic structure of the sentence in Figure 50 and its intonational
analysis: While the finally stressed word, yedi, is analyzed as forming its own PPh, the rising
tone realized on its stressed syllable is marked as a L+H* pitch accent, thus not indicating
whether the word forms one PPh or part of a larger PPh.
Now, let us turn to the analysis of the same phenomenon in Kamalı’s (2011) model.
Kamalı (2011) provides pitch track examples of sentences where prenuclear phrases are
composed of one word only. Therefore, a finally stressed word in such examples is always
phrase final, and the high tone on the final syllable of such prenuclear words is analyzed as a
phrase final tone, H-. Figure 51 below displays a sample analysis of prenuclear high tone on
finally stressed words according to Kamalı.
L+H*
86
Figure 51. Two prenuclear high tones on a finally stressed word are marked as
H-. From Kamalı (2011:68).
Since Kamalı (2011) does not provide sentences with any multi-word phrases, it is not
clear how a high tone on a phrase medial, finally stressed word would be analyzed in this model.
In short, neither Kan (2009) nor Kamalı (2011) distinguish phrase medial versus phrase
final high tone in the tonal tier on finally stressed words. In the current model, I distinguished
these two: If a finally stressed word appears ip-medially and hosts a high tone, this high tone is
marked by H* pitch accent reflecting its phrase medial status. If, on the other hand, it appears ip-
finally, this time the word final high tone is marked by H*H- boundary tone, indicating its
function of phrase final status. In Chapter 3, I compare H* and H*H- paradigmatically and
provide quantitative data from a production experiment to support the distinction between phrase
medial high tone and phrase final high tone on finally stressed words. But before doing that, I
will provide pitch track examples here where I compare both syntagmatically to show that these
two are indeed different.
The following pitch track in Figure 52 display typical pitch contour of a declarative
sentence produced in neutral focus context. The subject of this sentence is a four word NP,
fakültenin lokantalarının sandalyelerinin badanaları, where each word is non-finally stressed
and each word hosts a H* pitch accent on its stressed syllable within this NP. Although the rules
or domain of downstep in Turkish has not been systematically analyzed, it is a recurrent
observation in my data that pitch accents within the same ip undergo declination. This can be
seen in Figure 52. In order to exemplify declination within an ip/NP, I used words that are non-
finally stressed so that we know that the high tone realized on the stressed syllable marks lexical
H- H-
NUCLEAR
WORD
87
prominence on the stressed syllable rather than a phrase edge. As can be seen in Figure 52, peak
f0 on successive H* pitch accents within the subject NP/ip are realized at a lower level than that
of the previous one. However the phrase final tone, LH-, has the same f0 level as the preceding
High tone. This suggests that while successive H* pitch accents within a same prosodic phrase
are lowered along the declination slope, phrase final High tone does not follow this decliantion
slope, reaching about the same level of f0 as the preceding H* pitch accent .
Figure 52. Using non-finally stressed words to display declination within an
ip/NP.
The above observation clearly shows the difference between a phrase medial high tone
and a phrase final high tone on finally stressed wordsboundary tone. Following this, the
expectation would be the lack of declination for phrase final High tone on a finally stressed
word. This is exemplified in the following Figure 53 with a six word declarative sentence where
the subject NP which forms one ip is composed of three finally stressed words. It can also be
seen in this figure that high tone on the third word, oğlunun, which is phrase medial is lower than
the preceding High tone and yet the high tone on the fourth word, oyuncağı, which is ip-phrase
final, is slightly higher than the preceding High tone.
H*
H*
H*
H*
88
Figure 53. Each word within subject NP is finally stressed. The phrase medial
nouns within subject NP host a H* pitch accent and they undergo declination.
The phrase final high tone, which is marked by H*H- boundary tone, reaches a
slightly higher level than the preceding high tone.
This syntagmatic comparison of high tone on finally stressed words at two different
prosodic contexts show that they are different; a phrase final high tone is realized with higher f0
peak than a phrase medial high tone in its place. I will provide quantitative data in Chapter 3 that
compares these two tones in a paradigmatic relation and provide further support for the
distinction I make between H* and H*H- in the current intonational model.
Boundary tone marking the left edge of nuclear word, H*Hn (LHn)
This section is on the boundary tone that marks the left edge of the most prominent word
in declaratives. I have shown that a nuclear word in declarative sentences, including narrow
focus declaratives is marked by a pitch accent that has a highly compressed pitch range. This is
quite unusual from the perspective of prosodic typology because existing literature has
H*
H*
H* H*H-
89
recurrently shown that prominent word is often cued by stronger acoustic features such as higher
f0. Data shows that this is not the case in Turkish; i.e., sentence prominence in declaratives is not
marked by any substantial tonal event on the nuclear word itself. I have exemplified this by using
pitch track examples and quantitative data in Section 2.1.1. This unusual characteristics of the
nuclear pitch accent raises the question of whether the language has other intonational cues that
mark prominence. In this section, I will argue that the answer to this question is ‘yes.’ The main
claim of this section is that marking sentence prominence in declaratives, including focus,
involves two tonal events: (1) a pitch accent realized on the stressed syllable of the nuclear word
itself which is marked by !H* tone due to its compressed nature, (2) a high boundary tone (Tn)
realized on the final syllable of the immediatley preceding word. thus marking the left edge of
the most prominent word in the sentence. Therefore, pitch range compression on the nuclear
word is in a way compensated by an additional tone that marks the left edge of the nuclear word
in the sentence. A schematic representation of this tonal realization is given in Figure 54.
Figure 54. Schematic representation of prominence marking in declaratives.
σ σ σ σ] [σ σ ˈσ (σ)
Nuclear word
The ‘n’ diacritic in ‘Hn’ is used to specify that this edge tone is marking the nuclear word
and is different from ip-phrase final tone (T-). This Hn left edge tone marking sentence
prominence can be seen in the two figures below where the non-finally stressed adverbial,
tamamıyla (“completely”), is presented in two different surface positions. In Figure 55, the
adverb? is not immediately preceding the nuclear word, and in this position hosts only one tonal
event which is realized on its stressed syllable. On the other hand, when the same word
immediately precedes the nuclear word which is the case in Figure 56, this time the adverb
carries two tones, one on its stressed syllable and the other on its final syllable. I claim that this
second high tone, marked here as LHn, realized on the final syllable of the adverb has emerged
Hn
!H*
90
because the following word is the nuclear word and therefore marks sentence prominence
together with the nuclear pitch accent.
Figure 55. The non-finally stressed word, tamamıyla, does not immediately
precede the nuclear word and hosts a single High tone realized on its stressed
syllable only.
H*
NUCLEAR
WORD
91
Figure 56. The non-finally stressed word, tamamıyla, immediately precedes the
nuclear word and shows double f0 peaks, one on its stressed syllable and the
other on its final syllable.
These pitch track examples in Figure 55 and 56 suggest that lack of substantial tonal rise
on the prominent word in declaratives seems to be compensated by a second tonal event which is
marking the left edge of the prominent word. If this is the case, we should observe a similar left
edge prominence marking tone in the case of narrow focus sentences as well because prominence
in narrow focus sentences is also marked by a tone that has quite compressed pitch range. Then
the questions is: Is the left edge of narrow focused words also marked by a High tone? The
answer is ‘yes’. This is exemplified in the two pitch track examples below. Figure 57 is a sample
pitch track of a five-word sentence produced in the neutral context. The subject of this sentence
is a two word NP, alanyalı gemiciler, where the first word, alanyalı, is non-finally stressed. In
the neutral context, this non-finally stressed, first word carries a H* pitch accent only realized on
its stressed syllable. After H* pitch accent, there is no rising tone until the final syllable of the
following word, gemiciler, marking the end of an NP/ip.
H*
LHn
NUCLEAR
WORD
92
Figure 57. In the two-word subject-NP, alanyalı gemiciler, the non-finally stressed
firstword, alanyalı, hosts a post-lexical pitch accent realized on its stressed syllable.
Pitch track in Figure 58 displays the tonal realization of the same sentence, but this time
the second word, gemiciler, is narrowly focused as can be confirmed by the amplitude patterns
in the waveform (i.e., all words following the second word show reduced amplitude and their
pitch accents are all deleted). The non-finally stressed, first word, alanyalı, which host one tonal
event, pitch accent, in the neutral context (Figure 57), now hosts an additional pitch event
realized on its final syllable, marked by LHn tone.
H*
93
Figure 58. In the two-word subject-NP, alanyalı gemiciler, the non-finally
stressed first word, alanyalı, carries a high tone on its final syllable when it
immediately precedes the narrowly focused word.
Pitch track example in Figure 58 shows that when non-finally stressed word precedes a
narrowly focused word, a new high tone emerges at the right edge of this immediately
prefocused word; a tone that does not appear when the same word is not immediately prefocus
(Figure 57). This new tone on a non-finally stressed word triggered by being next to the nuclear
word is marked by the LHn label in the tonal tier.
So far, I have used non-finally stressed words to display the observation that tonal
marking of sentence prominence in declaratives is not confined to nuclear pitch accent only, but
requires a left edge high tone realized on the final syllable of the immediately preceding the
nuclear word (I will call this ‘immediately prenuclear word’). When the immediately prenuclear
word is non-finally stressed, an additional High tone emerges on the final syllable of this word.
Majority of the words being finally stressed in Turkish, the next question at this point is what
happens when the word that immediately prenuclear word is finally stressed. If it is the case that
prominence in declaratives triggers manipulation of f0 at its left edge, then we should expect to
see some kind of change on the high tone on a finally stressed word when it immediately
NARROWLY
FOCUSED
H*
LHn
94
precedes the nuclear word in the sentence compared to when it is not. The question at this point
then is: “Does a word final high tone undergo some kind of change when it immediately
precedes the nuclear word?” The answer is yes, and the change is in the direction of f0 boosting
when it is immediately prenuclear. I am going to provide quantitative data for this f0
manipulation under neutral condition in Chapter 3 where I compare phonetic realization of
different types of boundary tones. In this section, I will make this comparison by using sentences
produced under narrow focus condition. For this, I created five question-answer pairs. Each
answer was composed of four words, all of which were lexically stressed on their final syllables.
In all sentences, the third word (immediately preverbal) was the nuclear word. Five Turkish
native speakers produced these sentences under five conditions: narrow focus on each word, and
the neutral condition (base condition). The sentences were produced as an answer to a wh-
question (read by the author) that elicit the relevant narrow focus or neutral condition. A sample
question-answer pair is given in 21.
(21)
Q: Neriman kimi limana göndermiş?
Neriman who.ACC harbor.DAT send.PAST
“Who did Neriman sent to the harbor?”
A: Neriman Eylül’ü limana göndermiş.
Neriman Eylül.ACC harbor.DAT send.past
“Neriman sent Eylül to the harbor.”
Before moving onto statistical results, let us first look at sample pitch tracks. Figure 59
provides superimposed pitch tracks of the sentence given in 21 produced under four focus
conditions. Since we are only interested in immediately prenuclear rise at this point, pitch
contour of the same sentence produced when the first word is narrowly focused is not displayed
here. The three arrows on each peak show the High tone on a finally stressed word when the
following word is narrowly focused. It can be clearly seen in this figure that for each narrowly
focused word --word 2, word 3, word 4 -- immediately prefocus high tone is higher than its
neutral counterpart (black pitch contour).
95
Figure 59. Superimposed pitch tracks of the sentence, Neriman Eylülü limana
göndermiş, produced under four focus conditions.
Superimposed pitch tracks in Figure 59 show that when a finally stressed word is
immediately prenuclear (i.e., prefocus), high tone on its stressed syllable is realized with higher
f0 peak compared to its neutral counterpart. In order to compare the change in f0 due to focus, I
measured peak f0 value 10 ms before the right edge of word 1, word 2, and word 3 when they
were produced under neutral condition and when they were immediately preceding the narrowly
focused word. These f0 values were then converted into semitones to normalize among speakers.
Bar chart in Figure 60 displays the comparison of mean peak f0 values on the first three words
for neutral and prefocus conditions.
Figure 60. Mean peak f0 values of high tone on finally stressed words under
neutral and immediately prefocus conditions.
10
12
14
16
18
20
word1 word2 word3
Max f0 (semitone)
NEUTRAL
PREFOCUS *
*
96
Table 3. Results for the mean peak f0 values of high tone on finally
stressed words under neutral and immediately prefocus positions.
Intercept z p
word1 -0.7692 -3.368 0.00317 *
word2 -0.9053 -2.265 0.0776
word3 -3.7017 -14.171 <0.0001 *
For all words, maximum f0 on the finally stressed syllable was higher when the word was
immediately preceding the focused word than its neutral counterpart. This difference was found
to be significant for word 1 and word 3. The difference for word 2 did not come out significant.
The reason for this would be that, under the neutral condition, word 3 is the nuclear word and
word 2 is already preceding the nuclear word, and therefore maximum f0 on word 2 is already
boosted to mark sentence prominence under neutral condition. This boosting of f0 on the
immediately prenuclear high tone under neutral condition is exemplified using pitch tracks in
Figure 61 and Figure 62. Both figures exemplify the pitch contour of two SV intransitive
sentences preceded by a postpositional phrase (PP). SV intransitives can have two different
intonational patterns: (1) nuclear pitch accent can fall on the V, or (2) on the S
6
. In Figure 61,
nuclear pitch accent falls on the verb. When this is the case, f0 peak at the right edge of PP,
marked as H*H-, has almost the same f0 height as the preceding H* pitch accent. On the other
hand, in Figure 62, the nuclear pitch accent falls on the subject, therefore PP immediately
precedes the nuclear word. When this is the case, high tone on the final syllable of PP, marked as
H*Hn, reaches a higher level than the preceding H* pitch accent.
6
Details of shift in nuclear pitch accent in SV intransitives will be discussed in Chapter 3. For now, the reader can
just focus on immediately prenuclear f0 manipulation due to sentence prominence.
97
Figure 61. Sample pitch track of an SV intransitive sentence preceded by a PP.
“Babies sleep towards the evening”. Nuclear pitch accent is realized on the verb.
Figure 62. Sample pitch of a SV intransitive sentence preceded by a PP. “Babies
sleep towards evening”. Nuclear pitch accent is realized on the Subject.
NUCLEAR WORD
H*
H*H-
NUCLEAR WORD
H*
H*Hn
98
In this section, I have displayed pitch track examples and provided quantitative data from
a production experiment to show that tonal realization of sentence prominence in declaratives
involves a high tone realized on the final syllable of the immediately prenuclear word as well as
a nuclear pitch accent realized on the nuclear word itself. I used the diacritic “n” to distinguish
the prominence marking role of immediately prenuclear high tone in the intonational
phonological model of Turkish. When the immediately prenuclear High tone is realized on a
finally stressed word, I marked it by H*Hn because of the dual function of this High tone: (1) a
pitch accent, therefore “*”, (2) boundary tone marking the left edge of sentence prominence,
therefore “n”. On the other hand, when a non-finally stressed word immediately precedes the
nuclear word, high tone on the final syllable of the non-finally stressed word is marked by LHn
indicating its sole prominence marking function. Therefore Hn and LHn can be thought as the
allotones of the same underlying High tone that is triggered by sentence prominence.
The current study is not the first one to provide pitch track examples that display f0 boost
on the immediately prenuclear high tone. Similar observations have been made by Kan (2009)
and Kamalı (2011) as well but neither associated this immediately prenuclear high tone with
sentence prominence nor distinguished it in their inventory of tones. The following two pitch
track examples are taken from these two studies. Figure 63 is from Kamalı (2011) which displays
the pitch track of a four-word sentence where the third word, bunalanları, is the nuclear word.
As can be seen, there are two prenuclear High tones both of which are marked by H- boundary
tone. However, the second prenuclear High tone which immediately precedes the nuclear word
has a higher f0 peak than the first High tone. Kamalı (2011) says that “utterances such as [these]
are quite commonly found in the data, where the second prenuclear right edge is higher than that
of the first.” (p.68). According to Kamalı (2011), “The respective heights of the prenuclear peaks
indicate that a downstep process is not at work to lower the second peak in reference to the
previous. This could be interpreted in several ways. Downstep may be missing altogether, or
cannot operate across phrase boundaries. Alternatively, it could be that the relevant peaks are not
of the kind that induces or is sensitive to downstep.” In short, despite the consistent presence of
immediately prenuclear f0 boost, this High tone is not distinguished from other prenuclear High
tones in Kamalı’s model: Both of them are marked by a H- boundary tone.
99
Figure 63. Pitch track of a four word sentence from Kamalı (2011: 68). Third
word is the nuclear word and immediately prenuclear high tone is higher than
the initial high tone, however both prenuclear high tones are marked by H-.
This same fact can also be seen in the pitch tracks provided in Kan (2009). Figure 64
displays one such example where the typical tonal realization of a four word sentence, with the
third word being the nuclear word, produced in the neutral context is displayed. In this pitch
track example, second prenuclear f0 rise, the one that immediately precedes the nuclear word,
reaches a higher level than that of the first one. However, both prenuclear High tones are marked
as L+H*, therefore not distinguishing immediately prenuclear High tone from the other
prenuclear ones.
NUCLEAR
WORD
H-
H-
100
Figure 64. Pitch track of a four word sentence where the third word is the
nuclear word. High tone on the immediately prenuclear word is higher than that
of the first prenuclear high tone. from Kan (2009: 184)
Ayla twenty tangerine eat-PAST
‘Ayla ate twenty tangerines.’
In short, although f0 manipulation on the final syllable of the immediately prenuclear
word has been noticed and exemplified in the previous two intonational models, no direct
connection was made between this tone and its marking of sentence prominence; it is analyzed
the same way as other prenuclear high tones. In this subsection, I have shown that sentence
prominence causes systematic change on the pitch realized on the final syllable of the
immediately prenuclear word, therefore I distinguished the immediately prenuclear high tone in
the intonational phonology of Turkish and marked it by “n” diacritic. When the immediately
prenuclear word is non-finally stressed, prominence marking edge tone is marked by LHn, and
when it is finally stressed, it is marked by H*Hn tone.
A valid question at this point is how the left edge of a nuclear word is marked in
questions as the nuclear word in such sentences is realized by substantial pitch range expansion
(see Figures 36 and 37), unlike declaratives. Details of question intonation have not been
extensively analyzed, but the preliminary observations suggest that the left edge of the most
prominent word in questions is tonally realized by a Low tone unlike declaratives.. Pitch track in
Figure 65 displays this tonal realization. In Figure 65, the pitch contour on the immediately
L+H*
L+H*
101
prenuclar, non-finally stressed word, almanyaya, displays falling towards the end of the word
after the pitch accent. Details of what conditions f0 to fall in this context are not clear at this
point and requires further investigation but the bottomline here is that not only the tonal event on
the nuclear word is different in declaratives (compressed pitch range) versus questions (expanded
pitch range) but also the nature of prominence marking left edge tone is also different: In
declaratives it has quite expanded pitch range while in questions no such pitch range expansion
is observed. Instead, pitch contour on the immediately prenuclear word is falling in questions.
This, I mark by ‘Ln’ tone. I claim that the immediately prenuclear low tone is prominence
marking tone therefore ‘n’ diacritic’, rather than a regular phrase final tone, therefore not ‘-‘
diacritic, as it seems to emerge at the left edge of the most prominent word in questions, only.
Figure 65. Pitch track of a wh-question. Left edge of the wh-word is marked by
Ln tone.
Ln
102
Intonation phrase (IP) boundary tones (T%)
The largest tonally marked prosodic unit in Turkish prosodic structure is the intonation
phrase (IP), which corresponds to a full clause. Based on the corpus of data I have, an IP is
demarcated at its right edge by L% or H% tones. In the following subsection, I will provide pitch
tracks from various sentence types to exemplify IP boundary tones.
Low IP-boundary tone (L%)
Low IP boundary tone (L%) marks the right edge of declarative sentences, including
focus declaratives. In addition to yes/no questions, tag questions and imperatives are also
demarcated by L% boundary tone at their right edges. Figure 67 displays the tonal realization of
a five word declarative sentence produced in the neutral context. As can be seen in Figure 66,
right edge of the declarative is realized by a low tone which is marked by L% in the intonational
phonology of Turkish.
Figure 66. Sample pitch track of a five-word declarative sentence.
L%
103
Pitch contour in Figure 67 displays the tonal realization of the yes/no question
counterpart of the sentence given in Figure 66. A yes/no question in Turkish is formed by adding
the unstressable clitic –mI at the end of the phrase. In such sentences, the nuclear pitch accent is
realized on the stressed syllable of the word that the clitic –mI is attached to. After that, pitch
contour falls sharply to a low target which marks the right edge of the IP in yes/no question.
Figure 67. Pitch track of a yes/no question and a low IP-boundary tone (L%)
marks its right edge.
Another sentence type whose right edge is also marked by a low tone is tag questions.
Intonationally, I analyze tag questions as two separate Intonation Phrases; pre-tag and tag. This is
displayed in Figure 68. There are two reasons for this claim. First is that, each part of the
sentence; i.e., the pre-tag and tag, have a nuclear pitch accent. The pre-tag part has the tonal
contour of a regular declarative sentence with !H* marking the nuclear word and a high left-edge
boundary tone that marks the nuclear word (LHn). The tag part, howerver, displays the typical
contour of a yes/no question, i.e. a nuclear pitch accent with expanded pitch range followed by a
fall. So, in a tag construction, the pre-tag part behave like a regular declarative sentence and the
tag part like a yes/no question, each with its own NPA , !H* and ^H* respectively, followed by
L H* L H*H- L H* LH- L H* H*Hn L ^H* L%
lemaNIN ye!eN" alMANyaya araBAYla giD"yomu
leman’s nephew to germany by car going.Qpart
70
200
100
150
Pitch (Hz)
Time (s)
0 2.249
L%
104
L% tone. The second evidence is that there is usually a small pause between the pre-tag and the
tag parts, which is one of the signatures of an IP break.
Figure 68. Pitch track of a tag question, “Leman’s nephew is going to Germany
by car, isn’t he?”, and a low IP boundary tone (L%) marks its right edge.
Another sentence type that is marked by L% boundary tone is imperatives. Intonationally,
an imperative sentence behaves like a typical declarative sentence; rising pitch contour until the
nuclear pitch accented word, followed by a nuclear pitch accent which has compressed pitch
range, and L% at the end. This exemplified in Figure 69.
L%
PRE-TAG
TAG
105
Figure 69. Pitch track of an imperative sentence, “Call Alanya!”, and a low IP-
boundary tone (L%) marks its right edge.
High IP-boundary tone (H%)
The high IP boundary tone (H%) marks the right edge of wh-questions, echo questions,
and also the right edge of non-final coordinated clause. It is realized as a sharp rise on the
sentence-final syllable. In what follows, I will display pitch tracks to exemplify the realization of
H% for each sentence type.
One sentence type demarcated by high boundary tone, H%, is wh-questions. Turkish is a
wh-in situ language, i.e. no syntactic wh-movement is necessary to turn a declarative sentence
into a wh-question. Pitch track example in Figure 70 displays tonal realization of a 5-word wh-
question where the fourth word is the wh-word. The high tone marking the right edge of the
sentence can be clearly seen in this figure.
L H* LHn L !H* L%
aLANyayı aRAyın
alanya.acc call
70
300
100
150
200
250
Pitch (Hz)
Time (s)
1.396 2.319
L%
106
Figure 70. Pitch track of a wh-question and a high IP-boundary tone (H%).
A H% boundary tone also marks the right edge of a non-final coordinated clause in a
compound sentence. That is in a “C1 CONJ C2” construction where each “C” represent a
coordinate clause and “CONJ” stands for conjunction, the right edge of “C1” is demarcated by a
H%. This is exemplified in the pitch track in Figure 71 which displays the intonation of “C1
CONJ” portion of a “C1 CONJ C2” sentence.
H%
107
Figure 71. Pitch track of the first part of a coordinate clause, “Leman took
Kemal to the Restaurant and (went to her friend from there)”,. A high IP-
boundary tone (H%) marks its right edge.
Chapter summary
In this chapter, I have proposed a phonological model of Turkish intonation developed in
the AM framework of intonational phonology. I have proposed three tonally marked prosodic
units. These are (1) prosodic word which is marked at its left edge by a L tone, (2) intermediate
phrase (ip) which is demarcated at its right edge with a rising tone, (3) Intonation Phrase (IP)
which is marked either by a high or a low tone at its right edge.
With respect to pitch accents, I proposed three post-lexical pitch accents in the tonal
inventory: H*, !H*, and ^H*. A H* pitch accent is associated with prenuclear words that host a
post-lexical pitch accent. This is true for words that are finally-stressed as well as those which
are non-finally stressed. The other two pitch accents, !H* and ^H*, are proposed to mark nuclear
pitch accent in declaratives and questions, respectively. I have shown that, in a declarative
sentence, including focus declaratives, the nuclear word is realized by a tonal event that has
highly compressed pitch range. Therefore, I used !H* to mark the reduced pitch range on the
H%
108
nuclear word. Questions, on the other hand, displayed a complete opposite pattern with respect to
the tonal realization of the nuclear word. Unlike declaratives, informationally most important
word in questions is realized by expanded pitch range. Therefore, I proposed ^H* pitch accent to
mark sentence prominence, i.e., nuclear word, in questions.
The next prosodic level is the intermediate phrase. I have shown that the right edge of an
intermediate phrase is demarcated by a rising tone in declaratives. I proposed LH- and H*H- as
allotones to mark the right edge of an ip in declaratives. In addition, I have also proposed another
boundary tone type different from regular ip-final tone that demarcates the left edge the most
prominent word. I used the ‘n’ diacritic to distinguish prominence marking edge tone from
regular ip-final tone which is marked by ‘-‘ diacritic. In declarative sentences, the left edge of the
prominent word is marked by LHn if the immediately prenuclear word is non-finally stressed,
and by Hn if finally stressed (but since the syllable is also pitch accented, Hn is added right after
H*, i.e., H*Hn). In questions, left edge of the most prominent word is not marked by a rising
tone, but by a falling tone, demarcating the left edge of the nuclear word in questions. This is
labeled as ‘Ln’.
The highest prosodic level in the prosodic structure of Turkish is Intonation Phrase (IP)
which usually corresponds to a clause or a sentence. An IP is marked either by a low tone (L%)
or a high tone (H%) depending on the sentence type.
A full list of tonal inventory and their associations are provided in Table 3.
Table 4. List of pitch accents and boundary tones proposed in the current model and their
respective associations.
Association Target
Pitch accents H*, !H*, ^H*
Prosodic word-initial boundary tone L
ip-final boundary tones H-, LH-
IP-final boundary tones L%, H%
109
Chapter 3: Phonetics of Turkish intonation
3.1 General introduction
This chapter provides quantitative data to support the intonational model of Turkish
proposed in Chapter 2, in particular on prenuclear tones host by finally stressed words. As is well
known by now that regular stress pattern of Turkish assigns lexical stress on the word final
syllable. In the current intonational model, I distinguished three different types of prenuclear
high tones host by finally stressed words. These are prenuclear pitch accent; H*, high tone
marking the right edge of an intermediate phrase; H*H-, and high tone marking the left edge of
sentence prominence; H*Hn. The primary question addressed in this chapter is whether this
three-way tonal distinction made at the phonological level is reflected at the phonetic level, i.e.,
whether these tones are phonetically realized differently. This question is addressed in two
production experiments presented in section 3.2. and section 3.3. The first production experiment
is about the phonetic realization of H* versus H*H-. In the current model, I proposed H* as the
underlying prenuclear pitch accent. Therefore, when a finally stressed word is ip-phrase medial
and carries a high tone on its stressed syllable, this high tone is marked by H* pitch accent.
However, as majority of the words are lexically stressed on their final syllables, it is quite
common to find a finally stressed word as the last prosodic word in an intermediate phrase. This
fact raises the question of how to analyze a high tone on a finally stressed word when it is the last
word in an intermediate phrase because in this case this high tone has a dual function: (1) to
mark lexically stressed syllable, (2) to mark the right edge of an intermediate phrase. Two
previous intonational models, Kan (2009) and Kamalı (2011), approached this issue differently:
The former assumed a pitch-accent analysis for both phrase-final and phrase-medial high tones
on finally stressed words; i.e., according to this model, a high tone on a finally stressed word
within an IP was analyzed as a pitch-accent regardless of its phrase-medial or phrase-final status.
In Kamalı’s model, on the other hand, high tone on finally stressed words was analyzed as a
phrase final tone, H-. Unlike these two intonational models, I distinguished two different high
tones on a finally stressed word depending on its location in the prosodic structure: When a high
tone is realized on a phrase-medial finally stressed word, it is marked by a H* pitch accent which
110
reflects its phrase-medial status. On the other hand, when a finally stressed word is the last word
in an intermediate phrase, the word final high tone host by this prosodic word is marked by H*H-
tone which indicates that this high tone, in addition to marking lexical stress therefore “*”
diacritic, marks a bigger prosodic grouping; namely an intermediate phrase. The purpose of
Experiment 1 is to provide quantitative data from a production experiment to support this
phonological distinction of two high tones, H* and H*H-, proposed in the current model.
The second production experiment is about the phonetic realization of prominence
marking high tone on finally stressed words, i.e., H*Hn. In this current intonational model, I
made a novel proposal and suggested that a nuclear word in Turkish is marked by two tonal
events: a pitch accent on the nuclear word itself and a high tone realized on the final syllable of
the immediately prenuclear word which marks the left-edge of the nuclear word. Following this,
when the immediately prenuclear word is lexically stressed on its final syllable, a high tone
realized on this syllable again has two functions: (1) to mark lexical stress, (2) to mark the left
edge of the nuclear word. The question at this point is whether this prominence marking edge
tone, H*Hn, is phonetically different from the other high tones distinguished on finally stressed
words, H* and H*H-. The second production experiment addresses this question and aims to
provide quantitative data to support this tonal distinction in the current intonational model.
The third study presented in this chapter is a perception experiment. The aim of this
experiment is to provide further support for the novel proposal that marking sentence
prominence in Turkish involves a second tonal event, a left-edge tone, in addition to nuclear
pitch accent. To this end, I tested listeners’ sensitivity to prominence marking tones, H*Hn and
!H*, proposed in the current intonational model. The main question addressed in the perception
experiment is: “Do both prominence marking tones contribute to listeners’ perception of
sentence prominence?”
3.2 Experiment 1: Distinguishing H* and H*H-
The aim of this section is to investigate the nature of high tones on finally stressed words
that are ip phrase-medial and phrase-final. Two previous intonational models developed in the
AM framework, Kan (2009) and Kamalı (2011), analyzed high tone on finally-stressed words
differently. In Kan’s model, this tonal event was marked as a pitch accent, while Kamalı (2011)
111
analyzed them as phrase tones (H-). That is, according to Kan’s model, a high tone on a finally
stressed word within an IP is always a pitch-accent, while it is always a phrase tone according to
the model developed by Kamalı (2011). However, this uniform analysis of high tone, i.e., either
as a pitch accent or a phrase tone, does not distinguish between a high pitch realized on the final
syllable of a phrase-medial versus a phrase-final word. Therefore, in these model, the prosodic
structure of an utterance is not reflected in the hierarchical organization of the uttereance’s
intonational pattern.
In this respect, consider the following two utterances and their corresponding tonal and
prosodic analysis taken from Kan (2009). In Figure 72, the first PhP is comprised of one
prosodic word, ayla, which is lexically stressed on its final syllable. The rising tone on the final
syllable of this word which marks both lexical stress and the right edge of a PPh is marked by
bitonal L+H* pitch accent. This being the case, while the tonal contour reflects a smaller
prosodic unit; namely a prosodic word, the prosodic structure of the same sentence indicates that,
ayla, forms a bigger prosodic unit; a PPh. Therefore, the structure reflected in the tonal contour
of the sentence does not match with the hierarchical organization of its prosodic structure. In
Figure 73 on the other hand, the first PhP has two prosodic words. The first word within this
PPh, yaralı, is also lexically stressed on its final syllable, and the rising tone on its final syllable
is marked by bitonal pitch accent, L+H*. Tonal analysis in these two figures display that no
matter whether a prosodic word is phrase medial or phrase final, a high tone on its stressed
syllable is marked by a pitch accent at the tone tier, and this lack of distinction between phrase
medial and phrase final high tone at the tone tier results in a mismatch between the prosodic
structure of the sentence and the structure reflected in its tonal analysis.
112
Figure 72. Tonal analysis of a three word declarative sentence. The first word,
Ayla, is lexically stressed on its final syllable, and forms one PPh. Rising contour
on the first word is marked by bitonal L+H* which also marks the right edge of
PPh. Taken from Kan (2009:155)
L+H* !H* L- L%
[[Ayla]
PPh
[Almanya-yı özle-miş]
PPh
]
IP
Ayla Germany.ACC miss.EVıD
“Ayla missed Germany”
L+H*
113
Figure 73. Tonal analysis of a four word declarative sentence The first word,
yaralı, is lexically stressed on its final syllable and is phrase medial within a
PPh (yaralı yalovalılar). The rising tone on this PPh phrase medial tone is
marked by bitonal pitch accent L+H*. From Kan (2009:176)
L+H* L+H- !H* L- L%
[[Yaralı Yalovalı-lar]
PPh
[anı-lar-ın-ı yaz-mış]
PPh
]
IP
injured Yalovalı-PLU memoir-PLU-3SG.POSS-ACC write-EVıD
‘The people of Yalova who are injured have written their memoirs.’
Kamalı(2011), on the other hand, assumes that finally-stressed words in Turkish are
lexically not stressed. In this model, a high tone on a finally stressed word is analyzed as a phrase
tone (H-) of some Accentual/Major/Phonological Phrase. Since the prosodic phrases analysed in
this study are confined to one word only, it is not clear whether the H- is boundary tone of a
prosodic word or a major phrase. In addition, it is also not clear how a phrase medial high tone
on a finally stressed word would be analyzed with this model. Following pitch contour in Figure
74 displays one such analysis from Kamalı (2011).
L+H*
114
Figure 74. Tonal analysis of finally stressed words according to Kamalı (2011).
In the current intonational model as described in chapter 2, I distinguished phrase-medial
high tone on finally stressed words from a phrase-final one. A phrase-medial prenuclear high
tone is marked by a pitch accent, H*, and the phrase final one with a H*H-. This distinction in
the tonal tier also reflects the hierarchical prosodic structure of a sentence. That is, when high
tone on a word final syllable is marked by a H*H- tone, this indicates that H*H- demarcates the
right edge of an intermediate phrase. On the other hand, when it is marked by H*, this means that
the high tone is marking post-lexical prominence and that this prosodic word is phrase-medial.
This distinction between H* and H*H- on finally stressed words is exemplified in the tonal
analysis of a five-word sentence displayed in Figure 75. First two words in this sentence,
lemanın and oğlu, form one intermediate phrase. Both words are lexically stressed on their final
syllables and both host a high tone realized on their word final stressed syllable. According to the
current model, first high tone realized on the phrase medial finally stressed word, lemanın, is
marked by a H* pitch accent which also indicates its phrase-medial status, while the high tone on
phrase final word, oğlu, is marked by H*H- therefore indicating that this high tone in addition to
marking lexically stressed syllable also marks the right edge of a higher prosodic boundary;
namely an intermediate phrase.
H- H-
115
Figure 75. Distinction of phrase medial high tone vs phrase final high tone on
finally stressed words according to current model.
The aim of this chapter is to provide quantitative data to support this distinction between
phrase medial and phrase final high tones on finally stressed words within an IP. Previous studies
have shown that the strength of a prosodic boundary is reflected in the acoustic realization of
segments at the edges of a prosodic unit. For instance, the degree of final lengthening is stronger
at the end of stronger prosodic boundaries (e.g. Wightman et al. 1992, Jun & Fougeron 2000,
Tabain 2003a, Tabain and Perrier 2005), and the height of f0 peak is higher at stronger prosodic
boundaries (e.g. Beckman & Pierrehumbert 1986, Khan 2008, Jun 2010). In addition, the degree
of V-to-V coarticulation across stronger prosodic boundaries is weaker than that across weaker
prosodic boundaries (e.g. Cho 2005). Thefore, it is predicted that the final syllable of ip-final
word, carrying H*H-, will have a higher f0 peak and longer duration than when it is the last
syllable of a prosodic word, carrying a H* tone. It is also predicted that the V-to-V coarticulation
across the H*H- syllable will be weaker than that across the H* syllable. In this section, I will
provide the results of phonetic experiments confirming these predictions.
H* H*H-
116
3.2.1 Method
Materials
Materials were designed to compare H* and H*H- tones paradigmatically. To avoid any
effect of sentence length on the degree of f0 declination and the duration of syllables, five near
minimal pairs of five word sentences where the first two words and the initial syllable of the
third word are exactly the same. In each pair, the sentences differed in the size of the subject NP:
One (Group 1) contained three words and the other (Group 2) two words. In chapter 2, I have
shown that the last syllable of an NP is systematically realized with a rising tone (H*H-) marking
the right edge of an Intermediate Phrase. Therefore, in this stimuli, I manipulated the syntactic
grouping so that in one sentence, the second word is in the middle of a subject NP, carrying H*
(i.e., Group 1 in (22)) while in the other, the same second word is at the end of a subject NP,
carrying H*H- (i.e., Group 2 in (22).
22.
Group 1: {word1 word2
H*
word3}
ip
word4 word5.
a. {Yoldaki tuzak
H*
işaretler} önümüzü kapatıyor.
Road.LOC.ADJ trap sign.PLU front.POSS.ACC block.PROG
“The trap signs on the road are blocking our sight.”
Group 2: {word1 word2}
H*H-
ip
word3 word4 word5.
b. {Yoldaki tuzak}
H*H-
işareti tamamıyla kapatıyor.
Road.LOC.ADJ trap sign.ACC completely block.PROG
“The trap on the road is completely blocking the sign.”
According to Kan (2009), word2 in each Group receives a pitch accent (L+H*), but
according to Kamalı (2011), word2 in Group 2 receives a H- boundary tone marking a Major
Phrase. (The H tone of word2 in Group 1 is not analyzed in Kamalı (2011)). In the current
model, the high tone realized on the final syllable of word2 in Group 1 is in the middle of an ip
therefore, it would be marked with H*, but the final syllable of the same word in Group 2 is the
last syllable of an ip (corresponding to a right edge of NP), and so would be marked with H*H-.
117
Participants
Five native speakers of Turkish (three females and two males) participated in this
experiment. They were in their twenties or thirties. All speakers were born and raised in Turkey,
and have been living in Los Angeles for less than five years. The sentences were
pseudorandomized with fifteen filler sentences so that sentences in each pair were not ordered
next to each other. Each sentence was displayed on a computer screen, and the speakers read the
entire list 3 times. The recording was made using Praat at 16 bit sampling rate.
Measurements
Three types of acoustic measurements taken in each sentence and their criteria are as
follows.
i. Peak f0, Minimum f0, and Magnitude of f0-rise: The f0-peak was measured ten ms before the
end of the vowel in the final syllable of the target word to avoid any effect of microprosody, and
the minimum f0 value was measured on the preceding syllable. In order to normalize f0 values
across speakers, f0 values were converted into semitones and the magnitude of f0 rise was
calculated by the difference between the f0 peak and the minimum f0 values.
ii. Rhyme duration: The rhyme duration of the final syllable of word2 is measured. Since word3
always begins with a vowel, the rhyme duration of word2’s final syllable was measured as the
time between the vowel onset (V
1
) in Word2’s final syllable to the vowel onset (V
2
) of word3. A
vowel onset is defined as the point where the second formant energy begins on the spectrogram.
iii. V-to-V coarticulation: The degree of coarticulation between two vowels is quantified by
calculating the Euclidean distance in the F1 by F2 space between each vowel in V
1
C#V
2
context.
In all the sentences, V
1
= /a/, C = /k/, V
2
= /i/. These segments are chosen because the vowels
differ maximally in both F1 and F2 dimensions and the velar consonant is known to vary in its
place of articulation depending on the vowel context, thus less interfering with the tongue
gesture of the adjacent vowels (Stevens 1989).
118
3.2.2 Results
The statistical analysis was carried out in R using Linear Mixed-Effects Regression
(LMER) package ‘lme4’ (Baayeen et al. 2008). TONE TYPE (H* vs H*H-) was used as the fixed
factor, and SUBJECT and SENTENCE as random factors. The simplest and best fitting model for
each LMER analysis was derived via model comparisons.
In order to obtain p-values for the fixed effects, I used pvals.func in ‘languageR’ package
that computes the relevant p-values using Markov chain Monte Carlo sampling (default number
of samples = 10,000). When the best fitting model involved random slopes, significance of fixed-
effect was computed using likelihood-ratio test.
Peak f0, minimum f0, and magnitude of f0-rise
As predicted, the peak f0 values were higher at the Intermediate Phrase boundary (H*H-)
than at the Prosodic Word boundary (H*), but the difference did not come out significant
(Intercept= 13.25, β=0.73, SE= 0.4, t=1.84, p=0.07). However, I found that the minimum f0
value was significantly lower before the H*- syllable than before the H* syllable (Intercept=
12.4, β= -0.71, SE= 0.24, t= -2.95, p<0.05). Therefore, I compared the magnitude of f0 rise for
H* and H*H-, and found that magnitude of word final high tone rise at an Intermediate phrase
boundary was significantly larger than that of a Prosodic Word boundary (Intercept= 2.29,
β=1.435, SE=0.24, t=5.925, p<0.05). Figure 76 shows the average f0 rise in semitones in each
tone type condition. As shown in the figure, magnitude of f0 rise at the Intermediate Phrase
boundary (H*H-) is greater than that at the prosodic word boundary (H*).
119
Figure 76. Magnitude of mean f0 rise at Prosodic Word and Intermediate Phrase
boundaries.
Example pitch tracks of the sentences (22a) and (22b) are shown in Figure 77 and 78. In
Figure 77, the second word (tuzak) is phrase-medial, thus the f0 peak at the end of the second
word receives a pitch accent H*, while the f0 peak of the same word receives H*H- in Figure 78
because it marks the end of an Intermediate Phrase.
0
0.5
1
1.5
2
2.5
3
3.5
4
H* H*H-
Mean f0 rise (semitone)
Tone Type
120
Figure 77. Sample pitch track of the sentence (22a) illustrating H* at the end of
2
nd
word (tuzak), i.e., Prosodic Word boundary.
Figure 78. Sample pitch track of the sentence (22b) illustrating H*H- at the end
of 2
nd
word (tuzak), i.e., Intermediate Phrase boundary.
MAGNITUDE OF F0 RISE = 47.5 HZ
MAGNITUDE OF F0 RISE = 70.4
HZ
121
Rhyme duration
As in the magnitude of f0 rise, a significant effect of tone type was found on the temporal
properties of segments. As shown in Figure 79, rhyme duration is significantly shorter at the
prosodic word boundary (H*) compared to the Intermediate Phrase boundary (H*H-) (Intercept=
0. 12, β=0.067, SE=0.009, t=8.38, p < 0.05).
Figure 79. Rhyme duration (ms) of PW-final syllable (H*) vs. ip-final syllable
(H*H-)
Degree of coarticulation
Figure 80 displays the mean Euclidean distance in the F1 by F2 space between each
vowel in V
1
C#V
2
context. As hypothesized, there is less coarticulation (i.e., larger Euclidean
distance) across the Intermediate Phrase boundary than across the Prosodic Word boundary
(Intercept= 255, β=231.1, SE= 107.7, t=2.145, p<0.05).
0
0.05
0.1
0.15
0.2
0.25
H* H*H-
Duration (ms)
Prosodic Boundary
122
Figure 80. Euclidean distance (Hz) between vowel across a PW boundary (H*)
vs. an ip boundary (H*H-).
3.2.3 Experiment 1 discussion
In this experiment, I have shown that the phonetic realization of high tone on finally-
stressed words is different depending on its prosodic location. The magnitude of f0 rise is larger
at the end of a phrase (Intermediate Phrase/Phonological Phrase/Major Phrase) than in the
middle of a phrase. The syllable carrying the high tone is longer when it is the last syllable of an
Intermediate Phrase than of a Prosodic Word. Finally, the degree of V-to-V coarticulation is
weaker when the two vowels are separated by an Intermediate Phrase boundary than by a
Prosodic Word boundary.
The quantitative data found in the experiment therefore support the prosodic structure and
tonal categories proposed in the current model of Turkish intonation. In this current model, the
High tone on a stressed syllable of a prosodic word is labeled as H*, a pitch accent; the High
tone realized on the stressed syllable at the end of an ip, however, is labeled as H*H-, indicating
its function, as both a pitch accent and a boundary tone.
In this experiment, I only examined finally-stressed words, so an ip-final syllable is
always stressed, justifying H*H- symbol. However, as shown in chapter 2, the ip-final syllable
(or Phonological Phrase/Major Phrase-final syllable in Kan (2009) and Kamalı (2011)) was still
marked by a high tone when the syllable was not stressed. For example, consider Figure 81
0
100
200
300
400
500
600
700
800
H* H*H-
Euclidean Distance (Hz)
Tone Type
123
where the first two words are non-finally stressed and form a subject NP. Here, each word in the
subject NP has a High tone on its stressed syllable (H*) and the second word has an additional
High tone on its final syllable, which is also the last syllable of an Intermediate Phrase. Since
there is a low f0 target between the f0 peaks (one over the stressed syllable of the word and the
other at the end of the word), and the L tone is closer to the following H tone than the preceding
H tone (H*), I labeled ip-final f0 rise over an unstressed syllable as LH- (Note also the shallow
falling slope from H* in the first word, interpolating to the initial L boundary tone of the second
word in Figure 81. Thus, it is clear that the end of an ip is marked by a high boundary tone
regardless of whether the ip-final stressed syllable is labeled as H*H-, indicating both
prominence marking and the boundary marking functions; the High rising tone on the ip-final
unstressed syllable, however, is labeled as LH-, indicating its sole, boundary marking function.
Figure 81. Pitch track of the sentence “Restaurant’s chairs will be painted in
blue on Friday.” The second word (sandalyeleri) is ip-final and its stressed on
its second syllable. The second f0 rise on this word, labeled LH-, marks the right
edge of ip.
LH-
124
3.3 Experiment 2: Distinguishing H*H- and H*Hn
In the current intonational model, I distinguished a third type of high tone on finally
stressed words within an IP, H*Hn, and claimed that the function of this tone is to mark the left
edge of the nuclear word. In Experiment 1, I showed that two word final high tones, H* and
H*H-, are phonetically different justifying the tonal distinction made in the current intonational
model. The next question at this point is: “Is the third word final high tone, H*Hn, phonetically
realized differently from the other two word final high tones, H* and H*H-? The answer is “yes”
and the aim of this subsection is to provide quantitative data to show that H*Hn is phonetically
different from H* and H*H-. In order to address this question, I compared the immediately pre-
nuclear high tone, H*Hn, with a regular ip-final high tone, H*H- (i.e., H*H- syllable that is not
adjacent to the nuclear word). H*Hn was compared with H*H- instead of H* because informal
observation suggests that H*Hn is higher in f0 peak than H*H-. As shown in Experiment 1,
H*H- is higher in f0 peak than H* (although the results did not come out significant), and
magnitude of rise for H*H- was larger. If we find that f0 peak is higher and/or the magnitude of
f0 rise is larger for H*Hn than for H*H-, this would automatically suggest that H*Hn is different
from H* as well as H*H-.
3.3.1 Method
Materials
Two target tones, H*Hn and H*H-, are compared paradigmatically. To investigate
whether H*Hn is different from H*H-, I examined the tonal pattern of SV intransitive sentences.
SV intransitives are known to have two intonational patterns: nuclear stress falls on the subject
when the meaning of the verb is eventive but on the verb when the verb has generic meaning
(Sasse 1987, Zubizarreta & Vergnaud 2005, Kratzer & Selkirk 2007, Féry 2011).
Previous studies on Turkish distinguished two tonal patterns for SV intransitives as well.
However, the source of this distinction was not well disentangled. Üntak-Tarhan (2006) made a
clear-cut distinction and claimed that the difference in the intonation pattern of SV intransitives
was a result of whether the verb was unergative or unaccusative: Sentential stress fell on the
125
subject if the verb was unaccusative, and on the verb if it was unergative. This proposal by
Üntak-Tarhan (2006) was an impressionistic one. Kan (2009) put this impressionistic claim into
test and found that unergative/unaccusative distinction was not the source of different SV
intonation patterns: She found variation in both types of intransitives. Roots of this variability
was not investigated further, and is left to future inquiry (Kan 2009: 174).
Following previous findings on the source of two intonation pattern of SV intransitives;
i.e., location of sentence stress changed based on the generic versus eventive nature of the verb, I
did a pilot study to test whether this distinction would also explain the variability in Turkish SV
intonation patterns. Three Turkish native speakers produced six SV intrasitive sentences (three
with eventive reading, three with generic reading) five times, intermixed with fifteen filler
sentences. Speakers produced 3 sentences × 2 conditions × 3 speakers × 5 repetitions = 90
tokens: 45 with eventive reading, 45 with generic reading. Realization of sentence stress based
on the eventive versus generic nature of the verb is given in Table 5.
Table 5. Realization of sentence stress in SV intransitives based on eventive
versus generic nature of the sentence.
SV SV
Eventive 44 1
Generic 1 44
Distribution in Table 5 clearly show that speakers make a distinction on where they put
sentential stress in an intransitive sentence depending on whether the verb is generic versus
eventive. This finding, I used, in the current experiment, to manipulate the location of sentential
stress, therefore nuclear pitch accent: Nuclear accent will fall on the subject when the verb is
eventive, and on the verb when it is generic. This type of sentences will, therefore, provide a
clear data regarding the location of nuclear accent.
The experimental protocol consisted of sentence structure, “Postpositional Phrase +
Subject + Verb” (henceforth PP + S + V), five of which were eventive (PP-S-V) and five of
which were generic (PP-S-V) (the underlined part is where the nuclear accent falls).
According to the current intonation model, the last syllable of PP would be marked with
H*H-, but it would also be marked with H*Hn when the following “S” receives nuclear accent
126
(!H*), as in (23a). But the last syllable of PP would only be marked with H*H- when the “V”
receives a nuclear accent, as in (23b). An example sentence of (23a) and (23b) is shown in (24a)
and (24b), respectively. It is possible that both H tones are realized the same because the H*Hn
in (23a) is also marking the end of PP, thus H*H-.
(23)
(24) a. Sabaha karşı babamlar geldi. PP
H*Hn
S
!H*
V
Morning.dat towards dad.POSS come.PAST
“Towards the morning, my dad came.”
b. Sabaha karşı babalar horlar. PP
H*H-
S
H*Hn
V
!H*
Morning towards dad.PLU snore.AOR
“Towards the morning, dads snore.”
Subjects, procedures, and statistics
Five native speakers of Turkish (three females and two males) participated in this
experiment. They were in their twenties or thirties. All speakers were born and raised in Turkey,
and have been living in Los Angeles for less than five years. The sentences were
pseudorandomized with fifteen filler sentences so that sentences in each pair are not ordered next
to each other. Each sentence was displayed on a computer screen, and the speakers read the
whole the entire list 3 times. The recording was made using Praat at 16 bit sampling rate.
The statistical analysis was carried out in R using Linear Mixed-Effects Regression
(LMER) package ‘lme4’ (Baayeen et al. 2008). TONE TYPE (H* vs H*Hn) was used as the fixed
factor, and SUBJECT and SENTENCE as random factors. The simplest and best fitting model for
each LMER analysis was derived via model comparisons.
a. Eventive PP
H*Hn
S
!H*
V
b. Generic PP
H*H-
S
H*Hn
V
!H*
127
Measurements
Two types of measurements taken for each sentence and their criteria are as follows:
(1) Peak f0, Minimum f0, Magnitude of f0 rise: The f0-peak was measured ten ms before the end
PP to avoid any effect of microprosody, and the minimum f0 value was measured on the
preceding syllable. In order to normalize f0 values across speakers, f0 values were converted into
semitones and the magnitude of f0 rise was calculated by the difference between the f0 peak and
the minimum f0 values.
(2) Rhyme Duration: The rhyme duration of the final syllable of PP-final word was measured.
3.3.2 Results
Peak f0, minimum f0, and magnitude of f0-rise
As expected based on the preliminary observation from pitch track examples, I found that
peak f0 for H*Hn was significantly higher than that of H*H- (Intercept= 12.11, β=0.57,
SE=0.16, t=3.75, p<0.05). Minimum f0 was not significantly different (Intercept= 8.97, β=-
0.059, SE=0.35, t=-0.17, p=0.8), and naturally magnitude of f0 rise was larger for H*Hn
(Intercept= 3.75, β= 0.62, SE=0.31, t=1.99, p<0.05). Figure 82 shows the average f0 rise in
semitones in each tone type condition. As shown in the figure, magnitude of f0 rise for
prominence marking high tone (H*Hn) is greater than that of intermediate phrase marking one
(H*H-).
128
Figure 82. Comparison of magnitude of f0 rise in the H*H- vs H*Hn contexts.
A typical example is shown in Figures 84 and 85. In Figure 83, the verb, uyur, has a
generic meaning and receives !H* nuclear accent. The right edge of PP, akşama doğru, is
marked by a regular ip-final high tone, H*H-. The magnitude of f0 rise for H*H- is 63.28Hz. In
Figure 84, however, the subject ‘fish’ is nuclear accented and the right edge of PP is marked by
H*Hn. Magnitude of rise in this condition is 75.33Hz f0 rise.
0
1
2
3
4
5
H*H- H*Hn
Mean f0 rise (semitone)
Tone Type
129
Figure 83. Pitch track of the sentence “Towards the evening, babies sleep”. The
sentence has a generic reading of the verb, uyur, and the verb is nuclear accented.
Figure 84. Pitch track of the sentence “Towards the evening, fish burnt”. The
sentence has an eventive reading of the verb, and the subject, balıklar, is nuclear
accented.
MAGNITUDE OF F0 RISE=56.31HZ
MAGNITUDE OF F0 RISE=75.33HZ
130
Duration
Figure 85 shows mean rhyme duration of the syllable that carries H*H- vs. H*Hn.
Although the tendency is that segments are lengthened more when it carries a H*H- tone than a
H*Hn tone, the difference was not significant (Intercept=0.09, β=-0.003, SE= 0.01, t=-0.25, p >
0.05).
Figure 85. Rhyme duration in the H*H- versus H*Hn contexts
3.3.3 Experiment 2 discussion
In this experiment, phonetic realization of H*Hn and H*H- tones is compared. The
results show that the immediately pre-nuclear high tone, H*Hn, is realized with higher f0 peak
compared to H*H- peak.
The difference found between two f0 peaks in this experiment support the distinction
between H*Hn and H*H- proposed in the current phonological model. According to the current
model, H*n is realized on the final syllable of the immediately prenuclear word marking the left
edge of the nuclear word, while H*H- marks the right edge of an intermediate phrase. Also, since
the comparison between H* and H*H- in Experiment 1 showed that H*H- was phonetically
realized with greater magnitude of f0 rise than H* pitch accent, result of this current experiment
also shows that H*Hn is different from H*.
0.05
0.06
0.07
0.08
0.09
0.1
0.11
H*H- H*Hn
Mean Rhyme Duration (ms)
Tone Type
131
The primary finding of this experiment, i.e., high tone at the left edge of a nuclear word is
phonetically realized different from a regular phrase final high tone, confirms my claim that
sentence prominence in declaratives is marked by two tonal events in Turkish; one is the nuclear
pitch accent realized on the nuclear word itself which is phonetically realized by quite
compressed pitch range, and the second is the high tone realized on the final syllable of the
immediately prenuclear word marking the left edge of sentence prominence. This suggests that
encoding sentence prominence is not necessarily confined to a single tonal event; namely a
nuclear pitch accent. This present study have shown that a tone realized on a prosodic word other
than the nuclear word itself is also relevant in the marking of sentence prominence.
The question at this point is: “Do both of these prominence prominence tones, H*Hn and
!H*, contribute to the perception of the informationally most prominent word in the sentence. I
will present data from a perception study in the next subsection to address this question.
3.4 Experiment 3: A perception experiment
This subsection is about perception of sentence prominence in focus-neutral contexts. In
particular, I will test the contribution of two prominence marking High tones in the correct
identification of the informationally most prominent word in a sentence. In the current
intonational model described in Chapter 2, I proposed that sentence prominence in declarative
sentences is marked by two tonal events, an edge tone realized on the final syllable of the
immediately prenuclear word and a nuclear pitch accent realized on the stressed syllable of the
most prominent itself, which is realized by quite compressed pitch range. In Section 3.2., I
provided quantitative data from a production experiment to show that the immediately prenuclear
High tone was phonetically different from a regular intermediate phrase-final High tone
supporting the tonal distinction I made in the current phonological intonational model.
The question addressed in Experiment 3 is whether listeners use both of these tonal cues;
i.e. the nuclear pitch accent (!H*) and the prominence marking edge tone (H*Hn) in the correct
identification of the most prominent word in a sentence.
132
3.4.1 Method
Materials
Materials were designed to serve as auditory input for a prominence judgement task.
Eight target and sixteen filler sentences were recorded by a female Turkish speaker (not the
author) to present to listeners. Target sentences had the structure Post positional phrase + Subject
+ Verb (PP S V) with an eventive reading. As mentioned in section 3.2, in SV intransitive
sentences with an eventive reading, nuclear accent falls on the subject. Therefore in this
structure, the nuclear pitch accent, !H*, was realized on the subject, and the right edge of PP was
marked by the prominence marking left-edge tone, H*Hn. Sample pitch contour and its tonal
analysis is presented in Figure 86.
Figure 86. Pitch contour of a PP + S + V sentence with an eventive reading, and its
tonal analysis according to the current model.
H*
H*Hn
!H*
NUCLEAR WORD
133
All four words in all target items were stressed on their final syllables, and each preverbal
word carried a high tone on its stressed syllable as displayed in Figure 86; H* on the phrase-
medial word, H*Hn on the immediately prenuclear word and !H* on the nuclear word. In all
target sentences H*Hn had the highest f0 peak, H* the second and !H* had the lowest peak.
Each original pitch contour of a target sentence was then manipulated using PSOLA
implemented in Praat (ver. 5.4.06 (Boersma and Weenink 2015)) to modify two prominence
marking tones H*Hn and !H*. Each original pitch contour was modified twice: (1) to lower H*n,
(2) to delete nuclear pitch accent, !H*. As a result each target item appeared in three different
conditions: The first condition was the original condition where nothing in the pitch contour was
manipulated. I will refer to this condition as [+H*Hn +!H*]. The second condition was where
H*Hn is manipulated so that it had lower peak than the preceding H*, as a result it was not the
highest peak in the sentence, but still higher than the nuclear pitch accent. I will refer to this
condition as [-H*Hn +!H*]. Figure 87 displays manipulated pitch contour of the target sentence
shown in Figure 84.
Figure 87. Pitch track of a target sentence where f0 peak of H*Hn in the original
sentence is lowered. It is lower than the first f0 peak, H*, but still higher than the
nuclear pitch accent, !H*.
The third condition was where the pitch accent on the nuclear word was deleted. Therefore, pitch
contour displayed a falling slope following H*Hn instead of a level contour that spread over the
H*
- H*Hn
!H*
134
nuclear word. Figure 88 displays a sample contour as a result of this manipulation (The original
pitch contour is the one displayed in Figure 84).
Figure 88. Pitch track of a target sentence where the nuclear pitch accent !H* is
‘deleted’.
Peak f0 values for each high tone under three conditions are given in Table 6.
Table 6. Peak f0 values for each high tone under three conditions.
Tones
Sentence
H* H*Hn !H* -H*Hn -!H*
1. 243.4 269.1 193.4 215.6 172.3
2. 271.5 287.3 219.9 244.6 194.6
3. 221.5 248.5 185.4 205.1 163.7
4. 266.4 280.5 214.7 243.4 193.2
5. 249.5 278.8 204.8 227.2 182.8
6. 261.6 295.1 206.8 229.7 176.9
7. 255.6 272.8 202.9 227.8 181.1
8. 253.1 283.96 204.2 226.6 173.1
H*
H*Hn
- !H*
135
Target items (8 sentences × 3 conditions) were grouped into three sets and each set
contained only one condition, [+H*Hn +!H*], [-H*Hn, +!H*], [+H*Hn, -!H*], of each target
item. Table 7 displays the categorization of each sentence per condition. This categorization was
done so that a participant did not hear the same target sentence in more than one condition in one
sitting.
Table 7. Condition of each target sentence in each set.
SET 1 SET 2 SET 3
1. +Hn, +!H* -Hn, +!H* +Hn, -!H*
2. +Hn, +!H* +Hn, -!H* -Hn, +!H*
3. +Hn, +!H* +Hn, -!H* -Hn, +!H*
4. -Hn, +!H* +Hn, +!H* +Hn, -!H*
5. -Hn, +!H* +Hn, +!H* +Hn, -!H*
6. -Hn, +!H* +Hn, +!H* +Hn, -!H*
7. +Hn, -!H* -Hn, +!H* +Hn, +!H*
8. +Hn, -!H* -Hn, +!H* +Hn, +!H*
There were twenty-four sentences in each set, eight pseudorandomized target sentences
intermixed with sixteen non-experimental filler sentences. The filler sentences were recorded by
the same speaker who produced the target items. The fillers were created in a way that the
nuclear accent never fell on the immediately preverbal word. Eight of the fillers were produced
under narrow focus condition where a word other than the immediately preverbal one was
narrowly focused. Rest of the eight fillers were produced under neutral condition. Four of these
sentences had the structure S + SHORT ADV + BARE OBJ + V. In such structures in Turkish, nuclear
pitch accent falls on the short adverb (Untak-Tarhan 2006). The remaining four sentences had
the same structure as the target sentence, PP + S + V, but with a generic reading, so the nuclear
accent fell on the V.
136
Participants
Twelve native speakers of Turkish participated in the experiment. All of them were
graduate students at USC or UCLA, and none of them had any background or training in
linguistics. All participants confirmed they had no previous diagnosis of a hearing problems.
Procedure
Data collection procedure was divided into three parts, and the sentences in three sets
(Table 6) were presented to each participant one week apart, i.e., in the first week, a participant
heard sentences from Set 1 only, the second week from Set 2, and the third week from Set 3.
This way, a participant heard each target item under one of the three conditions only in each
week.
Powerpoint slides were used to present the data. Each slide contained the sentence
number and a play button. Orthoghraphic representation of the sentences were not provided to
listeners. They were able to move along each item at their own pace, and listen to each sentence
as many times as they wanted, however they were discouraged to replay the same sentence more
than three times. Participants were instructed to listen to determine the most prominent word in
each sentence. A prominent word is defined as the word that is highlighted for the listener, and
stands out from other non-prominent words.
Before participants did the first set, they were given a practice session of six sentences to
familiarize them with the task. Sentences in the training session also included sentences with
narrow and neutral focus conditions. None of the sentences with neutral condition had similar
structure to those used in the actual stimulus. After completing the practice session and asking
questions, participants listened twenty-four pseudo-randomized items (eight target and sixteen
filler items) binaurally over Bose QC15Q Acoustic Noise Cancelling Headphones at a
comfortable listening volume in a quiet room. Their responses were recorded by the author (12
participants × 8 test sentences × 3 conditions ([+H*Hn +!H*], [-H*Hn +!H*], [+H*Hn, -!H*]) =
288 judgements). These judgements, coded as TRUE if they identified the most prominent word
correctly and FALSE otherwise, used as the fixed factor in mixed-effects linear model using the
137
lmer function in the lme4 package (Baayen et. al. 2008). In addition to JUDGEMENT as the fixed
effect, the predictors in the model included SUBJECT and SENTENCE as the random factors.
3.4.2 Results
Figure 89 display the frequency of correct responses under three conditions ([+H*Hn
+!H*], [-H*Hn +!H*], [+H*Hn -!H*]). According to this figure, the highest correct identification
of the most prominent word occurred when both prominence marking cues (H*n, and !H*) was
present in the stimulus ([+H*Hn +!H*] condition), and listeners’ identification was lowest when
the prominence marking edge tone (H*Hn) was modified ([-H*Hn +!H*] condition). The lack of
nuclear pitch accent on the other hand did not have an effect on the identification of the sentence
prominence.
Figure 89. Frequency of correct identification of the most prominent word under three
conditions.
These observations were confirmed by the statistical analysis. According to the model
where listeners’ judgement was compared under three conditions, there was a significant effect
of condition (df=2, χ
2
=6.2188, p=0.04463). Results from multiple comparisons indicated that
the difference between condition [+H*Hn +!H*] and [-H*Hn +!H*] was significantly different
(Intercept=-0.7224, z=-2.001, p=0.0454), confirming our observation that lowering prominence
138
marking edge tone (H*Hn) affects listeners’ correct identification of the most prominent word in
the sentence. On the other, the difference between [+H*Hn +!H*] and [+H*Hn -!H*] did not
reach significance (Intercept=0.0786, z=0.201, p=0.8410) suggesting that presence or absence
of nuclear pitch accent did not make a difference in listeners’ judgement.
3.5 Local general discussion
The experiments presented in this chapter aimed to provide quantitative data to support
the phonological intonational model of Turkish described in Chapter 2, in particular prenuclear
high tones realized on finally stressed words. In the current model, I distinguished three separate
prenuclear High tones: H* pitch accent, H*H- ip phrase final tone, H*Hn prominence marking
left-edge tone. In Experiment 1 and Experiment 2, I showed that these three tones are
phonetically different supporting the tonal distinctions proposed in the current model. In this
current model, I marked a phrase medial High tone on a finally stressed word by H* pitch accent,
and a phrase final one by H*H-, and quantitative results from Experiment 1 supported this
distinction. This was one major difference of the current model from those proposed by Kan
(2009) and Kamalı (2011). In these models, no such distinction was made: Kan (2009) analyzed
a phrase medial and a phrase final High tone on finally stressed words as pitch accents, while
Kamalı(2011) treated word final High tones as phrase tones (The status of finally-stressed phrase
medial words is not discussed in Kamalı’s model).
The aim of Experiment 2 was to test whether prominence marking left-edge tone on
finally stressed words, H*Hn, was phonetically realized different from the other two high tones
on finally stressed words; H* and H*H-. To this end, H*Hn was compared to H*H- in
Experiment 2 because Experiment 1 already showed that H*H- was realized with greater f0 rise
than H*, and pitch track examples already showed that H*Hn had higher peak than H*H-,
therefore if I showed H*Hn was realized with higher f0 peak or greater magnitude of f0 rise than
H*H-, this would automatically mean that H*Hn was different from H*. The results indicated
that f0 peak for H*Hn was realized higher than that of H*H-. This finding supports the proposal
made in this current intonational model that sentence prominence in declaratives in Turkish is
marked by a left-edge High tone in addition to the nuclear pitch accent. This is a novel proposal
of this current model. Although previous intonational models displayed pitch contours that
139
display higher f0 peak for immediately prenuclear words, none of them associated this edge-
marking high tone with the marking of sentence prominence. Experiment 1 and 2 supported the
proposal made in this current model that there are indeed three kinds of word-final H tone when
it’s stressed (thus all are H* pitch accents). That is, I showed that the word-final high tones differ
in three ways: when it’s a phrase-medial simple pitch accent when it marks an ip-boundary, and
when it marks prominence of nuclear word. So, it can be said that the data from Experiment 1
and 2 justifies the way labeling proposed in the current model; H*, H*H-, H*Hn, where the
second High tone label after H* represents their additional function (other than being a pitch
accent, i.e. H*). I expect that the distinction between H*H- vs. H*Hn parallels to the distinction
between LH- vs. LHn which are purely phrasal tones realized at the end of the phrase when the
phrase final syllable is not stressed.
The aim of Experiment 3 was to test the role of two prominence marking tones, H*Hn
and !H*, in the perception of sentence prominence. The results from prominence judgement task
showed that listeners’ correct identification of the most prominent word in a sentence was
significantly lowered when the prominence marking left-edge tone, H*Hn, was taken away from
the acoustic signal. On the other hand, deleting the nuclear pitch accent did not have a significant
effect on correctly identifying the most prominent word. This finding from the perception
experiment shows that not only speakers use an additional tonal cue, H*Hn, other than the
nuclear pitch accent, to mark sentence prominence, but also that listeners make use of this
prominence cueing additional tone, H*Hn, to distinguish the most prominent word from other
words in the sentence. This finding that sentence prominence in Turkish declarative sentences
requires an additional tonal event other than the nuclear pitch accent has important theoretical
implications, especially in the debate of pitch-accent first versus stress-accent first theories. I
will return to this issue in the next chapter.
140
Chapter 4: Summary, discussions, and conclusion
This chapter provides a brief but comprehensive summary of the intonational model of
Turkish (§ 4.1) proposed in Chapter Two, as described using the AM framework of Intonational
Phonology introduced in Chapter One. In addition, new findings compared to previous
intonational models are presented in 4.2.1 followed by typological analysis of Turkish based on
phrasal prominence (§ 4.2.2) and the theoretical implications of the findings (§ 4.2.3) followed
by concluding remarks and future work.
4.1 Summary of the intonational phonology of Turkish
The primary goal of this study is to develop a phonological model of Turkish intonation.
By adopting the framework of Autosegmental-Metrical (AM) theory of intonational phonology
(Pierrehumbert 1980, Beckman & Pierrehumbert 1986, Pierrehumbert & Beckman 1988, among
others), the model interprets a continuous pitch contour as a series of discrete low (L) and high
(H) tonal targets aligned with metrically prominent syllables (i.e. pitch accents), and prosodic
boundary edges (i.e. boundary tones). The current model distinguishes three tonally marked
prosodic units: prosodic word (PW), intermediate phrase (ip), and Intonation Phrase (IP).
The smallest tonally marked unit is the prosodic word. Each prosodic word is marked by
an L tone at its left-edge. In addition, a prosodic word can host one of the three post-lexical pitch
accents — H*, !H*, ^H* — aligned with its metrically strong syllable. The two monotonal pitch
accents — !H* and ^H* — are associated with the nuclear word of the sentence depending on
the sentence type. In declaratives sentences, including narrow focus declaratives, a !H* pitch
accent marks the nuclear word. The current model adopts a downstepped pitch accent as the
nuclear tone in declarative sentences due to its highly compressed pitch range. In questions, on
the other hand, the tonal contour on the nuclear word has the highest peak in the sentence. This is
marked by an ^H* pitch accent in the current model. Post-lexical prominence on the prenuclear
prosodic word is marked by regular H* pitch accent.
Above the prosodic word, two larger prosodic units are identified which are demarcated
by a phrase boundary tone at their right edges: the intonation phrase (IP) and the intermediate
phrase (ip). The IP is the largest tonally marked prosodic unit which roughly corresponds to a
141
clause or a sentence. Right edge of an IP can be marked by a L% or H% depending on the
sentence type. Between the prosodic word and intonation phrase is the intermediate phrase which
marks the edges of smaller phrases. Right edge of an ip is marked by one of the two allotones —
H- or LH- — depending on the lexical prosody of the word: A H- marks the right edge of an ip
when the ip final word is lexically stressed on its final syllable. Because this ip-final syllable is
also pitch accented, this ip-final High tone is a combination of H* and H-, labeled as H*H-. On
the other hand, when the ip-final word is non-finally stressed, a LH- marks the right edge of an
intermediate phrase. The sole ‘-‘ diacritic indicates the single function of the tone as marking the
right edge of an intermediate phrase only.
In addition to intermediate phrase marking boundary tone, the current model introduces a
prominence marking edge tone which demarcates the left edge of the nuclear word. A
prominence marking edge tone is aligned with the final syllable of the immediately prenuclear
word and is phonetically realized with a higher f0 peak than that of a regular pitch accent or ip-
boundary tone. Prominence marking function of a boundary tone is marked by the ‘n’ diacritic,
Tn, to distiguish it from a regular phrase final tone (T-). Left edge of the nuclear word is
demarcated by one of the two allotones: Hn or LHn. The Hn is used when the immediately
prenuclear word is finally stressed. As in the ip-final stressed syllable, the Hn toned syllable is
also pitch accented, thus the label “H*Hn” is used to represent the dual function of the High
tone, i.e., a combination of H* and Hn. On the other hand, if the immediately prenuclear word is
non-finally stressed, LHn is used to mark the left edge of sentence prominence to indicate its sole
function as marking prominence only. The left edge of nuclear word is tonally marked in
questions as well, but unlike a rising tone in declaratives, the left edge of the most prominent
word in questions is marked by a low tone (Ln).
Having briefly summarized the intonational model proposed, I know move onto the
discussion of the results, in particular the implications of the tonal realization of sentence
prominence in Turkish with respect to prosodic typology and on the relation between nuclear
pitch accent and focus.
4.2 Discussions
Being developed based on a large corpus of data, the current phonological model of
Turkish intonation supplements the earlier models of Turkish intonation in three main aspects:
142
(1) finer distinction of the status of high tone on phrase-medial and phrase-final words that are
lexically stressed on their final syllables, (2) uniform analysis of nuclear pitch accent, (3)
discovering prominence marking edge tone. In what follows, I will first summarize the new
findings introduced in the current model, and then move onto the theoretical implications of
these findings.
4.2.1 New findings
Findings unique to the current model include distinguishing the phrase medial and phrase
final high tone on finally stressed words, tonal analysis of nuclear word, and introducing a new
prominence marking edge tone.
High tone on finally stressed words
One major difference of the current model from the previous intonational models of
Turkish is its ability to distinguish the function of phrase medial and phrase final high tone on
finally stressed words. The intonational model developed by Kan (2009) treats phrase medial and
phrase final high tone on finally stressed words uniformly as pitch accents, i.e. even if a finally
stressed word appears as the last word in a prosodic phrase (PPh in Kan’s analysis), the high tone
realized on the final syllable of this phrase final word is marked as a pitch accent. As a
consequence of this uniform treatment of phrase medial and phrase final high tones on finally
stressed words, the tonal analysis of a pitch contour does not reflect the true prosodic
organization of the sentence when the right edge of PPhs coincide with a finally stressed word.
Considering the fact that the regular stress pattern of Turkish is final stress, it becomes even
more important to provide a model that correctly couples the prosodic structure of a sentence
with its tonal structure. However, this is not possible with the assumptions made in Kan’s (2009)
model in which a high tone on a finally stressed word is always marked as a pitch-accent
regardless of its function.
Kamalı (2011), on the other hand, provides a completely different analysis. First of all,
words that are traditionally categorized as finally stressed are assumed to be not lexically
stressed in her model. This being the case, no pitch accent is associated with words that are
143
traditionally stressed on their final syllables. A high tone on the final syllable of this type of
words is treated as phrasal tones. Since the sentences analyzed in this study contained single
word phrases only, the tonal analysis seems compatible with the prosodic structure as each
prosodic word is also the last word in a prosodic phrase. However, how this model would
analyze high tone on phrase medial finally stressed words is not clear.
In the current model, the prosodic unit above the prosodic word (PW) is claimed to be an
intermediate phrase (ip). In this model, an ip-medial word-final high tone is analyzed as a pitch
accent, H*. But, when the ip-final word is finally stressed, it is assumed that the High tone has a
dual function, i.e., a pitch accent as well a boundary tone marking the right edge of an ip, thus
labeled as H*H-. The distinction made in the tonal categories is also supported by quantitative
data showing that ip phrase final high tone is phonetically realized by a larger pitch range than its
phrase medial counterpart.
Tonal analysis of the nuclear word
Unlike the previous intonational models, this current model makes a clear cut distinction
between prenuclear and nuclear pitch accents. A pitch accent on a prenuclear prosodic word is
marked by H* pitch accent while nuclear pitch accent is either a !H* (in declaratives) or a ^H*
(in questions). Such distinction is made because nuclear pitch accent in Turkish is realized by
quite a compressed pitch range in declarative sentences but with an expanded pitch range in
questions, unlike prenuclear pitch accents which display a regular rising nature (though the pitch
register is raised in questions, resulting in undershot of L tones).
This analysis is different from the way nuclear pitch accent is analyzed in the previous
models. Previous models analyzed declarative sentences only, therefore the comparison among
the models is based on this type of sentences.
Based on the pitch tracks presented, in Kan (2009) a nuclear pitch accent can be realized
either by a level tone (think of it as compressed pitch range), or by a rising tone. According to
Kan (2009), level tone on the nuclear word can be realized in one of the two ways: (1) it can
either sustain the same level as the previous high tone, which Kan marks by H*, or (2) it can be
realized with a level tone that is lower than the previous f0-peak. This second realization is
marked by !H* pitch accent, and this is the only point where Kan’s analysis of nuclear pitch
144
accent is similar to the analysis proposed in the current model. On the other hand, nuclear pitch
accent with compressed pitch range that sustains the same f0-level as the preceding f0-peak is
not observed in my data unless the nuclear word is too short, e.g. two syllables, therefore there is
not enough ‘time’ for the f0 lowering to be realized on the pitch contour, but such cases result
due to phonetic undershooting and is not part of the phonological model developed in this study.
In addition, Kan (2009) also reports nuclear pitch accents in declaratives that display substantial
f0 rise. This, Kan (2009) marks by L+H* or L+!H* pitch accents. However, I did not observe
such substantial f0-rise on the nuclear word in declarative sentences. Kamalı (2011) also reports
a similar observation that nuclear pitch accent does not display a substantial f0-rise. However,
she confines this property to nuclear words that are stressed on their final syllables only. When
we compare Kan’s analysis of nuclear pitch accent with that of the current model’s, we see that
Kan (2009) proposes two possible realization of the nuclear pitch accent: (1) compressed pitch
range, (2) substantial f0 rise. Based on the data collected for the current study, the second option,
i.e. substantial f0 rise, is not proposed in the current model. This is one difference between the
two models. The second difference is that although both models propose compressed pitch range
on the nuclear word, this compressed pitch range is always assumed to have a lower level than
the previous high tone in the current model while in Kan (2009), a nuclear pitch accent with
compressed pitch range can have the same f0 level as the previous high tone. In short, in Kan
(2009) nuclear pitch accent with compressed pitch range which has a lower level than the
previous f0-peak is one of the four possible realizations of nuclear pitch accent while in the
current model, it is the only tonal property in the intonational model ascribed to the nuclear pitch
accent of declaratives.
Nuclear pitch accent with compressed pitch range is also present in Kamalı’s (2011)
model however the way she analyzes this phenomenon is completely different from the current
analysis I propose or Kan’s analysis. In Kamalı (2011) the analysis of the tonal event on the
nuclear word depends on the lexical prosody of the word itself. In particular, when the nuclear
word is non-finally stressed, it is suggested that the nuclear pitch accent is realized by no f0-rise.
In her phonological model, Kamalı (2011), analyzes this property as L- tone, which is associated
with the left edge of MaP/PhP/AP, spreading rightward until the end of the nuclear word.
Therefore, no pitch accent is associated with the finally stressed nuclear words (remember that
according to Kamalı words that are traditionally stressed on their final syllables are lexically not
145
stressed). On the other hand, when the nuclear word is non-finally stressed, tonal event on the
nuclear word is not distinguished from prenuclear pitch accents; both marked by H*L lexical
pitch accent. In short, in Kamalı’s model, similar to Kan’s, compressed pitch range is one of the
possible realizations of the tonal event on the nuclear word. The difference between the two in
this respect is that in Kamalı whether the nuclear tone has compressed pitch range or not depends
on the lexical prosody of the nuclear word while in Kan, no such criterion is proposed. Unlike
the two previous models, I proposed that nuclear pitch accent is systematically realized by
compressed pitch range; either by no f0-rise or a very small f0-rise, and this is observed to be
true for both finally and non-finally stressed words. Following this observation, I proposed !H*
as the nuclear pitch accent aligned with the metrically-strong syllable of the nuclear word.
Prominence marking edge tone
The most novel proposal of this dissertation is its analysis of the tonal properties of
sentence prominence in Turkish. Unlike the two previous intonational models, the current model
associates sentence prominence with two tonal events; a nuclear pitch accent plus an edge-
marking boundary tone. The nuclear pitch accent is realized on the stressed syllable of the
nuclear word. Boundary tone which marks the left edge of nuclear word, on the other hand, is
aligned with the final syllable of the immediately prenuclear word. This phenomenon, although it
is rare, is similar to AP final boundary tone in Tokyo Japanese which is realized on the initial
syllable of the following AP (Beckman & Pierrehumbert 1986, Pierrehumbert & Beckman
1988).
Previous intonational models associate sentence prominence with a pitch accent only
(Kan 2009, Kamalı 2011). Although in both of the tonal models, pitch range expansion on the
final syllable of the immediately prenuclear word is exemplified in the pitch tracks provided, no
association between this boundary tone and sentence prominence has been suggested. In what
follows, I will discuss the implications of this proposal with respect to the prosodic typology of
languages based on phrasal prominence and what it implies on the relation between nuclear pitch
accent and nuclear stress.
146
4.2.2 Turkish as a head/edge marking language
The advent of the Autosegmental-Metrical model of intonational phonology in the 1980s
provided a uniform framework to describe the prosodic features of different languages using the
same prosodic categories. By now there are quite a few number of languages whose prosody has
been described within the framework of AM theory of intonational phonology. Describing the
prosody of languages using the same framework made it possible to compare and eventually
categorize languages based on their prosodic features.
Prosody, in the AM framework, is described based on two aspects: (1) prosodic structure
of an utterance, (2) prominence relations within the prosodic structure. A prosodic structure is a
hierarchical organization of prosodic units, and within a prosodic unit some words are more
prominent than others which is marked by segmental features such as vowel quality as well as
suprasegmental features such as pitch, duration, and/or amplitude.
As introduced in Chapter One, recent work on prosodic typology (Jun 2005, 2014)
classifies languages based on the type of prominence marking at phrasal level: phrasal
prominence can be marked culminatively (in a bottom-up fashion) when the head of a prosodic
unit such as a stressed syllable receives a pitch-accent (known as head-prominence), or
demarcatively by manipulating the prosodic phrasing of an utterance where the prominent word
is aligned with the right/left edge of some prosodic unit (known as edge-prominence). A third
possibility is to mark both the head of a phrase and its edge. Languages that avail the third option
to mark phrasal prominence are categorized as head/edge prominence languages.
One of the research questions asked in the current study is: Where does Turkish fall in
this typology: is it a head-marking, edge-marking, or head/edge marking language? Put another
way, is phrasal-level prominence in Turkish marked by a pitch accent, or by aligning the
prominent word with some phrase edges, or both.
Based on the pitch tracks analyzed as well as the phonetic experiments conducted to test
the validity of the phonological tones proposed in the current model, I proposed that phrasal
prominence in Turkish is realized by both a head marking pitch-accent and an edge marking
boundary tone. Following these observations and experimental results, I propose that Turkish is a
head/edge marking language similar to French and Bengali (Jun & Fugeron 2002, Khan 2008,
Jun 2005, 2014).
147
Categorizing Turkish as a head/edge prominence language also provides an explanation
to two unusual phenomena in Turkish intonation that would otherwise remain obscure:
compressed nature of the nuclear pitch accent, and pitch range expansion observed on the
immediately prenuclear word.. These two phenomena are unusual because nuclear pitch accent
which is aligned with the informational nucleus of the sentence often undergoes pitch range
expansion to highlight the informationally most prominent word in the sentence. Pitch range
expansion on the most prominent word in the sentence has been shown by many studies on many
different languages (Botinis et al. 1999, Gu and Lee 2007, Rump & Collier 1996, Xu & Xu 2005
to name a few). Does this mean that Turkish adopts an exactly opposite strategy, i.e. suppressing
f0 to mark the informationally most prominent word in a sentence?
The second phenomenon, i.e. higher f0 peak on the last syllable of the immediately
prenuclear word, is also unusual because the natural tendency of f0 contour is to undergo gradual
downdrift and narrowing of pitch range over the course of sentence if the words are in the same
prosodic unit; a determinant factor of f0 contour referred to as ‘declination’ in the literature
(Pierrehumbert 1980). However, when we look at the f0 contour in Turkish declaratives, we see
expanded pitch range on the final syllable of the immediately prenuclear word. This property of
Turkish intonation in isolation stands against the natural tendency of the f0 contour that has been
observed for other languages (e.g. Dutch (Collier 1975), English (Maeda 1976), Japanese
(Fujisaki et. al (1979), Finnish (Hirvonen 1970), Danish (Thorsen 1980), Swedish (Bruce 1977).
It was suggested by Bolinger (1978) that the phenomenon might be universal. One possibility at
this point would be to say that the phenomenon is not universal and Turkish provides evidence
that it is not.
When the two tonal phenomena are considered separately, we are left with two tonal
events in Turkish that are in stark contrast with what has been reported in the existing literature:
(1) pitch range compression on the nuclear word, and (2) pitch range expansion on the
immediately prenuclear word. However, if we combine two ‘unusual’ tonal events, the picture
that emerges on Turkish intonation does not look ‘unusual’: similar to what the existing literature
reports, informationally most prominent word in the sentence gives rise to pitch range expansion
in Turkish as well, however different from other languages, this pitch range expansion is not
realized on the prominent word itself but on the final syllable of the immediately prenuclear
word. Therefore, what looks like lack of declination on the surface is in fact pitch-range
148
expansion to mark sentence prominence. The difference between Turkish and other languages
then can be reduced to the alignment of intonational peak with the informational nucleus of the
sentence as schematized in Figure 90.
Figure 90. Schematic representation of the marking of sentence prominence in
Turkish declarative sentences compared to other languages.
This claim, i.e. in addition to nuclear pitch accent, there is an additional boundary tone
that marks sentence prominence in Turkish, is supported by qualitative as well as quantitative
data. I have shown that when a non-finally stressed word precedes a nuclear word, the word
bears two f0 rises: a H* pitch accent on its stressed syllable and a second f0 rise at the end of the
word. This second rise, however, did not emerge when the same word appeared in non-
prenuclear position. I have also shown, using quantitative data that, when the immediately
prenuclear word is finally stressed, high tone on its final syllable is realized by a larger pitch
range compared to regular phrase final high tone. The same phenomenon is observed in narrow
focus sentences as well.
Both of these findings, i.e. second high tone on the final syllable of a non-finally stressed
word when it is immediately prenuclear or pitch range expansion on immediately prenuclear
finally stressed word, point to the conclusion that the tonal event realized on the final syllable of
the immediately prenuclear word is triggered by the nuclear word. Based on this evidence, I
concluded that the high tone realized on the final syllable of the immediately prenuclear word
marks the left edge of the nuclear word. Functioning as prominence marker, it undergoes pitch
range expansion overriding the effect of declination. As a consequence, most prominent tonal
event in Turkish is not realized on the informationally most prominent word in the sentence but
on the final syllable of the immediately prenuclear word. What does this finding imply on the
PITCH RANGE
EXPANSION
PITCH RANGE
COMPRESSION
PITCH RANGE
COMPRESSION
PITCH RANGE
EXPANSION
NUCLEAR
WORD
NUCLEAR
WORD
OTHER LANGUAGES TURKISH
149
relation between nuclear pitch accent and nuclear stress? The next subsection addresses this
question.
4.2.3 On the relation between nuclear pitch accent and nuclear stress
What is the relation between nuclear pitch accent (NPA) and nuclear stress (NS), the
element with the highest prominence in the sentence? This section argues the relation between
the two based on the findings on the tonal realization of focus in Turkish. There are two distinct
views on the relation between NPA and NS: (1) The pitch-accent first theory (e.g. Selkirk 1984,
Gussenhoven 1984) argues that the grammar assigns pitch accents according to a set of
algorithms sensitive to predicate-argument relations, and the last pitch-accented word is
perceived as most prominent. (2) The stress-first theory argues that NS is computed either
directly from the syntactic structure (e.g. Cinque 1993, Kahnemuyipour 2004) or on the basis of
a metrical structure derived from the syntactic structure (e.g. Hale & Vergnaud 1987, Ladd 1996,
Zubizarreta and Vergnaud 2005, Zubizarreta 2014). NS is then aligned with NPA at the
intonational level. Under this view, NS is an abstract rhythmic notion, which is acoustically
identified, possibly in different ways across languages. In this section, I will argue that findings
from Turkish support the latter view.
The fundamental question that has been a matter of debate for decades is whether the
relationship between focus and prosodic prominence is part of grammar. While the answer is yes
for most of the researchers, some argued that the relationship is independent of grammar. One of
the well-known proponents of the theory that claims that accent is independent of grammar is
Bolinger (1972, 1982) who argues that pitch accent is directly reflecting speaker’s
communicative intent. Therefore, the only relationship between prosodic prominence and focus
is mediated by the informational status of the word, i.e. ‘given’ or ‘new’, that the accent is
aligned with, and cannot be predicted by grammar.
The theory proposed by Bolinger (1972, 1982) assumes a one-to-one relationship
between prosodic prominence and focus, therefore it is a theory of narrow focus only. However,
as has been argued by many researchers, pominence-focus relation cannot be reduced to a one-
to-one mapping and a theory between the relation between the two should explain multiple focus
domains, i.e. focus projection.
150
The concept of focus projection goes back, at least, to Chomsky (1971) and Jackendoff
(1972) according to whom focus is grammatically represented and that the location of Nuclear
Stress (NS), i.e. the rhythmically most prominent word, is fundamental in its identification. NS is
then aligned with the intonational nucleus of the sentence, namely the Nuclear Pitch Accent
(NPA). This approach directly links focus with nuclear stress, therefore stress-first. An
alternative approach proposed by Selkirk 1984, 1995, Gussenhoven 1984 argues that focus
identification is directly linked to the distribution of pitch accents, hence pitch-accent first, and
nuclear stress or its tonal counterpart, NPA, does not play any fundamental role in its
identification. NPA, in this approach, is defined as the last pitch accent in a series of pitch
accents in the relevant phrase and has no fundamental role in ascribing the focus domain of the
sentence. This brings us back to the main question we asked at the beginning of this section:
What is the relation between nuclear pitch accent and nuclear stress?
According to pitch-accent first theories, focus IS pitch accent. At first glance, this tight
relation between the two is appealing. When we look at the tonal realization of focus, where
informational prominence is marked by a pitch accent, the focused word does not only receive a
pitch accent but also undergoes pitch range expansion, i.e. f0 is a very strong correlate of focus
in the acoustic dimension in particular above the word level in many languages. Consider the
pitch contour in Figure 91, where nuclear pitch accent falls on the word ‘new’ in the sentence
‘Sarah bought a new car.’ As can be seen in this figure, the nuclear pitch accent is acoustically
most prominent peak in the entire sentence.
151
Figure 91. Pitch track example displaying pitch range expansion on the narrowly
focused word in English.
Pitch range expansion as a correlate of focus marking has been demonstrated for
languages other than English as well, e.g Greek and Swedish (Botinis et. al. 1990), Egyptian
Arabic (Hellmuth 2006). The same holds true even in a tone language where f0 cues distinguish
lexical items. For example, in Mandarin Chinese although the shape of the tone is maintaned to
encode lexical information, pitch range is manipulated to mark information structure (Peng et al.
2005).
The correlation between pitch accent and focus extends to the perception dimension as
well. Considerable amount of research has been dedicated to tease apart the role of f0 on the
perception of prominence at and above the word level. At the word level, f0 has been found to be
a less of a reliable cue. Beckman (1986) showed that English speakers relied less on f0 and more
on duration and intensity in their judgements of word level prominence. Similar results have
been reported by Turk and Sawusch (1996); duration and intensity as better predictors of word
level prominence and f0 less so. At the phrasal level, on the other hand, f0, acoustic correlate of
pitch accent, has been shown to be more of a relevant cue in the perception of prominence. A
systematic finding on the relation between f0 and the perception of prominence is that a higher f0
Time (s)
1.088 2.643
Pitch (Hz)
70
400
NEW
Susan bought a
car
152
peak corresponds to greater perceived prominence (Rietveld and Gussenhoven 1985, Ladd 1993)
with some corrections by listeners for declination (Pierrehumbert 1979).
The picture that emerges so far based on the findings in the previous literature is that
pitch accent plays a fundamental role on the production and perception of prominence. In
particular, speakers increase pitch range to ‘highlight’ informational prominence and this pitch
range expansion cue listeners in identifying sentence prominence. Findings as such might
support the arguments that directly link nuclear pitch accent with focus; i.e. pitch-accent first
theories. However as I will present below, the findings from Turkish show that the relation
between the two is not as tight.
So, what does the picture in Turkish look like? At first sight, it looks different from what
the existing literature suggests: In Turkish informationally most prominent word in the sentence
is marked by a pitch accent but this pitch accent has a quite compressed pitch range; a finding
that is in stark contrast with the findings from most languages reported so far. However, I have
shown that there is an additional tone, an edge tone, which also functions together with the
nuclear pitch accent to mark focus in Turkish. This additional tone, which I have shown to be
higher than a regular phrase final tone, is realized on the final syllable of the immediately
prenuclear word. Put another way, in Turkish acoustically most prominent tonal event,
prominence marking edge tone, is not coupled with the informationally most prominent word in
the sentence, but is realized on the immediately prenuclear word. This is the production side of
focus in Turkish. In terms of perception, I have shown that edge marking tone contributes
significantly to the correct identification of prominence while nuclear pitch accent does less so.
In short, in Turkish speakers align the most prominent tonal event with the final syllable of the
immediately prenuclear word while they compress the pitch range of nuclear pitch accent, and
listeners make use of the acoustically prominent tone on the immediately prenuclear word more
to correctly identify the most prominent word in the sentence and rely less on the nuclear pitch
accent for that purpose. This has been shown in the perception experiment presented in Chapter
Three: I have shown that when the prominence marking edge tone has a lower peak, listeners’
correct identification of focus significantly dropped while the presence/absence of nuclear pitch
accent did not have a significant effect on the perception of focus.
Although this picture might initially look different from what has been reported in the
previous literature on the production and perception of focus, a careful interpretation of the
153
findings shows that the results from Turkish is actually not any different. Let us summarize what
the previous literature reports on the relation between focus and pitch accent: speakers increase
pitch range on the prominent word to mark focus and this change facilitates perception of
prominence for listeners. What I have reported for Turkish with respect to the relation between
focus and pitch range expansion looks different in the sense that informationally most prominent
word in the sentence is not aligned with the intonational nucleus of the sentence. On the contrary,
nuclear pitch accent has the lowest peak in the sentence. However, the boundary tone marking
the left edge of nuclear word is realized by expanded pitch range. Therefore, as mentioned earlier
Turkish is not any different from the other languages with respect to the relation between focus
and pitch range expansion. The only difference is the alignment of intonational nucleus with
respect to informational nucleus of the sentence; in Turkish acoustically most prominent f0 peak
is not aligned with the nuclear word itself but with the right edge of the immediately prenuclear
word. And crucially, pitch range expansion at the right edge of the immediately prenuclear word
facilitates perception of prominence for the listeners, as is shown by a perception experiment in
Chapter Three. This suggests that similar to what has been reported in the literature with respect
to the relation between pitch range expansion and the perception of prominence such that larger
pitch range increases the rate of correct perception of prominence, results from Turkish complies
with this finding, except that the pitch range expansion contributing to the perception of
prominence is aligned with immediately prenuclear word, in other words pitch range expansion
on the preceding f0-peak increases the perceived prominence of the following word. This finding
can be related to what has been reported in Gussenhoven and Rietveld 1988. One of the
perception experiments in that study tests the dependency between the prominence of the first
peak and that of the second, and find that prominence of the first peak influences the prominence
of the second such that a higher first f0 peak increases the perceived prominence of the second.
In Turkish, pitch range expansion on the final syllable of the immediately prenuclear word
contributes to the perceived prominence of the nuclear word.
Coming back to our initial question one more time: What is the relation between nuclear
pitch accent and nuclear stress? Is there a direct relation between focus and pitch accent as
suggested by pitch-accent first theories? Findings from Turkish suggest that there is not.
Although in Turkish focus is marked by a pitch accent, that pitch accent is not the most
prominent accent in the sentence, and crucially it does not significantly contribute to the
154
perception of prominence. As the results from the perception experiment presented in Chapter 3
indicate, when the nuclear pitch accent is eliminated from the pitch contour, listeners correct
identification of sentence prominence was not significantly affected. What is more, if the
boundary tone is eliminated, the nuclear pitch accent on its own was not a sufficient cue for the
listeners. These findings suggest that NPA is neither a necessary nor a sufficient cue to mark NS,
most prominent element in the sentence, in Turkish. This implies that the relation between NPA
and NS is not NPA equals NS or vice-versa as is implied by pitch-accent first theories.
Therefore, there is no obvious way to model the relation (or the lack thereof) between NPA and
NS in Turkish. A more plausible way to look at Turkish phenomenon would be to consider NS as
an abstract rhythmic notion whose prominence is determined by the metrical structure rather than
by NPA as suggested by stress-first theories. In this metrical structure, the most prominent word
receives NPA, and the right edge of immediately prenuclear word is marked by an edge tone.
At this point, however, it is important to interpret the results from the perception
experiment correctly (Experiment 3). For the perception experiment, I only modified
fundamental frequency realized as two prominence marking high tones, H*Hn and !H*, and
asked listeners to identify the most prominent word in the sentence. The results showed that the
lack of nuclear pitch accent, !H*, did not significantly affect listeners’ correct identification of
sentence prominence, and that the left-edge marking H*Hn tone had a more significant role in
cueing sentence prominence for listeners. However this result should not be interpreted as H*Hn
is the only acoustic cue that signals sentence prominence for the listeners: It is quite possible that
listeners use other suprasegmental cues such as duration or amplitude on the prominent word
together with the H*Hn to identify sentence prominence. Although none of the experiments in
this study are designed to test acoustic correlates of sentence prominence under neutral focus
context, I have compared the duration of the stressed syllable of a NPA word with its non-NPA
counterpart in one near minimal-pair of sentence given in 27 below. This pair of sentence was
picked from the production experiment designed to compare H*Hn versus H*H- (Experiment 2).
Recall that I used two groups of SV intransitive sentences in that experiment: one group had
eventive reading where nuclear pitch accent fell on S 27a, and the second group had generic
reading therefore the NPA fell on the verb 27b. So, while the stressed syllable of S carries NPA
in 27a, it does not carry NPA in 27b. The relevant stressed syllable is written in capitals in 27.
Both sentences have same number of syllables for each word.
155
27.
a. Sabaha karşı babamLAR geldi.
Morning.dat towards dad.pos.plu come.past
“Towards the morning my dad came.”
b. Sabaha karşı babaLAR horlar.
Morning towards dad.plu snore.aor
“Towards the morning dads snore.”
I compared the vowel duration on the final syllable of S; in 27a, it carries the NPA, and in
27b it does not. The comparison is based on the production of five speakers of Turkish who
participated in Experiment 2. The bar chart in Figure 92 display mean duration of the vowel /a/
when it carries NPA (+NPA) and when it does not (-NPA) as produced by five speakers, and
each sentence is repeated three times.
Figure 92. Mean vowel duration of the stressed syllable when it carries the
nuclear pitch accent (+NPA) versus when it is not (-NPA).
This preliminary finding of changes in duration due to prominence suggests that although
the tonal event on the nuclear word is compressed, speakers enhance other acoustic features such
as duration to cue sentence prominence. How all these acoustic features, such as high tone
0.04
0.06
0.08
0.1
0.12
0.14
-NPA +NPA
Duration (sec)
156
realized on the final syllable of the immediately prenuclear word and increase in duration on the
prominent word itself interact to cue prominence to listeners require further investigation.
4.3 Concluding remarks and future directions
In this study, I set out to develop a phonological model of Turkish intonation using the
theoretical framework of Autosegmental-Metrical theory of intonation, and provide quantitative
data to support the tonal analysis proposed in the model. Based on a large corpus of data, I
proposed a full inventory of pitch accents and boundary tones. While doing so, I also compared
the current model with the two previous intonational model: Kan (2009) and Kamalı (2011).
The current model supplements the previous intonational models of Turkish in three main
aspects: First of all, unlike the two previous models, the current model makes a clear distinction
between phrase-medial and phrase-final high tone on finally stressed words. In the previous two
models, both the phrase-medial and the phrase-final high tone on finally stressed words were
both treated either as pitch accent (Kan 2009) or boundary tones (Kamalı 2011, although the
status of phrase medial ones are not clear in this model). In the current model, on the other hand,
a phrase-medial high tone is marked by H* pitch accent and a phrase-final high tone is marked
by H*H- consistent with its functions as demarcating the right edge of a bigger prosodic
boundary, namely an intermediate phrase. This distinction in the tonal tier correctly reflects the
prosodic structure of the sentence that the intonational contour is coupled with. Second of all,
the current model provides a more uniform analysis of the tonal marking of sentence prominence
in Turkish. Previous models ascribes multiple pitch accents for the nuclear word (Kan 2009) or
makes a distinction based on the lexical prosody of the nuclear word (Kamalı 2011). In this
current model, the nuclear pitch accent is associated with the !H* tone (for declaratives)
uniformly due to its compressed pitch range. As I have displayed using pitch track examples, this
compressed nature of nuclear pitch accent is not determined by the lexical prosody of the nuclear
word: The true underlying nature of the nuclear pitch accent is that it has compressed pitch range
and this is true for finally stressed as well as non-finally stressed nuclear words. Third, the
current model associates a second tone, an edge marking boundary tone, with the marking of
sentence prominence together with the nuclear pitch accent. This edge marking tone is realized
on the immediately prenuclear word and is realized with higher f0-peak than a regular phrase-
157
final boundary tone. Although previous models mention the f0-boosting on the immediately
prenuclear word, neither associates this with the marking of sentence prominence.
One of the research questions asked in this dissertation was where Turkish belongs to in
the prosodic typology of languages based on phrasal prominence: is it a head-marking language,
edge marking language, or head/edge marking language. After showing that both the head and
the edge of sentence stress is tonally marked, I categorized Turkish as a head/edge prominence
language. Interestingly, though, head marking tone, i.e. the nuclear pitch accent, has compressed
pitch range while the edge marking boundary tone has expanded pitch range. What is more, I
showed that prominence marking edge tone which is phonetically realized on the immediately
prenuclear word is more of a significant cue in the correct identification of prominence while
nuclear pitch accent did not seem to play a significant role in the perception of prominence. This
finding, I used to discuss the relation between nuclear pitch accent and nuclear stress in the
context of its implication for pitch-accent first (Selkirk 1984, 1995; Gussenhoven 1984) and
stress-accent first theories (Cinque 1983, Kahnemuyipour 2004, Halle & Vergnaud 1987, Ladd
1996, Zubizarreta & Vergnaud 2005, Zubizarreta 2014). Since NPA in Turkish is neither
necessary nor sufficient for the perception of prominence, I proposed that there is no direct link
between NPA and NS and suggested that the phenomenon in Turkish can be best explained by a
stress-first theory which treats NS as an abstract rhythmic notion, which is acoustically
identified, possibly in different ways across languages.
The findings in the current study raises questions on a number of points for future work.
The first point is the details of the interaction of prominence marking edge tone with other
acoustic cues such as duration and/or intensity in the perception of prominence. As I have briefly
touched upon in the discussion section, although nuclear pitch accent has a compressed pitch
range, the vowel on the stressed syllable of the nuclear word has longer duration. This suggests
that although nuclear pitch accent is compressed, there are other acoustic cues that are boosted to
highlight sentence prominence. Then the next question is how do these acoustic cues such as
longer duration interact with the prominence marking boundary tone to contribute to the correct
perception of prominence.
One interesting point reported in this study is the lack of any prosodic break between a
relative clause and the head noun it modifies. Interestingly, in Turkish a pre-relative clause and
the head noun it modifies are grouped into one prosodic phrase: (RC N)
NP
. The lack of a
158
prosodic break in between the RC and the head noun would have interesting implications with
respect to attachment preferences across languages. The phenomenon is that when the head of a
RC is a complex noun phrase (e.g. the servant of the actress who was on the balcony) speakers of
some languages, such as English or Arabic, prefer low attachment, i.e. the relative clause who
was on the balcony modifies the actress in the sentence Someone shot the servant of the actress
who was on the balcony. Speakers of languages such as Spanish, German or Japanese on the
other hand prefer high attachment, i.e. the RC modifies the servant. Fodor (1988) claims that this
cross-linguistic difference in attachment preferences is due to the differences in the prosody of
languages; since languages differ in their prosody, attachment preferences would vary
accordingly. For example, the default prosodic grouping of such structures for Japanese speakers
is (RC) // (NP1 NP2); bigger prosodic break between the RC and the two noun phrases, and this
grouping results in higher attachment preferences, i.e. RC is attached to NP2 (Jun & Koike 2008,
Jun 2010). In Turkish, the prosodic grouping is different from that of the Japanese. I have shown
that speakers do not put a prosodic break between the RC and the noun it modifies. This implies
that unlike Japanese speakers, Turkish speakers would prefer low attachment when a relative
clause modifies a complex head noun because the prosodic phrasing would be (RC NP1 NP2). Is
that the case and if so then how would Turkish speakers group the relative clause and the
complex head noun if they mean to attach RC to NP2, i.e. higher attachment?
Finally, the model proposed in this study is an intonational model of standard Turkish.
Dialectal variations would also be worthwhile to study. It would especially be very interesting to
see if sentence prominence in other dialects is marked the same way: a nuclear pitch accent with
compressed pitch range and a boundary tone with expanded pitch range which is phonetically
aligned with the final syllable of the immediately prenuclear word.
These are some of the points that are raised by the findings in the current study. I hope
that with the current phonological model of Turkish intonation as the starting point, other
colleagues would join me in studying Turkish prosody from all subfields of linguistic research.
159
References
Arnhold, A. (2014) Prosodic structure and focus realization in West Greenlandic, in Sun-Ah Jun
(ed.) Prosodic Typology II: The Phonology of Intonation and Phrasing. Oxford
University Press. pp. 216-251.
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed
random effects for subjects and items. Journal of Memory and Language 59: 390-412.
Barker, C. (1989). Extrametricality, the cycle, and Turkish word stress. In Phonology at Santa
Cruz 1, ed. by Junko Itô & Jeffrey Runner, 1-33. Santa Cruz: Linguistic Research Center.
Beckman, M. (1986). Stress and non-stress accent. (Netherlands Phonetic Archives 7).
Dordrecht: Foris.
Beckman, M., & Pierrehumbert, J. (1986) Intonational structure in Japanese and English.
Phonology Yearbook, 3: 255–309.
Boersma, P. & Weenink,D. (2015). Praat: doing phonetics by computer [Computer program].
Version 5.4.06, retrieved 21 February 2015 from http://www.praat.org/
Bolinger, D. (1972). Accent is predictable (if you are a mind reader). Language 48, 633-644.
Bolinger, D. (1978). Intonation across languages. In J. H. Greenberg et. al., 471-524.
Bolinger, D. (1982). Intonation and its parts. Language 58, 505-533.
Botinis, A., Fourakis, M., Gawronska, B. (1990). Focus identification in English, Greek, and
Swedish. In Proceedings of the 14
th
International Congress of Phonetic Sciences: 1557-
1560, San Francisco.
Bruce, G. (1977). Swedish word accents in sentence perspective. Travaux de L’Institut de
Linguistique de Lund, CWK Gleerup, Malmö.
Cho, T. (2005). Prosodic strengthening and featural enhancement: evidence from acoustic and
articulatory realization of /a,i/ in English. Journal of the Acoustical Society of America
117: 3867-3878.
Chomsky, N. (1971). Deep structure, surface structure, and semantic interpretation. Semantics,
ed. by Danny Steinberg & Leon Jakobovits. Cambridge: University Press, 183-216.
Chomsky, N. (2000). Minimalist inquiries: The framework. In Step by step: Essays on minimalist
syntax in honor of Howard Lasnik, ed. Roger Martin, David Michaels, and Juan
Uriagereka, 89-156. Cambridge, MA: MIT Press.
160
Chomsky, N. (2001). Derivation by phase. In Ken Hale: A life in language, ed. Michael
Kenstowicz, 1-52. Cambridge, MA: MIT Press.
Collier, R. (1975). Physiological correlates of intonation patterns. Journal of the Acoustical
Society of America 58, 249-255.
Cruttenden, A. (1997). Intonation. Cambridge: Cambridge University Press, 2
nd
edition.
Crystal, D. (1969). Prosodic systems and intonation in English. London: Cambridge University
Press.
Demircan, Ö. (1983). Türkçe ezgilemeye giriş. TDAY- BELLETEN. Ankara.
Erguvanlı, T. E. (1984). The function of word order in Turkish grammar. Berkeley, CA:
University of California Press.
Féry, C. (2011). German sentence accents and embedded prosodic phrases. Lingua 121: 1906-
1922.
Fox, A. (2000). Prosodic features and prosodic structure: The phonology of suprasegmentals.
Oxford: Oxford University Press.
Fry, D. B. (1955). Duration and intensity as physical correlates of linguistic stress. Journal of the
Acoustical Society of America 27, 765-768.
Fry, D. B. (1958). Experiments in the perception of stress. Language and Speech 1, 120-152.
Fujisaki, M., Hirose, K., & Ohta, K. (1979). Acoustic features of the fundamental frequency
contours of declarative sentences in Japanese. Research Institute of Logopedics and
Phoniatrics Annual Bulletin No. 3, 163-172, University of Tokyo.
Godjevac, S. (2000). An autosegmental/metrical analysis of Serbo-Croatian intonation. Ohio
State University Working Papers in Linguistics 54: 79-142.
Gordon, M. (2004). A phonological and phonetic study of word-level stress in Chickasaw.
International Journal of Americal Linguistics 70, 1-32.
Göksel, A. & Özsoy, S. (2000). Is there a focus position in Turkish? In Studies on Turkish and
Turkic Languages: Proceedings of the 9
th
International Conference on Turkish
Linguistics, ed. Aslı Göksel & Celia Kerslake, 219-228. Wiesbaden: Harrassowitz
Verlag.
Gu, W., & Lee, T. (2007). Effects of tonal context and focus on Cantonese f0. ICPHS,
Saarbrücken.
Gussehoven, C. (1984). On the grammar and semantics of sentence accents. Dodrecht: Foris.
161
Hayes, B. (1989). The prosodic hierarchy in meter. Rhythm and meter, ed. by P. Kiparsky and G.
Youmans, 201-260. Phonetics and Phonology, 1. Academic Press, New York.
Hayes, B. & Lahiri, A. (1991). Bengali intonational phonology. Natural Language and
Linguistic Theory 9. 47-96.
Hellmuth, S. (2006a). Focus-related pitch range manipulation (and peak alignment effects) in
Egyptian Arabic. Speech Prosody, Dresden.
Hirvonen, P. (1970). Finnish and English communicative intonation. Publications of the
Phonetic Department of the University of Turku, No. 8.
Inkelas, S. (1999). The exceptional stress-attracting suffixes in Turkish: Representations versus
the grammar. In The prosody-morphology interface, ed. René Kager, Harry van der
Hulst, and Wim Zonneveld, 134-187. Cambridge: Cambridge University Press.
Inkelas, S. & Orgun, C. O. (1998). Level (non)ordering in recursive morphology: evidence from
Turkish. In Morphology and its relation to phonology and syntax, ed. Patrick Farrell,
360-392.
Inkelas, S. & Orgun, C. O. (2003). Turkish stress: A review. Phonology 20, 139-161.
Ipek, C. (2011). Phonetic realization of focus with no on-focus pitch range expansion in Turkish.
Proceedings of the 17
th
International Congress of Phonetic Sciences. Hong Kong, China.
Jackendoff, R. S. (1972). Semantic interpretation in generative grammar. Cambridge MA: MIT
Press.
Jassem, W., Morton, J., & Steffen-Batóg, M. (1968). The perception of stress in synthetic
speech-like stimuli by Polish listeners. Speech Analysis and Synthesis I, 289-308.
Jun, J., J. Kim, H. Lee, & S.-A. Jun (2006) “The Prosodic Structure and Pitch Accent of
Northern Kyungsang Korean” Journal of East Asian Linguistics 15:4.
Jun, S.-A. (1993). The phonetics and phonology of Korean prosody. Doctoral Dissertation, Ohio
State University, Columbus.
Jun, S.-A. (1995b). Asymmetrical prosodic effects on the laryngeal gesture in Korean.
Phonology and phonetics evidence: Papers in Laboratory Phonology IV, ed. by Connell,
B. & Arvaniti, A., 235-253. Cambridge University Press, England.
Jun, S.-A. (1998). ‘The Accentual Phrase in the Korean Prosodic Hierarchy’, Phonology 15.2,
189–226.
Jun, S.-A. (2005). Prosodic typology: the phonology of intonation and phrasing. Oxford
University Press.
162
Jun, S.-A. (2010). The implicit prosody hypothesis and overt prosody in English. Language and
Cognitive Processes 25, 1201-1233.
Jun, S.-A. (2011). Prosodic markings of complex NP focus, syntax, and the pre-/post-focus
string. In the Proceeding of the 28
th
WCCFL, 214-230.
Jun, S.-A. (2014). Prosodic typology: by prominence type, word prosody, and macro-rhythm.
Prosodic Typology II: The new development in the phonology of intonation and phrasing,
ed. by Sun-Ah Jun. Oxford University Press.
Jun, S.-A. & Fugeron, C. (2002). A phonological model of French Intonation. In A. Botinis [Ed],
Intonation: Analysis, Modeling and Technology, 209-242. Kluwer Academic Publishers.
Jun, S.-A. & Koike, C. (2008) “Default Prosody and Relative Clause Attachment in Japanese”,
Japanese-Korean Linguistics 13, Stanford: CSLI.
Jun, S.-A., Vicenik, C., & Löfstedt, I. (2007). Intonational phonology of Georgian. UCLA
Working Papers in Phonetics 106, 41-57. Los Angeles: University of California.
Kabak, B. & Vogel, I. (2001). The phonological word and stress assignment in Turkish.
Phonology 18, 315-360.
Kahnemuyipour, A. (2004). The syntax of sentential stress. Doctoral Dissertation. University of
Toronto. Published (2009). OUP.
Kaisse, E. (1985). Some theoretical consequences of stress rule in Turkish. In Chicago
Linguistics Society 21, 199-209.
Kamalı, B. (2008). Phases of incorporation and a three-way distinction of direct objects in
Turkish. Talk given at the Workshop on Altaic Formal Linguistics 5, SOAS, London,
May 2008.
Kamalı, B. (2011). Topics at the PF interface of Turkish. Unpublished Doctoral Dissertation,
Harvard University.
Kan, S. (2009). Prosodic domains and the syntax-prosody mapping Turkish. Unpublished MA
thesis. Boğaziçi University.
Karlsson, A. M. (2014) The Intonational phonology of Mongolian, in Sun-Ah Jun (ed.) Prosodic
Typology II: The Phonology of Intonation and Phrasing. Pp. 187-215.
Keane, E. (2007b). Investigating the intonational phonology of Tamil. Poster presented at the
satellite meeting of the 16
th
International Congress of Phonetic Sciences (ICPhS),
Saarbruecken, Germany.
163
Khan, S. D. (2008). Intonational phonology and focus prosody of Bengali. Unpublished Doctoral
Dissertation, University of California, Los Angeles.
Kidwai, A. (2000). XP-Adjunction in Universal Grammar. New York: Oxford University Press.
Kratzer, A. & Selkirk, E. O. (2007). Phase theory and prosodic spell-out: The case of verbs. The
Linguistic Review 24, 93-135.
Ladd, D. R. (1980). The structure of intonational meaning. Bloomington and London: Indiana
University Press.
Ladd, D. R. (1996/2008). Intonational Phonology. 1
st
and 2
nd
edition. Cambridge: Cambridge
University Press.
Lees, R. (1961). The phonology of modern standard Turkish. Bloomington, IN: Indiana
University.
Levi, S. (2005). Acoustic correlates of lexical accent in Turkish. Journal of the International
Phonetic Association 35, 73-97.
Lewis, G. (1967). Turkish grammar. Oxford: Oxford University Press.
Liberman, M. (1975). The intonational system of English. Doctoral Dissertation, MIT.
Liberman, M. & Prince, A. (1977). On stress and linguistic rhythm. Linguistic Inquiry 8, 249-
336.
Maeda, S. (1976). A characterization of American English intonation. Unpublished Doctoral
Dissertation, MIT.
Nash, R. (1973). Turkish intonation: An instrumental study. Mouton, The Hague.
Nespor, M. & Vogel, I. (1986). Prosodic phonology. Dordrecht: Foris.
Nielsen, K. (2005). Kiche Intonation. In UCLA Working Papers in Phonetics 104, 45-60.
Özsoy, S. (2004). Türkçe’nin yapısı; Sesbilim [The structure of Turkish: Phonology]. Istanbul,
Boğaziçi University.
Özge, U. (2003). A tune-based account of Turkish information structure. Masters’s thesis.
Middle East Technical University.
Peng, S.-h., Chan, M., Tseng, C.-y., Huang, T., Lee, O. J. & Beckman, M. (2005) Toward a Pan-
Mandarin System for Prosodic Transcription, in Sun-Ah Jun (ed.) Prosodic Typology:
The Phonology of Intonation and Phrasing, Oxford Univ. Press. pp. 230-270.
Pierrehumbert, J. (1980). The Phonology and Phonetics of English Intonation. Unpublished
Doctoral Dissertation, MIT.
164
Pierrehumbert, J. & Beckman, M. (1988). Japanese tone structure. MIT Press.
Rump, H. H., & Collier, R. (1996). Focus conditions and the prominence of pitch-accented
syllables. Language and Speech 39, 1-17.
Sadat-Tehrani, N. (2007). The intonational grammar of Persian. Doctoral Dissertation,
University of Manitoba.
Sasse, H. J. (1987). The ‘thetic’-‘categorical’ distinction revisited. Linguistics 25, 511-580.
Selkirk, E. (1984). Phonology and syntax: the relation between sound and structure. Cambridge,
MA: MIT Press.
Selkirk, E. (1986). On derived domains in sentence phonology. Phonology 3, 371-405.
Selkirk , E. (1995). Sentence prosody: Intonation, stress, and phrasing. The Handbook of
Phonological Theory, ed. by John Goldsmith. Cambridge, Mass.: Blackwell.
Sezer, E. (1981). On non-final stress in Turkish. Journal of Turkish Studies 5, 61-69.
Sluijter, A. & van Heuven, V. (1996). Spectral balance as an acoustic correlate of linguistic
stress. Journal of the Acoustical Society of America 100, 2471-2485.
Sluijter and van Heuven (1996a)
Stevens, N. (1989). On the quantal nature of speech. Journal of phonetics 17, 3-46.
Tabain, M. (2003a). Effects of prosodic boundary on /aC/ sequences: articulatory results. Journal
of the Acoustical Society of America 113: 516-531.
Tabain, M. & Perrier, P. (2005). Articulation and acoustics /i/ in pre-boundary position in
French. Journal of Phonetics 33: 77-100.
Thorsen, N. (1980). A study of the perception of sentence intonation -- Evidence from Danish.
Journal of the Acoustical Society of America 67(3), 1014-1030.
Turk, A. & Sawusch, J. (1996). The processing of duration and intensity cues to prominence.
Journal of the Acoustical Society of America 99, 3782-3790.
Üntak, T. A. (2006). Topic in syntax-phonology interface in Turkish: Sentential stress and
phases. Master’s thesis, Boğaziçi University.
Vallduvi, E. & Engdahl, E. (1996). The linguistic realization of information packaging.
Linguistics 34(3), 459-520.
Van der Hulst, H. & Van de Weijer, J. (1991). Topics in Turkish phonology. In Turkish
linguistics today, ed. by Hendrik Boeschoten & Ludo Verhoeven, 11-59. Leiden: Brill.
165
Viana, C. & Frota, S. (2007). Towards a P_ToBI. http://www.fl.ul.pt/dlgr/SonseMelodias/P-
ToBI/P-ToBI.htm (collaborators: Isabel Falé, Flaviana Fernandes, Isabel Marcarenhas,
Ana Isabel Mata, Helena Moniz & Marina Vigário)
Wightman, C. W., Shattuck-Hufnagel, S., Ostendorf, M., & Price, P. J. (1992). Segmental
durations in the vicinity of prosodic phrase boundaries. Journal of the Acoustical Society
of America 91: 1707-1717.
Xu, Y. & Xu, C. X. (2005). Phonetic realization of focus in English declarative intonation.
Journal of Phonetics 33, 159-197.
Zubizarreta, M. L. & Vergnaud, J. R. (2005). Phrasal stress, focus, and syntax. In M. Everaert &
H. van Riemsdijk(eds) The syntax companion. Cambridge: Blackwell.
Zubizarreta, M. L. (2014). Nuclear stress and information structure. The Oxford Handbook of
Information Structure, ed. by Caroline Féry & Shin Ishihara, Oxford: Oxford University
Press.
Abstract (if available)
Abstract
The aim of this study is to provide a phonological model of Turkish intonation using the framework of Autosegmental-metrical model of intonational phonology.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Prosody in contact: Spanish in Los Angeles
PDF
Prosodic recursion and syntactic cyclicity inside the word
PDF
Sound sequence adaptation in loanword phonology
PDF
Connecting phrasal and rhythmic events: evidence from second language speech
PDF
The prosodic substrate of consonant and tone dynamics
PDF
The planning, production, and perception of prosodic structure
PDF
Building adjectival meaning without adjectives
PDF
The Spanish feminine el at the syntax-phonology interface
PDF
Copy theory of movement and PF conditions on spell-out
PDF
Sources of non-conformity in phonology: variation and exceptionality in Modern Hebrew spirantization
PDF
Speech production in post-glossectomy speakers: articulatory preservation and compensation
PDF
A unified syntactic account of Mandarin subject nominals
PDF
Individual differences in phonetic variability and phonological representation
PDF
Tone gestures and constraint interaction in Sierra Juarez Zapotec
PDF
A reduplicative analysis of sentence modal adverbs in Spanish
PDF
Investigating the production and perception of reduced speech: a cross-linguistic look at articulatory coproduction and compensation for coarticulation
PDF
Articulatory knowledge in phonological computation
PDF
The locative alternation: unaccusative constructions and subject position
PDF
The case of a person: The person case constraint in German
PDF
Effects of speech context on characteristics of manual gesture
Asset Metadata
Creator
Ipek, Canan
(author)
Core Title
The phonology and phonetics of Turkish intonation
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Linguistics
Publication Date
04/22/2015
Defense Date
12/04/2014
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
intonation,OAI-PMH Harvest,prosodic typology,prosody,Turkish
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Zubizarreta, Maria Luisa (
committee chair
), Goldstein, Louis (
committee member
), Jun, Sun-Ah (
committee member
), Saltarelli, Mario (
committee member
)
Creator Email
canan.ipek@gmail.com,ipek@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-557156
Unique identifier
UC11299542
Identifier
etd-IpekCanan-3365.pdf (filename),usctheses-c3-557156 (legacy record id)
Legacy Identifier
etd-IpekCanan-3365.pdf
Dmrecord
557156
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Ipek, Canan
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
Tags
intonation
prosodic typology
prosody
Turkish