This study investigates consonant-related F0 perturbations (“CF0”) in French and Italian by compa... more This study investigates consonant-related F0 perturbations (“CF0”) in French and Italian by comparing the effects of voiced and voiceless obstruents on F0 to those of voiced sonorants. The voiceless obstruents /p f/ in both languages are found to have F0-raising properties similar to American English voiceless obstruents, while F0 following the (pre)voiced obstruents /b v/ in French and Italian patterns together with /m/, again similar to English [Hanson (2009). J. Acoust. Soc. Am. 125(1), 425–441]. In both languages, F0 is significantly depressed, relative to sonorants, during the closure for voiced obstruents, but cannot be differentiated from sonorants following the release of oral constriction. These findings are taken as support for a model on which F0 perturbations are fundamentally the result of laryngeal maneuvers initiated to sustain or inhibit phonation, regardless of other language-particular aspects of phonetic realization.
Journal of the Southeast Asian Linguistics Society, 2017
This paper presents an acoustic and perceptual study of the r>h shift in the variety of Khmer spo... more This paper presents an acoustic and perceptual study of the r>h shift in the variety of Khmer spoken in Giồng Riềng district, Kiên Giang province, Vietnam. In Phnom Penh Khmer, /r/ is realized as [h] in syllable onsets and onset clusters, and accompanied by lowered pitch, breathiness, and in some cases a change in the quality of the following vowel. In Kiên Giang Khmer, the r>h shift is accompanied by pitch lowering, but without changes in aspiration or vowel quality, and spectral measures did not indicate substantial differences in voice quality. Consistent with their productions, users of this dialect appear to rely solely on differences pitch to identify these lexical items. We discuss the implications of our findings for Khmer dialectology, mechanisms of sound change, and variation in the realization of rhotics more generally.
We examine the degree of correspondence between musical and linguistic tone sequences in a corpus... more We examine the degree of correspondence between musical and linguistic tone sequences in a corpus of 20 Vietnamese popular songs. Our data suggest that text-setting constraints in Vietnamese, as in some other Asian tone languages, are based primarily on the direction of the pitch transition from one syllable/note to the next. Borrowing some musical terminology, we may say that ‘similar motion’ is favoured, ‘oblique motion’ is allowed in certain cases, and ‘contrary motion’ is disfavoured. As with other Asian tone languages, the definition of these three types of motion depends on the categorisation of the linguistic tones; our current best model achieves a rate of 77% similar motion. We hypothesize that avoidance of contrary motion may be as or more important than achieving correspondence between tonal and melodic transitions in Vietnamese popular song.
Journal of the International Phonetic Association, 2017
The Tai dialect spoken in Cao Bằng province, Vietnam, is at an intermediate stage between tonal r... more The Tai dialect spoken in Cao Bằng province, Vietnam, is at an intermediate stage between tonal register split and the accompanying transphonologization of a voicing contrast into a dual-register tone system. While the initial sonorants have completely lost their historical voicing distinction and developed a six-way tonal contrast, the obstruent series still preserves the original voicing contrast, leaving the tonal split incomplete. This paper presents the first acoustic study of tones and onsets in Cao Bằng Tai. Although f0, VOT, and voice quality were all found to play a role in the system of laryngeal contrasts, the three speakers considered varied in terms of the patterns of acoustic cues used to distinguish between onset types, particularly the breathy voiced onset /b̤ /. From the diachronic perspective, our findings may help to explain why the reflex of modal prevoiced stops (*b) can be either aspirated or unaspirated voiceless stops.
In this chapter, we address the role of contact in the evolution of tone in mainland Southeast As... more In this chapter, we address the role of contact in the evolution of tone in mainland Southeast Asia (MSEA). We present an overview of the phonetic, phonological, and genetic characteristics of MSEA tone systems, emphasizing the rich variability of tonal realization found in the region. Next, we discuss the ways in which languages can become tonal, reviewing evidence for the spread of tone through contact as well as for the idea that much of the observed tonality on the ground in modern MSEA might be traced to a small number of ‘tonogenetic events’ rather than a large number of borrowings. In light of this discussion, we consider whether a re-evaluation of the notion of tone as a canonical indicator of ‘linguistic area’ more generally is warranted.
Proceedings of the 18th International Congress of Phonetic Sciences, 2015
We report new experimental evidence on consonant-induced F0 perturbations in two languages with p... more We report new experimental evidence on consonant-induced F0 perturbations in two languages with prevoiced stops, French and Italian. A positive correlation between duration of voicing lead and F0 at the onset of post-release voicing is observed, consistent with the predictions of an automatic or biomechanical account of the source of this effect. While the findings do not strictly rule out a role for onset F0 as a controlled enhancement, they support the proposal that, if anything, the enhancement is of [-voice] or [stiff] rather than [+voice].
Proceedings of the 18th International Congress of Phonetic Sciences, 2015
Madurese is a language with a three-way laryngeal contrast and an unusual consonant-vowel co- occ... more Madurese is a language with a three-way laryngeal contrast and an unusual consonant-vowel co- occurrence restriction. We provide new data on the phonetic realisation of Madurese stops from a sample of 15 native speakers by examining VOT, f0 and two acoustic correlates of voice quality, H1*-H2* and H1*-A3*. Our data indicate that while f0 distinguishes voiced from voiceless (aspirated and unaspirated) stops, at least one voice quality measure contrasts voiced and voiceless aspirated stops with voiceless unaspirated stops, suggesting that the relationship between these features may be more complex than has previously been assumed. Madurese appears to be best described as ‘register system’ of the Mon-Khmer type, albeit one in which pitch and voice quality are dissociated.
Languages of Southeast Asia: The State of the Art, Mar 2015
Mainland Southeast Asia (MSEA) is often described as the quintessential Sprachbund in which langu... more Mainland Southeast Asia (MSEA) is often described as the quintessential Sprachbund in which languages belonging to different language families converge as a result of contact. In this paper, we look in detail at the evidence for convergence of a specific phonological feature, tone, as expressed by two of its phonetic correlates, pitch and voice quality. Based on a database of 197 languages and dialects, we assess the extent of tonal diversity in MSEA languages and construct a statistical model of the degree to which tonal inventories can be predicted on the basis of geographic proximity, genealogical relatedness and population size. We find that the most robust predictors of tonality in MSEA languages are family and word shape.
10th International Seminar on Speech Production, 2014
Onset clusters in Khmer (Cambodian) often appear with an acoustic transition between consonants, ... more Onset clusters in Khmer (Cambodian) often appear with an acoustic transition between consonants, but the phonological status of these elements is indeterminate. If transitions result from gestural separation, they may disappear in fast speech. Acoustic analysis of data from 10 speakers shows that vocalic transitions in Khmer are found in largely predictable set of consonantal contexts. While their presence is modulated by speech rate, they never disappear completely, in some cases becoming more rather than less frequent in fast speech. Clusters contain- ing transitions are generally longer in duration than those that do not, and are also longer than monosyllables containing a lexical schwas, but the transitions do not show any spectral evidence of a distinct gestural target. The possible interpretations of these findings are discussed in the context of the range of articulatory variation known to occur in the implementation of speech rate.
Unlike many languages of Southeast Asia, Khmer (Cambodian) is not a tone language. However, in th... more Unlike many languages of Southeast Asia, Khmer (Cambodian) is not a tone language. However, in the colloquial speech of the capital Phnom Penh, /r/ is lost in onsets, reportedly supplanted by a range of other acoustic cues such as aspiration, a falling- or low-rising f0 contour, breathy voice quality, and in some cases diphthongization, e.g. /krɑː/ ‘poor’ > [kɔ̀ɑ], [kʰɔ̌ɑ], [kɔ̤ɑ̤], /kru:/ ‘teacher’ > [khùː] [kʰǔː], [kṳː]. This paper presents the results of production and perception studies designed to shed light on this unusual sound change. Acoustic evidence shows that colloquial /CrV/ forms differ from reading pronunciation forms in terms of VOT, f0, and spectral balance measures, while a pair of perceptual studies demonstrate that f0 is a sufficient cue for listeners to distinguish underlying /CrV/-initial from /CV/-initial forms, but that F1 is not. I suggest that this sound change may have arisen via the perceptual reanalysis of changes in spectral balance, coupled with the coarticulatory influence of the dorsal gesture for /r/.
In the colloquial Phnom Penh dialect of Khmer (Cambodian), lexical use of F0 is emerging together... more In the colloquial Phnom Penh dialect of Khmer (Cambodian), lexical use of F0 is emerging together with an intermediate VOT category and breathy phonation following the loss of /r/ in onsets (e.g. /kruː/ ‘teacher’ > [khṳ̀ː]). I show how this incipient tonogenesis might arise in a series of computational simulations tracing the evolution of multivariate phonetic category distributions in a population of ideal observers. Acoustic production data from a fieldwork study conducted in Phnom Penh was used as the starting point for the simulations. After establishing that the basic framework predicted relative stability over time, two possible responses to a phonetic production bias were considered: one in which agents correctly identified the source of (and thereby compensated for) the effects of the bias, and one in which agents misattributed the acoustic effects of the bias as a property of the onset. Good qualitative fits to the empirical production data were found for the latter group of learners, while the outcome for compensating learners resembled production data from a related dialect. These results are consistent with the sudden and discontinuous nature of many sound changes, and suggest that what appear to be enhancement effects may also emerge under different assumptions about the number of cue dimensions accessible to or deemed relevant by the learner.
Proceedings of the 35th Annual Conference of the Cognitive Science Society, 2013
We consider the problem of language evolution in a population setting, focusing on the case of co... more We consider the problem of language evolution in a population setting, focusing on the case of continuous parameter learning. While theories of phonetic change tend to emphasize the types of transmission errors that could give rise to a shift in pronunciation norms, it is challenging to develop a model that allows for both stability as well as change. We model the acquisition of vowel-to-vowel coarticulation in both single- and multiple-teacher settings, considering progressively more restrictive prior learning biases. We demonstrate that both stability and change are possible at the population level, but only under fairly strong assumptions about the nature of learning and production biases.
This chapter argues for the role of probabilistic enhancement in phonologization through computat... more This chapter argues for the role of probabilistic enhancement in phonologization through computational simulation of an ongoing sound change in Seoul Korean. Two challenges faced by a phonologization model of sound change are addressed: explaining which cues are selected for phonologization, and explaining why phonologization is often accompanied by dephonologization. It is proposed that cues are targeted for enhancement as a probabilistic function of their statistical reliability in signaling a contrast. Simulation results using empirically derived cue values are taken to support the idea that loss of contrast precision may drive the phonologization process.
This paper provides a detailed sketch of the main phonetic and phonological characteristics of th... more This paper provides a detailed sketch of the main phonetic and phonological characteristics of the Hanoi variety of Northern Vietnamese.
This study investigated the perceptual dimensions of tone in Vietnamese and the effect of dialect... more This study investigated the perceptual dimensions of tone in Vietnamese and the effect of dialect experience on listener’s prelinguistic perception of tone. While Northern Vietnamese tones are cued by a combination of pitch and voice quality, Southern Vietnamese tones are purely pitch based. 30 listeners from two Vietnamese dialects (10 Northern, 20 Southern) participated in a speeded AX discrimination task using northern stimuli. The resulting reaction times were used to compute an INDSCAL multidimensional scaling solution and were submitted to hierarchical clustering analysis. While the analysis revealed a similar three-dimensional perceptual space structure for both listener groups, corresponding roughly to f0 offset, voice quality, and contour type, the relative salience of these dimensions varied by dialect: Southern listeners were more likely to confuse tones produced with nonmodal voice quality, whereas Northern listeners found tones with similar pitch excursions to be more confusable. The results of hierarchical clustering of the stimuli further support an analysis where low-level perceptual similarity is influenced by primary dialect experience.
This paper explores the learnability of covert contrasts (impressionistically homophonous categor... more This paper explores the learnability of covert contrasts (impressionistically homophonous categories that can be reliably distinguished at the phonetic level) through a series of model-based clustering simulations using human production data. Allowing the models to learn both the number and parameters of those categories provides a way to explore the potential stability of category structures. The results indicate that while a statistical learner can be quite effective at inducing covert contrasts, success depends crucially on the number and distributional characteristics of the relevant cue dimensions.
This paper describes a probabilistic model of phonetic category merger. By modeling phonetic cate... more This paper describes a probabilistic model of phonetic category merger. By modeling phonetic categories as Gaussian mixtures, merger may be understood as a model selection problem: a distinction between two categories is predicted to collapse when the optimal model contains just a single mixture component. Intuitively, two categories should merge when their distributions overlap along some number of dimensions; the questions are how much, and in how many dimensions? The results indicate that while a statistical learner can be quite effective at inducing extremely subtle distinctions, providing an existence proof that covert contrasts can persist in the acquisition of phonetic categories, the learner’s success depends heavily on the number and distributional characteristics of the relevant cue dimensions.
This study investigates consonant-related F0 perturbations (“CF0”) in French and Italian by compa... more This study investigates consonant-related F0 perturbations (“CF0”) in French and Italian by comparing the effects of voiced and voiceless obstruents on F0 to those of voiced sonorants. The voiceless obstruents /p f/ in both languages are found to have F0-raising properties similar to American English voiceless obstruents, while F0 following the (pre)voiced obstruents /b v/ in French and Italian patterns together with /m/, again similar to English [Hanson (2009). J. Acoust. Soc. Am. 125(1), 425–441]. In both languages, F0 is significantly depressed, relative to sonorants, during the closure for voiced obstruents, but cannot be differentiated from sonorants following the release of oral constriction. These findings are taken as support for a model on which F0 perturbations are fundamentally the result of laryngeal maneuvers initiated to sustain or inhibit phonation, regardless of other language-particular aspects of phonetic realization.
Journal of the Southeast Asian Linguistics Society, 2017
This paper presents an acoustic and perceptual study of the r>h shift in the variety of Khmer spo... more This paper presents an acoustic and perceptual study of the r>h shift in the variety of Khmer spoken in Giồng Riềng district, Kiên Giang province, Vietnam. In Phnom Penh Khmer, /r/ is realized as [h] in syllable onsets and onset clusters, and accompanied by lowered pitch, breathiness, and in some cases a change in the quality of the following vowel. In Kiên Giang Khmer, the r>h shift is accompanied by pitch lowering, but without changes in aspiration or vowel quality, and spectral measures did not indicate substantial differences in voice quality. Consistent with their productions, users of this dialect appear to rely solely on differences pitch to identify these lexical items. We discuss the implications of our findings for Khmer dialectology, mechanisms of sound change, and variation in the realization of rhotics more generally.
We examine the degree of correspondence between musical and linguistic tone sequences in a corpus... more We examine the degree of correspondence between musical and linguistic tone sequences in a corpus of 20 Vietnamese popular songs. Our data suggest that text-setting constraints in Vietnamese, as in some other Asian tone languages, are based primarily on the direction of the pitch transition from one syllable/note to the next. Borrowing some musical terminology, we may say that ‘similar motion’ is favoured, ‘oblique motion’ is allowed in certain cases, and ‘contrary motion’ is disfavoured. As with other Asian tone languages, the definition of these three types of motion depends on the categorisation of the linguistic tones; our current best model achieves a rate of 77% similar motion. We hypothesize that avoidance of contrary motion may be as or more important than achieving correspondence between tonal and melodic transitions in Vietnamese popular song.
Journal of the International Phonetic Association, 2017
The Tai dialect spoken in Cao Bằng province, Vietnam, is at an intermediate stage between tonal r... more The Tai dialect spoken in Cao Bằng province, Vietnam, is at an intermediate stage between tonal register split and the accompanying transphonologization of a voicing contrast into a dual-register tone system. While the initial sonorants have completely lost their historical voicing distinction and developed a six-way tonal contrast, the obstruent series still preserves the original voicing contrast, leaving the tonal split incomplete. This paper presents the first acoustic study of tones and onsets in Cao Bằng Tai. Although f0, VOT, and voice quality were all found to play a role in the system of laryngeal contrasts, the three speakers considered varied in terms of the patterns of acoustic cues used to distinguish between onset types, particularly the breathy voiced onset /b̤ /. From the diachronic perspective, our findings may help to explain why the reflex of modal prevoiced stops (*b) can be either aspirated or unaspirated voiceless stops.
In this chapter, we address the role of contact in the evolution of tone in mainland Southeast As... more In this chapter, we address the role of contact in the evolution of tone in mainland Southeast Asia (MSEA). We present an overview of the phonetic, phonological, and genetic characteristics of MSEA tone systems, emphasizing the rich variability of tonal realization found in the region. Next, we discuss the ways in which languages can become tonal, reviewing evidence for the spread of tone through contact as well as for the idea that much of the observed tonality on the ground in modern MSEA might be traced to a small number of ‘tonogenetic events’ rather than a large number of borrowings. In light of this discussion, we consider whether a re-evaluation of the notion of tone as a canonical indicator of ‘linguistic area’ more generally is warranted.
Proceedings of the 18th International Congress of Phonetic Sciences, 2015
We report new experimental evidence on consonant-induced F0 perturbations in two languages with p... more We report new experimental evidence on consonant-induced F0 perturbations in two languages with prevoiced stops, French and Italian. A positive correlation between duration of voicing lead and F0 at the onset of post-release voicing is observed, consistent with the predictions of an automatic or biomechanical account of the source of this effect. While the findings do not strictly rule out a role for onset F0 as a controlled enhancement, they support the proposal that, if anything, the enhancement is of [-voice] or [stiff] rather than [+voice].
Proceedings of the 18th International Congress of Phonetic Sciences, 2015
Madurese is a language with a three-way laryngeal contrast and an unusual consonant-vowel co- occ... more Madurese is a language with a three-way laryngeal contrast and an unusual consonant-vowel co- occurrence restriction. We provide new data on the phonetic realisation of Madurese stops from a sample of 15 native speakers by examining VOT, f0 and two acoustic correlates of voice quality, H1*-H2* and H1*-A3*. Our data indicate that while f0 distinguishes voiced from voiceless (aspirated and unaspirated) stops, at least one voice quality measure contrasts voiced and voiceless aspirated stops with voiceless unaspirated stops, suggesting that the relationship between these features may be more complex than has previously been assumed. Madurese appears to be best described as ‘register system’ of the Mon-Khmer type, albeit one in which pitch and voice quality are dissociated.
Languages of Southeast Asia: The State of the Art, Mar 2015
Mainland Southeast Asia (MSEA) is often described as the quintessential Sprachbund in which langu... more Mainland Southeast Asia (MSEA) is often described as the quintessential Sprachbund in which languages belonging to different language families converge as a result of contact. In this paper, we look in detail at the evidence for convergence of a specific phonological feature, tone, as expressed by two of its phonetic correlates, pitch and voice quality. Based on a database of 197 languages and dialects, we assess the extent of tonal diversity in MSEA languages and construct a statistical model of the degree to which tonal inventories can be predicted on the basis of geographic proximity, genealogical relatedness and population size. We find that the most robust predictors of tonality in MSEA languages are family and word shape.
10th International Seminar on Speech Production, 2014
Onset clusters in Khmer (Cambodian) often appear with an acoustic transition between consonants, ... more Onset clusters in Khmer (Cambodian) often appear with an acoustic transition between consonants, but the phonological status of these elements is indeterminate. If transitions result from gestural separation, they may disappear in fast speech. Acoustic analysis of data from 10 speakers shows that vocalic transitions in Khmer are found in largely predictable set of consonantal contexts. While their presence is modulated by speech rate, they never disappear completely, in some cases becoming more rather than less frequent in fast speech. Clusters contain- ing transitions are generally longer in duration than those that do not, and are also longer than monosyllables containing a lexical schwas, but the transitions do not show any spectral evidence of a distinct gestural target. The possible interpretations of these findings are discussed in the context of the range of articulatory variation known to occur in the implementation of speech rate.
Unlike many languages of Southeast Asia, Khmer (Cambodian) is not a tone language. However, in th... more Unlike many languages of Southeast Asia, Khmer (Cambodian) is not a tone language. However, in the colloquial speech of the capital Phnom Penh, /r/ is lost in onsets, reportedly supplanted by a range of other acoustic cues such as aspiration, a falling- or low-rising f0 contour, breathy voice quality, and in some cases diphthongization, e.g. /krɑː/ ‘poor’ > [kɔ̀ɑ], [kʰɔ̌ɑ], [kɔ̤ɑ̤], /kru:/ ‘teacher’ > [khùː] [kʰǔː], [kṳː]. This paper presents the results of production and perception studies designed to shed light on this unusual sound change. Acoustic evidence shows that colloquial /CrV/ forms differ from reading pronunciation forms in terms of VOT, f0, and spectral balance measures, while a pair of perceptual studies demonstrate that f0 is a sufficient cue for listeners to distinguish underlying /CrV/-initial from /CV/-initial forms, but that F1 is not. I suggest that this sound change may have arisen via the perceptual reanalysis of changes in spectral balance, coupled with the coarticulatory influence of the dorsal gesture for /r/.
In the colloquial Phnom Penh dialect of Khmer (Cambodian), lexical use of F0 is emerging together... more In the colloquial Phnom Penh dialect of Khmer (Cambodian), lexical use of F0 is emerging together with an intermediate VOT category and breathy phonation following the loss of /r/ in onsets (e.g. /kruː/ ‘teacher’ > [khṳ̀ː]). I show how this incipient tonogenesis might arise in a series of computational simulations tracing the evolution of multivariate phonetic category distributions in a population of ideal observers. Acoustic production data from a fieldwork study conducted in Phnom Penh was used as the starting point for the simulations. After establishing that the basic framework predicted relative stability over time, two possible responses to a phonetic production bias were considered: one in which agents correctly identified the source of (and thereby compensated for) the effects of the bias, and one in which agents misattributed the acoustic effects of the bias as a property of the onset. Good qualitative fits to the empirical production data were found for the latter group of learners, while the outcome for compensating learners resembled production data from a related dialect. These results are consistent with the sudden and discontinuous nature of many sound changes, and suggest that what appear to be enhancement effects may also emerge under different assumptions about the number of cue dimensions accessible to or deemed relevant by the learner.
Proceedings of the 35th Annual Conference of the Cognitive Science Society, 2013
We consider the problem of language evolution in a population setting, focusing on the case of co... more We consider the problem of language evolution in a population setting, focusing on the case of continuous parameter learning. While theories of phonetic change tend to emphasize the types of transmission errors that could give rise to a shift in pronunciation norms, it is challenging to develop a model that allows for both stability as well as change. We model the acquisition of vowel-to-vowel coarticulation in both single- and multiple-teacher settings, considering progressively more restrictive prior learning biases. We demonstrate that both stability and change are possible at the population level, but only under fairly strong assumptions about the nature of learning and production biases.
This chapter argues for the role of probabilistic enhancement in phonologization through computat... more This chapter argues for the role of probabilistic enhancement in phonologization through computational simulation of an ongoing sound change in Seoul Korean. Two challenges faced by a phonologization model of sound change are addressed: explaining which cues are selected for phonologization, and explaining why phonologization is often accompanied by dephonologization. It is proposed that cues are targeted for enhancement as a probabilistic function of their statistical reliability in signaling a contrast. Simulation results using empirically derived cue values are taken to support the idea that loss of contrast precision may drive the phonologization process.
This paper provides a detailed sketch of the main phonetic and phonological characteristics of th... more This paper provides a detailed sketch of the main phonetic and phonological characteristics of the Hanoi variety of Northern Vietnamese.
This study investigated the perceptual dimensions of tone in Vietnamese and the effect of dialect... more This study investigated the perceptual dimensions of tone in Vietnamese and the effect of dialect experience on listener’s prelinguistic perception of tone. While Northern Vietnamese tones are cued by a combination of pitch and voice quality, Southern Vietnamese tones are purely pitch based. 30 listeners from two Vietnamese dialects (10 Northern, 20 Southern) participated in a speeded AX discrimination task using northern stimuli. The resulting reaction times were used to compute an INDSCAL multidimensional scaling solution and were submitted to hierarchical clustering analysis. While the analysis revealed a similar three-dimensional perceptual space structure for both listener groups, corresponding roughly to f0 offset, voice quality, and contour type, the relative salience of these dimensions varied by dialect: Southern listeners were more likely to confuse tones produced with nonmodal voice quality, whereas Northern listeners found tones with similar pitch excursions to be more confusable. The results of hierarchical clustering of the stimuli further support an analysis where low-level perceptual similarity is influenced by primary dialect experience.
This paper explores the learnability of covert contrasts (impressionistically homophonous categor... more This paper explores the learnability of covert contrasts (impressionistically homophonous categories that can be reliably distinguished at the phonetic level) through a series of model-based clustering simulations using human production data. Allowing the models to learn both the number and parameters of those categories provides a way to explore the potential stability of category structures. The results indicate that while a statistical learner can be quite effective at inducing covert contrasts, success depends crucially on the number and distributional characteristics of the relevant cue dimensions.
This paper describes a probabilistic model of phonetic category merger. By modeling phonetic cate... more This paper describes a probabilistic model of phonetic category merger. By modeling phonetic categories as Gaussian mixtures, merger may be understood as a model selection problem: a distinction between two categories is predicted to collapse when the optimal model contains just a single mixture component. Intuitively, two categories should merge when their distributions overlap along some number of dimensions; the questions are how much, and in how many dimensions? The results indicate that while a statistical learner can be quite effective at inducing extremely subtle distinctions, providing an existence proof that covert contrasts can persist in the acquisition of phonetic categories, the learner’s success depends heavily on the number and distributional characteristics of the relevant cue dimensions.
Uploads
the acquisition of phonetic categories, the learner’s success depends heavily on the number and distributional characteristics of the relevant cue dimensions.
the acquisition of phonetic categories, the learner’s success depends heavily on the number and distributional characteristics of the relevant cue dimensions.