UC Berkeley PhonLab Annual Report

The Berkeley Phonetics Machine is a Linux virtual machine imageproduced and used by the UC Berkeley Phonology Lab asa platform for phonetic research. It contains a full data analysisstack based on Python and R and also specialized tools for phoneticresearch. The machine is designed as a flexible and productiveplatform for established and novel research agendas thatcan be easily shared and reproduced. We list the software availablein the machine, which includes many command-line toolsfor acoustic analysis and media file manipulation, as well asspecialized Python libraries. We also discuss the use of this machinein the Phonology Lab and in phonetics courses. The overallexperience with the machine has been positive, as facultyand graduate students are able to share and execute scripts ina common working environment. Undergraduate students haveless opportunity to master the virtual machine environment butbenefit from simplified instructions and fewer installation andoperating problems. The primary difficulty that we have encounteredhas been with a few underpowered student computersthat cannot run the virutual machine or do not run it well.

Lindblom et al. (1995) proposed two modes of listening to speech: a “what” mode, in whichlisteners focus on meaning, and a “how” mode, where listeners attend to details ofpronunciation. This theory fits with Hickok and Poeppel’s (2004, 2007) more recent dualstream model of speech perception. What conditions then are necessary for modulating the useof one listening mode or the other? Following observations concerning the effect of higher levellinguistic information on speech perception (Cole & Jakimik 1980, etc.), I will detail the resultsof two experiments which consider how structural and semantic context (word predictability)interact with the listener’s attention to phonetic details. The experiments use the phoneticaccommodation or imitation paradigm (Goldinger 1998, etc.) as a tool to determine whatphonetic details subjects noticed after hearing target words in a variety of contexts. The firstexperiment compares the degree of accommodation in isolated phrase vs. sentence context.The second experiment considers how the variable of word predictability within the context ofa sentence influences the degree of accommodation. The results suggest listeners attend moreclosely to sub-phonemic details of pronunciation when less structural and semantic context ispresent and that contextual predictability modulates phonetic attention.

Social and Structural Constraints on a Phonetically-Motivated Change in Progress: (str) Retraction in Raleigh, NC

The “many to one” problem arises when trying to map inversely from acoustic patterns tovocal tract configurations. In one famous demonstration, Atal et al. (1978) searched through theacoustic outputs of a synthetic vocal tract for vowels that matched each other exactly on the frequenciesof the first three resonances (F1-3) and found that for each vowel tested [i], [a] and [u] there wereseveral vocal tract configurations that gave the same formant frequencies. This result has been used toshow that one (whether speech technologist or listener) cannot inversely map from acoustics to derive aunique possible vocal tract shape. We synthesized vowels using the formant frequencies reported byAtal et al. and show that listeners can detect differences between them even though the vowels areidentical in the first three formants. Our conclusion is that listeners may not be as troubled by a manyto-oneproblem as has been assumed before.

On Reconstructing Tone in Proto-Niger-Congo

Social network structure, accommodation, and language change

Scalar tone shift as evidence for morphology without morphemes∗

Multiple Exponence in the Lusoga Verb Stem

The Autosegmental Approach to Tone in Lusoga

Previous research [1, 2] shows that articulatory variability is reduced for people with flatter palates. It has been hypothesized [1] that this is because the mapping between articulation and acoustics is more linear for flatter than for more domed palates. A combination of two synthesizers were used to model how vocal tract anatomy influences the mapping of articulation onto acoustics, using American English /r/ as a test case. A retroflexable tongue tip was added to the articulatory parameters. Two additional palate shapes and a sublingual cavity that appears during /r/ production were also added to the synthesizer. A Python script searched the articulatory-acoustic space for vocal tract configurations that resulted in a low F3 (the hallmark acoustic cue for /r/) for each palate. Palate shape influences not only the overall sensitivity of the articulatory-acoustic mapping, but also the effect of each individual articulatory parameter on F3.

Sounds that are contrastive in a language are rated by listeners as being more differentfrom each other than sounds that don’t occur in the language or sounds that areallophones of a single phoneme. The study reported in this paper replicates this findingand adds new data on the perceptual impact of learning a language with a new contrast.Two groups of speakers of the Nanjing dialect of Mandarin Chinese were tested. Onegroup was older and had not been required to learn standard Mandarin as school children,while the other younger group had learned standard Mandarin in school. Nanjing dialectdoes not contrast [n] and [l], while standard Mandarin does. Listeners rated the similarityof naturally produced non-words presented in pairs, where the only difference betweenthe tokens was the medial consonant. Pairs contrasting [n] and [l] were rated by olderNanjing speakers as if two [n] tokens or two [l] tokens had been presented, while thesesame pairs were rated by younger Nanjing speakers as noticeably different but not asdifferent as pairs that contrast in their native language.

Underlying Representations and Bantu Segmental Phonology

As it is well known, noun class prefixes are low tone in Narrow Bantu and classes 1, 3, 4, 6(a), 9,and 10 have nasals (Meeussen 1967). However, just outside Narrow Bantu, noun class prefixesare usually high tone and the nasals are typically missing. A dichotomy is found in GrassfieldsBantu where Eastern Grassfields resembles Narrow Bantu but the Ring and Momo sub-groups ofWestern Grassfields have high tone prefixes and lack nasals except sporadically. Drawing ondata from Babanki and other Ring languages, we show that this relationship is not accidental. Ina number of contexts where we expect a high tone prefix, a stem-initial NC cluster requires thatit rather be low. We provide some speculations in this paper as to why nasals should beassociated with low tone, an issue that has not been fully addressed in the literature on consonanttypes and tone.

Morphologically assigned accent and an initial three syllable window in Ese’eja

This ultrasound and acoustics study of American English /r/ and /s/ investigates whether variability in production is related to individual differences in vocal tract morphology, namely the shape of the palate. There was reduced articulatory variability for flatter palates for both /r/ and /s/. There was a relationship between acoustics and palate shape for /s/ only, where flatter palates had increased acoustic variability, but for /r/, there was no relationship between palate shape and acoustic variability.

The Effects of Singing on Speech in Geriatric Voice: An Acoustic Study

This work examines three- to six-year-old children’s acquisition of the Spanishpassive. This structure, a notoriously difficult concept for early learners, exhibits great variationin age of acquisition cross-linguistically. Spanish, with two passive constructions, is an ideal casestudy for the role of frequency in the development of the passive. This study utilizes data fromCHIEDE, a spontaneous oral corpus spanning more than 20,000 words of child speech. Only alimited number of studies examining the passive have utilized spontaneous corpus data; as aresult, it is unclear if lexical semantic patterns are due to experimental or task effect - an issuethat only the inclusion of natural data can resolve. Results show that children only produce oneof two possible forms of the Spanish passive. Their production is also limited to action verbs.Finally, while children as young as 3;0 produce the passive, cross-sectional data show thebeginnings of a downward U-shaped developmental pattern. These results are explained in termsof acquisition by analogy as children utilize previously-acquired structures to create abstractsyntactic representations.

In Search of Prosodic Domains in Lusoga

A phonetic study of the vowel spaces of 535 young speakers of Californian Englishshowed that participation in the California Vowel Shift, a sound change unique tothe West Coast region of the United States, varied depending on the speaker’s selfidentifiedethnicity. For example, the fronting of the pre-nasal hand vowel varied byethnicity, with White speakers participating the most and Chinese and South Asianspeakers participating less. In another example, Korean and South Asian speakers ofCalifornian English had a more fronted foot vowel than the White speakers. Overall,the study confirms that CVS is present in almost all young speakers of CalifornianEnglish, although the degree of participation for any individual speaker is variable onaccount of several interdependent social factors.

An Ultrasound Analysis of Tswefap Back Consonants