|
USC Phonetics and Phonology Group
|
|||||
homefacultycourseseventsphonetics
|
Open USC Phonetics and Phonology Group Home in new window.Research Projects at the USC Phonetics
Laboratory: Phonetics and Laboratory Phonology
|
||||
 Photo credit: Stacey Halper |
 |
SPAN: Speech Production and Articulation
kNowledge
Group
Real-Time Magnetic Resonance Imaging for Speech Production
Faculty: Shri Narayanan (USC EE,
Ling, CS), Krishna Nayak (USC EE), Sungbok Lee (USC
EE & Ling), Dani Byrd (USC Ling), Richard Leahy (EE)
Graduate Students: Abhinav Sethy (EE), Erik Bresch (EE), Stephen
Tobin (Ling), Jason Adams (CS MS '06)
Post-doc: Jon-Fredrik Nielsen (EE)
Undergraduates: Celeste de Freitas, David Hunt, Nathaniel Go
Funding: NIH R01DC007124
The long term goals of this project are to wed
state-of-the-art technology for imaging the vocal tract with a
linguistically informed analysis of the speech tasks or goals requisite
in the production of spoken language. Magnetic resonance imaging
(MRI) has served as a valuable tool
for studying static postures in speech production. Now, recent
improvements in temporal resolution are making it possible to examine
the dynamics of vocal tract shaping during fluent speech using
MRI. Our team has developed an approach for MRI image
reconstruction rates
of 24 images per second, making veridical real-time movies of speech
production
possible for the first time without X-rays (Narayanan, Nayak, Lee
Sethy,
& Byrd, to appear), providing us exquisite information about the
spatiotemporal
properties of speech gestures in both the oral and pharyngeal portions
of
the vocal tract. Our long-term goal is to understand the aspects
of
vocal tract shaping that are critically controlled during speech, both
for
sounds known to be complex in geometry (e.g., /r/ & sibilant
fricatives)
and for sounds known to be complex in their temporal structuring (e.g.,
/l/
& diphthongs). An understanding of vocal tract shape as a
fundamentally
dynamic aspect of linguistic organization will do much to add to the
field’s
current—basically static (i.e., postural & fixed
time-point)—approach
to describing the production of speech. An appropriate
understanding
of how sounds are produced in space and time is fundamentally a
phonological
question in that it bears directly on the phonological representation
of
segmental units, a representation that we take to be intrinsically
articulatory
and dynamic.
The published study below uses spiral k-space acquisitions with a low flip-angle gradient echo pulse sequence on a conventional GE Signa 1.5T CV/i scanner. This strategy allows for acquisition rates of 8-9 images per second and reconstruction rates of 20-24 images per second, making veridical movies of speech production now possible. Segmental durations, positions, and inter-articulator timing can all be quantitatively evaluated. Data show clear real-time movements of the lips, tongue, and velum. Sample movies and data analysis strategies are presented in the JASA article below and at sail.usc.edu/production/rtmri/jasa2004.
S. Narayanan, K. Nayak, S. Lee, A. Sethy & D. Byrd (2004) An approach to real-time magnetic resonance imaging for speech production. Journal of the Acoustical Society of America, 115, 1771-1776. [pdf]
E. Bresch, J. Adams, A. Pouzet,
S. Lee, D. Byrd, S. Narayanan (to appear). Semi-automatic processing of
real-time MR image sequences for speech production studies. 7th
International Seminar on Speech Production, Ubatuba, Brazil. [pdf]
Synchronized and noise-robust
audio recordings during realtime magnetic resonance imaging scans
Erik Bresch, Jon Nielsen, Krishna Nayak, and Shrikanth Narayanan
J. Acoust. Soc. Am. 120, 1791 (2006) [pdf]
Some of our SPAN team (Narayanan & Nayak, Byrd, Lee [left to right])
 |
|
Real time MRI video phonetics sounds tutorial at tbe SPAN website
An articulatory view of Kinyarwanda coronal harmony
Walker, Rachel, Dani Byrd, and Fidèle Mpiranya
Phonology 25, 499-535, 2008
Funding: NIH DC03172
Coronal harmony in Kinyarwanda causes alveolar fricatives to become postalveolarpreceding a postalveolar fricative within a stem. Alveolar and postalveolar
stops, affricates and palatals block coronal harmony, but the flap and non-coronal
consonants are reported to be transparent. Kinematic data on consonant production
in Kinyarwanda were collected using electromagnetic articulography.
The mean angle for the line defined by receivers placed on the tongue tip and
blade was calculated over the consonant intervals. Mean angle reliably distinguished
alveolar and postalveolar fricatives, with alveolars showing a lower tip
relative to blade. Mean angle during transparent non-coronal consonants showed
a higher tip relative to blade than in contexts without harmony, and the mean
angle during transparent [m] was not significantly different than during postalveolar
fricatives. This is consistent with a model where Kinyarwanda coronal
harmony extends a continuous tip-blade gesture, causing it to be present during
‘transparent’ segments, but without perceptible effect.
Underlying Invariance & Surface Variability in Speech Production: Modeling Phrasal Effects
Dani Byrd
(USC) and Elliot Saltzman (Haskins
Laboratories & Boston University)
Funding: NIH DC03172
We explore the puzzling juxtaposition of underlying invariance of control and surface variability in performance during speech production, and outline how a dynamical systems approach can contribute to solving this puzzle. Articulatory patterning at phrase edges is used as an example of how the surface expression of underlyingly invariant phonological units can vary in a linguistically principled way. We use computational simulations of these phrase boundary effects as prosodically-induced local temporal slowing. This slowing is generated by dynamical effects on the parameter specification of articulatory gestures. This focus allows us to examine a specific view of how underlying temporal characteristics of linguistic units can be modulated for communicative ends in the production of a particular utterance.
D. Byrd & E. Saltzman. (2003) The elastic phrase: Modeling the dynamics of boundary-adjacent lengthening. Journal of Phonetics, 31,2, pp 149-180. [pdf protected by Academic Press]]
Sound Patterns
in Language: The Relation Between Vowels and Syllable
Prominence
PI: Rachel Walker
Graduate Research Assistant:
Erika Varis
This project investigates how the relative prominence of a
syllable affects the way vowels are pronounced. Syllables with increased
prominence, such as stressed syllables, are often focal in vowel
patterns.
For example, many languages only show their full range of vowel
distinctions
in stressed syllables and they reduce vowel quality elsewhere. The
specific
questions for this study are: what is the range of vowel patterns
sensitive
to syllable prominence? and why are only certain relations between
vowels
and prominence attested? The project undertakes a typological
investigation
of prominence-sensitive vowel patterns, including vowel deletion, vowel
reduction, vowel/consonant metathesis, and vowel assimilation. A central
discovery is that these patterns avert vowel qualities that are
expressed
only in a non-prominent syllable. This gives rise to a hypothesis that
the
patterns have evolved so as to reduce perceptual difficulty. The
connected
theoretical proposal is that the patterns share a mental construct in
the
form of constraints that prevent vowel qualities from being expressed
in a
non-prominent position alone. A success of this model is its prediction
of a
common outcome by diverse processes across languages.
Funded by the Advancing Scholarship in the Humanities and Social
Sciences
Initiative, Office of the Provost, University of Southern California
Functional
data analysis of prosodic effects on articulatory timing
Sungbok Lee, Dani Byrd,
Jelena Krivokapic, Ben Parrell
Funding: NIH DC03172
An application of functional data analysis
(FDA)
(Ramsay & Silverman, 1997) for linguistic experimentation is
explored.
The time-warping function provided by FDA is shown to offer novel
advantages
in the investigation of articulatory timing. Traditionally,
articulatory
studies examining the effects of linguistic variables such as prosody
on
articulatory timing have relied on kinematic landmarks to define speech
intervals of interest. However, we present a novel approach that
allows
the analysis of the entire, continuous kinematic trajectories obtained
in
various experimental conditions, specifically, in the presence or
absence
of a phrase boundary. FDA time warping functions after alignment
of
test and reference (control) signals indicate slowing of articulator
movement
as the speech stream recedes from the phrase boundary. This is a
theoretically
predicted pattern (Byrd & Saltzman, 2003), which would be more
difficult
to validate with a traditional interval-based approach. However,
there
exists tokens for which FDA is problematic, and some potential remedies
are outlined. Despite certain limitations, generally, FDA is
shown
to be a useful tool for characterizing timing patterns in linguistic
experimentation
based on continuous kinematic trajectories.
Prosodic boundary gestures or “pi‐gestures” (Byrd and Saltzman, J. Phonetics, 2003) have been introduced to model the local slowing or lengthening of articulatory gestures in the vicinity of phrase boundaries. Computational modeling of articulatory dynamics is an important tool in assessing the predicted effects of pi‐gestures of varying boundary strength on constriction gestures in varying contexts. We simulate pi‐gestures within the TaDA task dynamics computational model [Nam and Kim, JASA, 116, 172–185 (2004)] and examine how functional data analysis can provide a tool for connecting articulatory lengthening with underlying pi‐gesture activation strength. Specifically, the model is applied to the articulatory synthesis of two sequences: [CV#CV] and [CVC#CV], where C is bilabial or alveolar. The pi‐gesture’s midpoint is coordinated synchronously with the midpoint of following consonant’s constriction gesture, and pi‐gesture activation strength and duration are manipulated. Results indicate that pi‐gesture activation strength has a much stronger effect on slowing than its duration. The slowing effect is asymmetrical, skewed earlier than the midpoint in the pi‐gesture interval. After removing linear‐time slowing effect (i.e., after length normalization), the slowing effect is slightly stronger in [CV#CV] than in [CVC#CV]. The strength of pi‐gesture also affects spatial articulatory characteristics depending on constriction location and sequential context.
S.
Lee, Benjamin Parrell, Dani Byrd. (2009) Computational
modeling of
juncture strength using articulatory synthesis of prosodic gestures.
Proceedings of the 157th Meeting, Portland, Oregon, 18-22 May 2009.
On
the temporal scope of phrase boundary effects
Dani Byrd, Jelena Krivokapic, Sungbok Lee
Funding: NIH DC03172
Acoustic
lengthening at prosodic boundaries is well explored, and the
articulatory bases for this lengthening are becoming better understood.
However, the temporal scope of prosodic boundary effects has not been
examined in the articulatory domain. The few acoustic studies examining
the distribution of lengthening indicate that boundary effects extend
from one to three syllables before the boundary, and that effects
diminish as distance from the boundary increases. This diminishment is
consistent with the pi-gesture model of prosodic influence [Byrd and
Saltzman, J. Phonetics 31, 149–180 (2003)]. The present experiment
tests the preboundary and postboundary scope of articulatory
lengthening at an intonational phrase boundary. Movement-tracking data
are used to evaluate durations of consonant closing and opening
movements, acceleration durations, and consonant spatial magnitude.
Results indicate that prosodic boundary effects exist locally near the
phrase boundary in both directions, diminishing in magnitude more
remotely for those subjects who exhibit extended effects. Small
postboundary effects that are compensatory in direction are also
observed.
<>
Cross-linguistic differences in prosodic organization
Emily Nava, Maria Luisa Zubizaretta, Louis Goldstein
NSF Grant BCS-0444088
USC Provost Grant: Advancing Scholarship in the Humanities and Social
Sciences
Languages differ in the realization and distribution of prominent
pitch events at the phrasal level, and we argue that these cross-linguistic
differences also reflect differences in language-specific organization at
the syllable level. The goal of the current project is to establish the
connection between events at the rhythmic level and pitch accents at the
phrasal level by detailing the hierarchy of prosodic organization for two
typologically different languages, Spanish and English. Additionally, the
project includes data from the speech of adult second language learners of
English whose first language is Spanish, which provides insight as to the
relevant stages of acquisition of native-like prosodic proficiency –in turn
contributing to linguistic theory as to the nature of the architecture of
prosodic organization. Data analyzed from the first of two experiments
reveal that English and Spanish differ regarding pitch accent placement for
certain information structure contexts, and that second language learners
initially transfer prosodic patterns from their first language. Data from a
second experiment analyzed using a forced alignment technique revealed that
vowel durations across lexical categories do not differ significantly in
Spanish but are significantly different in English. Additionally, those
second language speakers whose pitch accent placement was native-like in
English also had native-like vowel durations in English. Additional
experiments are currently in progress that examine organization at the foot
level in monolingual English and Spanish and second language English.
E. Nava & M.L. Zubizarreta. (In press). Deconstructing the Nuclear Stress
Algorithm: Evidence from Second Language Speech. The Sound patterns of
Syntax, eds. N. Erteschik-Shir & L. Rochman. Oxford University Press.
E. Nava & M.L. Zubizarreta. 2008. Prosodic Transfer in L2 Speech: Evidence
from Phrasal Prominence and Rhythm. Proceedings of Speech Prosody 2008, ed.
Plinio Barbosa, Sandra Madureira and Cesar Reis. Campinas, Brazil.
At the juncture of prosody, phonology, and phonetics—The interaction of phrasal and syllable structure in shaping the timing of consonant gestures.
Dani Byrd and Susie Choi
Presented at LabPhon 10 Paris 2006
Funding: NIHDC03172
Over the preceding nine Laboratory Phonology conferences, leading research in linguistic speech experimentation has clearly established that abstract phonological structure is directly reflected in the spatiotemporal details of speech production. In fact this softening of the clear severance between phonetics and phonology and an appreciation of the complexity of their relationship has been a prominent contribution of the LabPhon tradition. In recent LabPhons, the study of the phonetics-phonology interface has been extended to the phonetics-prosody interface, and we have, again, seen that abstract structural properties, this time of sentences rather than words, have a complex and rich effect on the articulatory details of speech production (see e.g., LabPhon 5). This has led to new theoretical conceptions regarding how phrase boundaries should be represented and to new computational models of their realization in speech (e.g., LabPhon 5 & 8, Byrd & Saltzman JPhon 2003). In the study described below, we move from two areas of laboratory phonology that are reasonably well-explored—the effects of syllable structure and of phrasal structure on articulatory overlap—to a consideration of how these phonological and prosodic structural properties interact in shaping articulatory timing.
It remains unknown how syllable and phrasal structure interact in determining intra- and intergestural coordination for consonant clusters preceding (in coda position), spanning (heterosyllabic), and following (in onset position) a phrase boundary. The prosodic gestural model of phrase boundaries (Byrd & Saltzman JPhon 2003) specifically predicts longer durations and less overlap for both consonants of the CC sequence due to its representation of a phrasal juncture as a prosodic gesture extending in time and causing a slowing of the clock that controls the pacing of gestural activation. However, this approach suggests that the strongest effect will be on the consonant most local to the boundary, i.e., on C2 for codas, C1 for onsets, and both consonants comparably in a heterosyllabic sequence. It further predicts that the durational and overlap effects will grade with the strength of the boundary.
We conducted an articulator movement tracking (EMMA) experiment with a design fully crossing: syllable position of a CC cluster (word-final, cross-word, and word-initial), adjacent/intervening phrase boundary type (word only, intermediate phrase, intonational phrase, and utterance), and three clusters ([sp], [sk], [kl]). Three subjects participated, and the tongue tip, lip aperture, and tongue rear were tracked for a total of 756 utterances. Gestural overlap was calculated both as absolute overlap (the time between peaks of C1 and C2) and relative overlap (the proportion of the way through C1 that the C2 peak occurred).
The results demonstrate significant main effects as well as two-way interactions of boundary type and syllable position on the duration of the individual consonants. The predictions were supported in that the heterosyllabic and coda sequences showed the larger phrasal effect on C2 duration and the onset and heterosyllabic sequences on C1 duration. Further, this lengthening graded with relative boundary strength such that four discrete degrees of lengthening are observed for the pooled data, and three or four for the individual subject data. Both absolute and relative intergestural overlap also exhibited two-way interactions of boundary type and syllable position. Cluster type was also a significant factor. In addition to intergestural overlap decreasing for clusters spanning a phrase boundary, clusters situated both before and after a boundary also showed decreases in overlap, though somewhat smaller. That is, prosodic structure had clear effects on word-internal segment-to-segment timing in words preceding and following a phrase boundary, in addition to timing effects across the boundary. These overlap changes are not, we show, a simple consequence of individual gestural lengthening. In fact, these types of timing changes are predicted by simulations within the prosodic gestural model of phrase boundaries. Finally, onset clusters were less overlapped than codas at every prosodic position (as predicted from early findings at the phrase-medial word-level.) There is also some suggestion in the data that onset clusters are more ‘resistant’ to prosodic perturbation; this cohesion is suggestive of the gestural molecule structure proposed by Browman and Goldstein (Phonetica 1992) for s-stop clusters.
In sum, the predictions of the prosodic gesture model regarding individual consonant lengthening in CC clusters are supported: both consonants lengthen, but the consonant gesture closest to the phrase boundary lengthens more. These effects grade with juncture strength. There is an interaction of phrase boundary type and syllable position on CC overlap such that while CC sequences spanning a boundary are most affected, CC sequences preceding and following a phrase juncture are also affected in a way that grades with juncture strength. This study extends the on-going efforts of the LabPhon community to understand the relation between abstract linguistic structure and low-level phonetic detail to an examination of the interaction of two different types of structure—syllabic and prosodic—in shaping articulatory timing. Simultaneously, it seeks to test and refine a theoretical model of the prosody-phonetics interface.
Interacting effects of syllable
and
phrase position on consonant articulation
Dani Byrd, Sungbok Lee, Jason Adams, Daylen Riggs
Funding: NIH DC03172
The complexities of how prosodic structure, both at the phrasal and syllable levels, shapes speech production have begun to be illuminated through studies of articulatory behavior. The present study contributes to an understanding of prosodic signatures on articulation by examining the joint effects of phrasal and syllable position on the production of consonants. Articulatory kinematic data were collected for five subjects using electromagnetic articulography (EMA) to record target consonants (labial, labiodental, & tongue tip), located in (1) either syllable final or initial position and (2) either at a phrase edge or phrase-medially. Spatial and temporal characteristics of the consonantal constriction formation and release were determined based on kinematic landmarks in the articulator velocity profiles. The results indicate that syllable and phrasal position consistently affect the movement duration; however, effects on displacement were more variable. For most subjects, the boundary-adjacent portion of the movement (constriction release for a pre-boundary coda and constriction formation for a post-boundary onset) are not differentially affected in terms of phrasal lengthening—both lengthen comparably.
Prosodic complexity and phrase length as factors in pause duration
<> Jelena KrivokapicResearch on influences on pauses has mainly focused on the impact of syntax, discourse and prosodic structure on the likelihood of pause occurrence and on the impact of syntactic structure on the duration of pauses within an utterance. Very little is known about what factors, apart from syntactic factors, play a role in determining the length of pauses between utterances or phrases. This experiment examines the effect of prosodic structure and phrase length on pause duration. Subjects read 24 English sentences varying along the following parameters: a) the length in syllables of the intonational phrase preceding and following the pause and b) the prosodic structure of the intonational phrase preceding and following the pause, specifically whether or not the intonational phrase branches into smaller phrases. In order to minimize variability due to speech rate and individual differences, speakers read sentences synchronously in dyads (Cummins 2002, Zvonik & Cummins 2002). The results show that length has a significant effect on pause duration both pre- and post-boundary for all dyads and that prosodic complexity has a significant post-boundary effect for some dyads. The possible reasons for the observed pause duration effects and the implications of these results on the question of incrementality in speech production are discussed.
<>Relating the performance and perception of phrasal boundaries
Jelena Krivokapic
This study examines the correspondence between the production of prosodic structure and perceptual judgments regarding prosodic boundary strength. In the production component of the study, 3 speakers read sentences in which one specific juncture is manipulated (varying in syntax, position in sentence, and phrase length) to elicit phrasal boundaries of differing strengths. Lengthening and pause duration at the boundary were measured. The same sentences, in written form, were presented to a second different group of 3 speakers who provided estimates of the strength of the target boundary on a scale of eight degrees. The results of the production and the estimation portions of the experiment demonstrate significant correlations between the production boundary strength as reflected in durational properties at the juncture and the boundary strength as estimated in the judging task. These correlations are roughly linear. Further, in both the production and perception domains, a range of boundary strength is exhibited rather than a small discrete set of boundary types. We also examine whether speakers’ own boundary strength estimates agree with their productions to a greater extent than estimates of other speakers.
Funding: NIH DC0317
Back to Top
A study of articulatory kinematics of American English diphthongs
Sungbok Lee and Dani ByrdDiphthongs are acoustically characterized by formant
movements between
initial and final vocalic segments [cf, Holbook and Fairbanks, J.
Speech Hear. Res., (1962); Thomas Gay, J. Acoust. Soc. Am., (1968)]. It
has been observed that the initial (onset) and final (offset) portions
may not correspond to typical monophthongal vowels and that there
exists more acoustic variability in the offset portions than in the
onset portions. Spectral transition rates have been found to be
different from diphthong to diphthong. In the current study, the
articulatory properties of American English diphthongs are investigated
in tandem with their acoustic properties. We collected both acoustic
and articulatory data of five diphthongs and seven monophthongs in
[b‐t] context embedded in a carrier sentence using electromagnetic
articulography. The data were recorded from two male and one female
subjects, and each diphthong was repeated five times. The main analysis
focus is on the kinematic characteristics of the three different tongue
sensors (tongue tip, tongue dorsum, and tongue rear) in each diphthong
production and the comparisons to monophthongs that are close to either
the onset portion or the offset portion of the given diphthong.
S. Lee, Dani Byrd. (2009) A study of articulatory kinematics of
American
English diphthongs. Proceedings of the 157th Meeting, Portland, Oregon,
18-22 May 2009. [pdf]
Supported by NIH
Looking for the Spanish Imperative Intonation
Sergio Robles-Puente
The intonation of both statements (Come la tarta / He eats the cake) and questions (¿Come la tarta? / Does he eat the cake?; ¿Quién come la tarta?/ Who eats the cake? ) in Spanish has been widely studied in the field. However, not much attention has been paid to the intonation of imperatives (¡Come la tarta! / Eat the cake!). Our knowledge about of these kinds of sentences is so limited that their status as primitives when it comes to Spanish intonation has even been called into question. Moreover, if we take into account that there is
regionally-conditioned intonational variation in other types of utterances like statements or questions, it is also expected that imperative commands may differ notably from Latin and Caribbean varieties to the Peninsular
ones. In order to get a clearer answer to the question of whether there is a Spanish imperative intonation or not, this study collects data not only to make more generalizations about imperatives in Spanish, but to compare the strategies speakers use both within and among the different Peninsular dialects.
Funding: Del Amo Research Award (Summer 2009)
Older Projects
<>The influence of phonemic vowel length on the voicing effect
Rebeka Campos-Astorkiza
This study tests the influence
of
phonemic vowel length on the realization of the voicing effect, i.e.,
the
phonetic process by which vowels tend to be longer before voiced
obstruents
than before voiceless ones. The literature on the voicing effect has
identified
a number of factors that influence the degree of this effect (Hussein
1994),
among them the presence of phonemic vowel length (e.g. Keating 1985).
However
there appear to be no published reports of experiments to test this
claim
about the influence of contrastive vowel length. In order to test the
hypothesis
that the presence of phonemic vowel length attenuates the voicing
effect,
it is necessary to isolate phonemic vowel length from other possible
conditioning
factors. This can be done by testing a language where length is
contrastive
for a subset of its vowel qualities, i.e., a language that has some
unpaired
vowel for the long-short contrast. The prediction is that the vowel
without
a short/long counterpart will exhibit a stronger voicing effect than
vowels
part of a long/short contrast. Lithuanian shows such an asymmetrical
system.
Lithuanian mid vowels lack a contrast for duration; they are always
long.
Thus, the hypothesis is that the voicing effect will be greater for
/e:,
o:/ than for the other vowels.
Acoustic data from native speakers of Lithuanian was collected. The
stimuli
consisted of bisyllabic non-sense words of the shape CV1C1C2V, where V1
could
be any of the Lithuanian vowels and the sequence C1C2 was either /kS/
or
/gZ/. The results show that the difference in vowel duration before
voiced
obstruents and before voiceless ones, i.e., the voicing effect, is
greatest
for /e:/ and /o:/ (p<.05), compared to the other vowels. Our
experiment
concludes that the vowels unpaired for length (/e:, o:/) are more
impacted
by the voicing effect. Vowels with a long/short counterpart are
influenced
to a lesser degree. This supports our hypothesis. More generally, this
conclusion
provides evidence for the influence of phonemic contrast on phonetic
realization,
previously discussed in relation to coarticulation (Manuel 1999) and
the
cues to stress (Berinstein 1978). Furthermore, the asymmetrical
Lithuanian
system suggests the importance of minimal contrast in the phonological
representation.
If a vowel differs from another vowel only in length, then it minimally
contrasts
for length. Our experiment shows that vowels minimally contrastive for
length
behave differently from vowels that do not minimally contrast for
length.
What
is raddoppiamento?
Length and prosody in Italian
Rebeka
Campos-Astorkiza
Raddoppiamento fono-sintattico in Italian has received much attention in the literature. However, the phenomenon seems to be far from explained and understood. Traditionally, Raddoppiamento refers to a lengthening process that affects word-initial consonants that follow a word ending in a stressed vowel. Furthermore, prosodic and syntactic constraints have been posed that prevent this process from taking place (Nespor 1977, 1979). Unfortunately, most of the analyses and conclusions regarding Raddoppiamento instances lack a solid empirical foundation. This project aims at shedding light by introducing considerations about the nature of the segments in the Raddoppiamento environment and different prosodic contexts. We consider not only consonants but also vowels as possible lengthened segments and examine whether their behavior patterns with that of consonants. Second, two different prosodic contexts are considered. The two relevant words are placed either phrase-internally or at the boundary of a phrase. According to previous analyses, Raddoppiamento is unexpected at the phrase boundary (Nespor 1977, Vogel 1977). Lastly, stressed and unstressed environments are tested.
The results showed that lengthening took place
in the traditional Raddoppiamento environment, i.e., when word1 ends in
stressed vowel and word2 begins with a consonant and there is no
intervening boundary between them. On the other hand, when the
initial segment in word2 was a vowel, this did not lengthen. This
result shows that any attempt at explaining the process must deal with
the fact that only consonants are subject to lengthening. As far as the
final vowel in word1 is concerned, this was significantly longer when
it carried the stress than when it was unstressed. Finally, the
presence of a boundary did not block the process categorically.
At the phrase juncture, the initial consonant in word2 was
significantly longer when the preceding vowel was stressed than when
the latter was unstressed. In the view of this empirical evidence, some
of the accounts of Raddoppiamento will have to be revisited in order to
accommodate the data.
J.
Acoust. Soc. Am. 116, 2645 (2004)
Some
novel allophonic and phonemic phenomena in Biscayan Basque
Rebeka Campos-Astorkiza
An acoustic study of novel allophonic and phonemic phenomena in the isolate language Basque is presented. The focus is on speakers of the Biscayan dialect. First, Basque shows a spirantization process by which voiced plosives are produced as approximants, particularly intervocalically. Interestingly, we find that Basque /ld/ sequences, where spirantization is not expected [Hualde (1991) Basque Phonology], are realized as a lateral approximant followed by a voiced lateral fricative. Second, in this variety of Basque, the historical three-way contrast among sibilants (two alveolars and one postalveolar) has been reduced to a two-way distinction. The original contrast, still found in other varieties, between a laminal alveolar and an apical alveolar has merged with different results depending on the continuancy of the sibilants. Third, Basque presents a contrast between trill and flap intervocalically. However, elsewhere this is neutralized, and the precise realization of this segment varies from trill to frication. Finally, the Basque five-vowel inventory allows for almost any sequence of two vowels. The same vowel sequence might be a diphthong (tautosyllabic) or a hiatus (heterosyllabic) depending on the lexical item. That is, diphthongs and hiatus are contrastive.
J. Acoust. Soc. Am. 118, 1901 (2005)ChAIM (Children and Adults Interacting with Machines)
a past project: See current CHIMP
website at SAIL Lab
Faculty: Shrikanth
Narayanan (USC Elec. Eng.), Dani Byrd (USC Linguistics), Elaine
Andersen (USC Linguistics; USC Neuroscience Program), Alex Potamianos
(Bell Labs)
Students: Suzanne Curtin, Laurie Gerber, Alison Bryant, Serdar
YeldrimUndergraduate
Students (past and present): Sudha Arunachalam (now at U Penn), Dylan
Gould, Abe Kazemzadeh (now a grad student at USC), Sonia Khurana
Funding: USC Integrated Multimedia Systems Center Undergraduate
Research Project (funded by NSF) and USC Provost's Office Undergraduate
Research Programs and Zumberge Interdisciplinary Grant
Enabling spoken language
capability as a part of immersive multimedia interfaces adds
naturalness and efficiency to human-machine interactions. Two crucial
requirements for multimedia human-machine interfaces are robustness
(i.e., minimal performance degradation under trying conditions) and
adaptabilty (i.e., responding to the user in a customized manner).
Undergraduate researchers, mentored by graduate students, are
investigating challenges to robustness & adaptability in the
field of automatic speech recognition: the challenge of recovering
from system errors made while interacting with capable adult users, and
the challenge of adapting to child-users. This is undertaken via
linguistically-informed data analysis of
two large databases of adults and children interacting with a
computer via spoken language.Back to Top
Phonetic foundations of final /s/ patterning in South-Central
Castilian Spanish
Ana Sanchez-Muñoz
Numerous studies have shown that in many
dialects of Spanish, the phoneme /s/ in final position may have several
realizations depending on a variety of factors (e.g. Lipski 1986;
Terrell 1979, 1981; Widdison 1995, 1996). However, there are few
analyses
of dialectal regions such as the area of South-West Central Castilian
Spanish, which because of its geographical location does not entirely
belong to any of the dialectal areas described for
Castilian. This study explores what factors may be having an effect
on the different realizations of final /s/. It considers aspiration
([h]), deletion, velarization ([x]), or realization of the sibilant
([s]). Data were collected from six native speakers under a controlled
task. This experiment aimed at production of final /-s/ in three
grammatical words (los, mis and estos/éstos), taking into
account the following factors: 1) The sound that follows the target
final /s/ (consonant or vowel); 2) Whether the target word is phrase
final or not; 3) Whether the target word carries focal accent or not.
The results show that the first two factors are
highly significant whereas the third one is not. It is furthermore
observed that certain realizations of final /s/ are restricted by the
type of sound that follows it, specifically for [x] after /k/. The
results show clear patterns of /s/ realization as [s] mainly occurs
before vowels and in prepausal position and [ø] mainly before
consonants. It is argued that s-lenition in Spanish can be explained in
terms of two variations in the gestural score: changes in magnitude and
overlap among gestures. These results help understand in greater depth
the mechanisms leading to the different realizations of syllable-final
/s/ in Spanish as characterized in the proposed hypothesis.
A three-way VOT contrast in
final position: Data from Armenian
Narineh Hacopian (now at Inquira)
Standard Eastern Armenian has a three-way VOT contrast among its oral stop series. While it is rare for such a 3-way contrast to be preserved in final position, Standard Eastern Armenian is claimed to do just this (Ladefoged & Maddieson 1996; Vaux 1998; Khachaturian 1988). This phenomenon provides an ideal opportunity to explore how prosodic structure influences the realization of a complex and delicate system of contrast maintained by temporal and saliency distinctions. The cues to this contrast, including VOT, closure duration, and burst amplitude, are examined in a variety of segmental and prosodic environments. The experiment evaluates effects of intonation phrase final, intermediate phrase final, word final and syllable final positions. We find that the 3-way VOT contrast is maintained in these prosodic domains in which the stop consonants are final. Speakers make significant VOT distinctions among the target consonants within the same boundary condition, and across prosodic conditions, larger prosodic domains exhibit longer VOT values.
(2003) Journal of the International Phonetic Association, 33(1) 51-80. [pdf]
Nasalization, neutralization, and merger in English front vowels
Jerry Liu (now at Google)
In American English, there exists a tense/lax distinction among the front vowels [i]/[I] and [e]/[E]. These often create minimal pairs in a number of different environments, however, before a velar nasal, the contrast is not preserved and is usually considered to have conflated to only the set of lax vowels. However, Ladefoged (2001) states that for the high front vowels “many younger Americans pronounce ‘sing’ with a vowel closer to that in ‘beat’ rather than to that in ‘bit.’” An acoustic study was conducted with nine young native Californians to examine whether raising occurs before the velar nasal, as has been anecdotally observed for the front high vowels, and if so, whether it also occurs for the front mid vowels, which have not previously been reported to raise from the lax. The results indicate that these subjects produce a vowel intermediate in formant frequencies (and sometimes, though not always, in duration) between the tense and lax vowel in the velar nasal environment. Further, this occurs for both the high and mid front vowels.
J.
Acoust. Soc. Am. 116, 2630 (2004)
Durational cues to focus in a scrambling language
Fetiye Karabay
This paper examines the phonetic structure of focus in Turkish. Focus is marked in different ways in different languages. Three most common ways of focus marking are via pitch perturbation, higher intensity and lengthening (O'Shaughnessy 1979, O'Shaughnessy & Allen 1983, Cooper et al 1985, Wells 1986). In many languages such as English, German, and Dutch, duration, amplitude, and pitch combine to give the effect of perceived stress in words that are in focus (Selkirk 1984, 1994; Gussenhoven 1983, 1994). Other languages, such as Danish and Spanish, use only pitch and amplitude as cues to focus, and, in these languages, durational effects play an unimportant role (Noteboom & Kryut 1987, Toledo 1989). Turkish is a scrambling language where the word order is flexible. While basic word order is SOV, almost all 6 permutations are possible in many sentences. In Turkish, broad focus is marked with word order as well as a pitch perturbation, typically a H tone. Generally, the accented word preceding the verb is in focus, unless the verb is in utterance initial position, in which case the verb itself is in focus. In narrow focus, however, the focused word does not immediately precede the verb. An experiment is conducted to examine whether focus in Turkish is marked by durational lengthening in addition to pitch perturbation and word order. The hypothesis that duration does not have an importance in broad focus due to the presence of the very salient pitch and word order cues is investigated. Further, we hypothesize that durational effects may play an important role in narrow focus since it is otherwise not as saliently marked in that it lacks the canonical word order cue. Data is collected from four native Turkish speakers in a controlled task. Stimuli include (i) four sentences, with phonetically identical subjects and objects, in which word order specifies broad focus and (ii) a second parallel set of sentences with narrow focus (i.e. in which the focused noun is not in canonical per-verb position) for comparison. Seven repetitions of each sentence are recorded and digitized. Phonetic lengthening in broad and narrow focus will be measured from the waveforms and statistically evaluated to confirm or refute the experimental hypotheses.
Stress effects on voicing and
frication of voiced obstruents in North-Central Spanish
Carolina Gonzalez (now at Florida State
University)
Funding: Del Amo Fellowship
In Spanish, /b, d, g/ are
usually spirantized to voiced approximants in all syllabic contexts
after a continuant sound. However, in North-Central Peninsular Spanish
(NCS), spirantization interacts with coda devoicing,
yielding voiceless fricatives. In the majority of cases, coda
/b, d, g/ occur in stressed syllables. This work examines whether
stress is a factor in the likelihood of frication and devoicing of coda
/b, d, g/ in this dialect. An acoustic study was conducted of nine
native speakers from NCS. These speakers were tested
on nonce words with /b, d, g/ in coda position in both stressed
and unstressed syllables. Measurements were made of vowel and
consonant duration, presence and absence of frication and voicing, and
voicing duration. The results show that frication is more likely in
stressed syllables than in unstressed syllables.
This suggests that in stressed syllables, a higher subglottal pressure
produces higher airflow across the glottis, which favors frication. In
turn, frication inhibits voicing due to conflicting aerodynamic
requirements between the two. We conclude that stress is a factor in
spirantization, and that it may indirectly affect
the voicing properties of /b, d, g/.
C. Gonzalez. (2002) Phonetic variation in voiced obstruents
in North-Central Penninsular Spanish. Journal of the
International Phonetic Association.32(1). [pdf]
Understanding the production
of English word stress in speaking
Dani Byrd
Teruhiko Fukaya (now at Sugiyama Jogakuen
University)
Funding: James H. Zumberge Individual Research Grant Proposal Category
I: New Faculty Start-Up
This project investigates the relation between the spatiotemporal orchestration of the vocal movements used to produce speech and a particular aspect of linguistic structure—word stress. The proposal describes a study employing a previously collected articulatory database of nearly a thousand spoken productions of words differing only in their stress pattern. This database was collected using a magnetometer system to track articulator movement during speech and includes data on tongue, lip and jaw trajectories for three subjects. The kinematic characteristics of these movements will be analyzed as a function of stress condition and phrasal prominence. The proposed study will profile the manner in which articulatory behavior is shaped by the linguistic dimension of word stress.
An articulatory examination of word-final
flapping at phrase edges and interiors
Dani Byrd
Teruhiko Fukaya (now at Sugiyama Jogakuen
University)
Funding: NIH
Formulations
of
flapping as a symbolic phonological rule suggest clear articulatory
differences between flaps and stops, and often offer no overt
explanation for why phrase boundaries should block the
alternation. The present study explores the articulatory
foundation of the distinction between flaps and non-flaps in word-final
position. We examine kinematic and acoustic data for these
articulations in phrase-final and -medial positions and in falling
and level stress contours. It is shown that a discrepancy exists
between acoustic and articulatory durational patterning while acoustic
durations of flaps are shorter than those of non-flaps overall, their
articulatory durations are not uniformly so.
It is important to consider multiple potential articulatory sources
both spatial and temporal for the acoustic shortness that characterizes
flaps, including spatial reduction, temporal articulatory shortening,
and changes in intergestural coordination.
The kinematic data indicate that different sources of flap shortness
exist for different speakers and different prosodic conditions.
These results imply that flaps are not categorically different from
non-flaps in the articulatory domain, as traditional formulations of
flapping as a symbolic phonological rule would suggest.
We conclude that the gradient variability in the spatiotemporal
patterning of tongue tip constrictions yields acoustic shortening
in flaps.
T.
Fukaya & D. Byrd (2005) An articulatory examination of word-final
flapping at phrase edges and interiors. Journal of the International
Phonetics Association 35, 1. [pdf]
Coarticulatory information in natural speech stimuli is crucial for infant recognition
of syllable sequences
Suzanne Curtin (USC Linguistics; now
at U of Pitts.), Toben Mintz (USC Psychology), Dani Byrd (USC
Linguistics)
Funding: NIH, SSHC, USC Zumberge Grant
This research was conducted in the USC Language Development
Lab (T. Mintz, Director).
Recent research has determined that coarticulatory information in speech provides important cues to early word segmentation. This experiment investigates whether 7-month old infants' ability to recognize a string requires the presence of appropriate coarticulatory information in the speech familiarization stream. Following familiarization to a string of CV syllables, infants were tested to determine if sequences that co-occurred in the familiarization string were preferred over those in which the syllables did not appear adjacently during familiarization. Further, the test phase was conducted so that the items had either appropriate or inappropriate coarticulation information. The results indicate that infants tested on items with appropriate coarticulation listened significantly longer to strings that had appeared during familiarization than to the appropriately coarticulated control strings that never occurred together during familiarization. Interestingly, when presented with inappropriate coarticulation test items, infants showed no preference for previously familiarized strings over the non-co-occurring syllable strings. We conclude that infants are sensitive to coarticulation in recognizing sequences in a speech stream. Furthermore, coarticulatory cues, in combination with other cues to segmentation, greatly enhance recognition of syllable sequences. These results suggest that coarticulation plays an important role in early word segmentation.
S. Curtin, T. H. Mintz, & D. Byrd. (2001) Coarticulatory cues enhance infants' recognition of syllable sequences in speech. BUCLD 25: Proceedings of the 25th Annual Boston University Conference on Language Development. [pdf]
Back to Top
Production
ease as a basis for
phonological coocccurence restrictions
Rachel Walker
Narineh Hacopian (now at Inquira)
Undergraduate Students: Gayane Arabyan, Mariko Taki
Funding: James H. Zumberge Individual Research Grant Proposal Category
I: New Faculty Start-Up
A widespread pattern is the
exclusion of similar sound elements in a word is known
as a phonological cooccurrence restriction. When word formation
produces a structure containing sounds that violate a cooccurrence
restriction, the phonological form of the word is altered in
one of two ways in order to obey the condition: the prohibited sounds
either dissimilate (become less similar) or assimilate (become
identical). We are investigating a hypothesis that bans on similar but
different elements have a foundation in psycholinguistic
processing. By conducting speech error experiments, we will investigate
the factors contributing to production ease in words containing
consonants that differ in voicing/nasality. The results will
have important implications for shaping a crosslinguistic typology of
voicing/nasality cooccurrence effects. Explaining phonological
cooccurrence restrictions in terms of maximizing production and
processing ease is an exciting new interdisciplinary research direction
combining linguistics and psychology. This research promises not only
to bring new understanding to the study of widespread cooccurrence
patterns, but also to mark a significant
advance in determining what universal factors underlie properties
of human language.
D. Byrd
Copyright © 2009 All rights reserved.
Information in this document is subject to change without notice.
Copyright Notice: The documents distributed here have been provided as a means to ensure timely dissemination of scholarly and technical work to individuals on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.