Phorum

The Berkeley Phonetics, Phonology and Psycholinguistics Forum ("Phorum") is a weekly talk and discussion series featuring presentations on all aspects of phonology, phonetics and psycholinguistics. We meet on Fridays from 3 to 4:30 pm. Phorum is organized by Katie Russell and Maks Dąbkowski. Our emails are respectively "krrussell" and "dabkowski" @berkeley.edu.


Spring 2023 Schedule

January 20

Sarang Jeong (Stanford): The relation between perception and production in an ongoing sound change: A pilot experiment on the younger group's perception of Korean three-way stop contrast

The purpose of this study is to investigate whether individual language users’ perception is parallel to their production when their language community is going through a sound change. When a sound change is in progress, do speakers with the new and old variants have perceptual maps that reflect their own respective variants? Or do they share similar perceptual maps despite having different variants in production? I will address this question by examining Korean stop contrast whose primary cue has been shifting from VOT to F0. I hypothesize that there will be a discrepancy between listeners’ perceptual maps and their production patterns. Through an auditory identification experiment, this preliminary study explores one of the predictions made by the hypothesis, which relates to the perception of the younger age group.

January 27

Mykel Brinkerhoff (UC Santa Cruz): Testing the Laryngeal Complexity Hypothesis: Evidence from Santiago Laxopa Zapotec

Most descriptions about the interaction between tone and phonation are based on South East and East Asian languages (e.g., Masica 1976, Thurgood 2002, Michaud 2012, Brunelle & Kirby 2016, Kuang 2017). These descriptions have lead to widespread claims about what is possible in languages with tone and phonation, specifically that tone and phonation are co-dependent. This means that certain tones will bear specific phonations or that certain phonation types only appear with certain tones. For example, Mandarin’s tone 3 is associated with creaky voice (Hockett 1947) and Vietnamese’s low falling tone is associated with breathy voice (Thurgood 2002). This, however, is not the true for Oto-Manguean languages where tone and phonation are independent from one another (Silverman 1997).
Despite being independent from one another, tone and phonation need to interact in a way that is perceptually salient. This need for perceptual saliency led to the Laryngeal Complexity Hypothesis (LCH; Silverman 1997, Blankenship 1997, 2002). The LHC’s basic premise is that for tone and phonation to be best perceived there needs to be an ordering or phasing between the gestures for tone and phonation. If there was no strict ordering of the gestures for tone and phonation then the cues for tone and/or phonation will be interrupted. DiCanio’s (2012) study exploring the LCH in Itunyoso Trique found exactly this interference. DiCanio showed that when there is a large overlap between the gestures for tone and phonation that the f0 signal is perturbed by the glottal gestures for phonation.This paper investigates the LCH’s role in Santiago Laxopa Zapotec (SLZ), an understudied Oto-Manguean language spoken by approximately 1000 people in the municipality of Santiago Laxopa in the Sierra Norte of Oaxaca, Mexico. 
Preliminary results of a generalized additive mix model show that the LCH is correct in arguing for the phasing between tone and phonation cues. 

February 3

No meeting

February 10

Reiko Kataoka (UC Davis): Effects of Lexical Frequency on Phonetic Recalibration

Lexically guided perceptual learning (LGPL) is a special case of perceptual recalibration for categorization of speech sounds, whereby upon repeated exposure to an ambiguous speech sound in a disambiguating lexical context listeners come to categorize that ambiguous sound in a way that is consistent with the lexical information (Norris, McQueen, & Cutler, 2003). Various factors that may affect this recalibration have been examined in the literature (see Samuel & Kraljik, 2009).

Building on the current body of evidence, this talk will present the study that examines the effect of lexical frequency on the extent of recalibration. Listeners were exposed, in a form of a lexical decision task, to an ambiguous sound [?], midway between [f] and [s], which was embedded in [s]-biasing high-frequency word frames (e.g., tenni[?]) for a high frequency (HF) group or low-frequency word frames (e.g., hubris[?]) for a low frequency (LF) group. After that exposure, all listeners engaged in a phonetic categorization task, classifying tokens from an [ɛf]-[ɛs] continuum either as [ɛf] (f-response) or [ɛs] (s-response). The HF group recognized the exposure words faster and more often than the LF group during the lexical decision task and both groups exhibited evidence of LGPL, but there was no significant group difference in the extent of recalibration. The results will be discussed as they relate to the effect of lexical frequency in perceptual learning in general and in LGPL more specifically.

February 17

No meeting

February 24

Nay San (Stanford): Improving access to language documentation corpora using self-supervised models for speech

Many hours of speech from language documentation projects often remain inaccessible to a variety of interested parties — from speakers within the language communities, to language teachers, to linguists. This inaccessibility largely stems from the fact that recording speech is far simpler and considerably less time consuming than annotating the speech with relevant information. While machine-assisted workflows have the potential to help speed up this annotation process, conventional speech processing systems trained in a supervised setting require many hours of annotated speech to achieve sufficient performance to yield any time savings over manual annotation — leaving many projects in a Catch-22. In this talk, we examine how recently developed self-supervised models for speech such as wav2vec 2.0 have dramatically decreased the amount of annotated data required to implement machine-assisted workflows in language documentation projects and report on the benefits and challenges of using them based on several case studies.

March 3

Julia Swan (San José State): Monophthongal /ow/ among Nordic Americans in Puget Sound: A Case of Functional Reallocation

When describing regional features of English and ethnolinguistic repertoires of English, sociolinguists often posit substrate influence from non-English languages. A feature may be introduced by speakers of a non-English community language and subsequently used by English dominant speakers with identity or stance motivations (Fought 2010). How exactly this happens has been difficult to trace and is rarely as simple as direct language transfer. Ethnolinguistic features may derive from the heritage language of the community but be reallocated to other sociolinguistic meanings by the second generation (Gnevsheva 2020). Older and younger individuals with the same generational status deploy ethnolinguistic variation differently depending on the timing of their birth relative to salient cultural events in the community (Sharma & Sankaran 2011). Features associated with ethnolinguistic repertoires may evolve to become markers of regional identity or local stances (Labov 1963). An analogous situation is observed in dialect contact where diverse varieties of the “same” language come into contact and the distinct features resulting from this contact are functionally reallocated to different linguistic environments over subsequent generations (Britain 1997, Trudgill 1985).

Monophthongal /ow/ is a notable feature of the English in Washington State, along with other parts of the Upper Midwest. The current acoustic analysis explores this feature in the interview speech of 30 first, second and third-generation Swedish and Norwegian immigrants to the Puget Sound area born in the 1920s to 1940s. The interviews were collected as part of the Nordic American Voices Oral History Project (NAV) at the Nordic Heritage Museum in Seattle, WA. Analyzing 4,623 /ow/ tokens using a trajectory length measure and visualizations of formant trajectories, this article describes the focusing and reallocation of monophthongal /ow/ among children and grandchildren of Nordic American immigrants. For diachronic perspective, these data are also compared to conversational data from Seattle young adults (not necessarily of Nordic heritage) collected in 2014-2015.

First generation immigrants who are native speakers of Norwegian and Swedish exhibit more monophthongal /ow/ as indicated by significantly shorter trajectory length measures and trajectory shape differences. Subsequent generations of Norwegian and Swedish Americans display allophonic re-functionalization determined by following voicing environment: monophthongal /ow/ is relegated to the pre-voiceless environment. A similar process of focusing and reallocation of dialect variants in contact led to a phonologized raising pattern in Fenland English (Britain and Trudgill 2003: 251). The pattern of monophthongal /ow/ has become a marker of English in Washington State (among other places). Comparison data from contemporary young adults suggests that this pattern remains robust in Seattle-area young adults born before about 1990, but may be declining among speakers born in the 1990s and later. This project documents the complexity of contact-induced language change and offers insight into the influence of non-English languages on the development of ethnolinguistic and regional varieties of English. The study provides a reminder that despite large-scale language shift, the children and grandchildren of immigrants may impact the salient linguistic features of English in the local environment and likewise deploy such features as markers of their local orientation and identities.

March 10

Amber Galvano (UC Berkeley): A Q-theoretic approach to cross-dialectal Spanish <st> production

The variably pre- and post-aspirated realization of Spanish <s> + voiceless stop (e.g. ‘pasta’ → [pa(h)t(h)a]) has become synonymous with Andalusian Spanish, spoken in southern Spain (e.g. Ruch & Peters, 2014; Torreira, 2006). Previous accounts (e.g. Parrell, 2012) within Articulatory Phonology (AP) (Browman & Goldstein, 1986) explain the Andalusian production patterns via gestural phasing and extensive coarticulatory overlap. In this talk, I will present spontaneous speech data from speakers of Western Andalusian, Buenos Aires, and North-Central Peninsular Spanish, each of which demonstrates a unique envelope of fine-grained phonetic variation in the intervocalic <st> context. Then, bearing this data in mind, I will introduce a novel representational approach based in Q-theory (Inkelas & Shih, 2013) which explains the cross-dialectal variation in production as variation in the linking of segments, subsegments, and marginal subsegments, in accordance with phonetic, social, and other potential constraints. I propose that this type of account offers a more fruitful synchronic model than AP alone, especially for prevalent “transitional” productions with both pre- and post-aspiration. This approach also makes useful diachronic predictions about how these sequences have evolved and may continue to do so.

March 17

Simon Todd (UC Santa Barbara): Building implicit linguistic knowledge through passive exposure

In this talk, I outline recent work that investigates how humans can build implicit lexical and phonotactic knowledge of a language they don't speak simply through being exposed to it regularly, and how this process may be affected by structural and social aspects. I look first at the implicit learning of Māori, the Indigenous language of Aotearoa New Zealand, by New Zealanders who are surrounded by it in everyday life but can't speak it. By applying computational modeling to experimental results, I show that non-Māori-speaking New Zealanders have a surprising amount of lexical and phonotactic knowledge of Māori, which is best explained by the assumption that they have an implicit memory store of approximately 1,500 distinct morphemes. Through consideration of machine learning of morphological segmentation, I describe how this learning is facilitated by the heavy use of compounding in Māori. I then turn to examine the implicit learning of Spanish, which has a much lower relative degree of compounding, in California and Texas. I show that Californians and Texans who don't speak Spanish have implicit lexical and phonotactic knowledge of it, much like non-Māori-speaking New Zealanders do of Māori, but that the form of this knowledge appears to be affected by the morphological differences between the languages. Furthermore, I show that the strength of implicit knowledge evidenced by non-Spanish-speaking Californians and Texans is affected by the attitudes they hold toward Spanish and its speakers, supporting previous work that demonstrates how listeners' attitudes affect not only the conscious actions they take in response to a speaker, but also the way they unconsciously process and represent speech.

March 24

No meeting (spring break)

March 31

No meeting (spring break)

April 7

Marie Tano (Stanford): Stancetaking and the construction of Black American identities in the United States 

This study accounts for variation found in the use of an ethnically-marked language variety spoken by a Nigerian-born Black American. Methodologically, it takes on a meticulous stance-based analysis regarding the use of specific phonetic features. Drawing on previous work which uses the intra-speaker paradigm to investigate the construction of identity in interaction (Podesva, 2007; Sharma, 2011), this paper examines the ways in which a US-based immigrant indexes the various identities with which they align. A qualitative analysis of self-recorded data revealed behavior distinct from previous literature on addressee-influenced style shifting, but consistent with literature on Diasporic linguistic repertoires that include a variety of standardized and ethnic linguistic varieties. Overall, the analysis demonstrates how a participant's social stances drives their linguistic behavior and allows for them to build and negotiate their identity categories (Rickford & McNair-Knox, 1994; Bell, 2006). 

April 14

Laura Kalin (Princeton)

April 21

Georgia Zellou (UC Davis)

April 28

Christopher Cox (Aarhus University)

May 5

No meeting (RRR week)


Fall 2022 Schedule

September 2

CJ Brickhouse (Stanford): Revisiting California’s apparent low-back merger: a lot of thoughts about LOT and THOUGHT

Since Moonwomon (1991), linguists have observed an apparent merger in the low-back vowels of California English speakers based on overlap in F1-F2 space (e.g., D’Onofrio, et al. 2016; Holland 2014; Kennedy and Grama 2012), but Wade (2017) demonstrates that apparent mergers may be distinguished along dimensions other than F1 and F2 frequency. Presenting two apparent-time analyses, I evaluate whether Californians’ low-back vowels are truly merged in production using wordlist data from ~400 speakers across 5 regions of California. I replicate previous findings of vowel convergence in formant space but demonstrate a simultaneous divergence in vowel duration over that same period. These findings suggest that speakers might have maintained a contrast, and that the low back vowels might not have merged in California English. These findings inform the design of future perceptual experiments, demonstrating a need to manipulate length in addition to formant frequency to account for this additional potential dimension of contrast in California English. I conclude with an argument against the exclusive use of the 2-dimensional F1-F2 plane, and I suggest ways of incorporating more holistic analyses of vowel quality into the workflow of future studies.

September 9

Emily Grabowski (UC Berkeley): Exploring phonetic time series analysis and representations

Until recently, phonetic analyses have hinged on researcher-determined features that are derived from spectral information in the data set and measured at a fixed point, such as formants, fundamental frequency, VOT, etc. While these measures have been unmistakably useful for phonetic analysis, more recent studies in acoustic phonetics have included an expansion towards including dynamic behavior of measures in analysis. In this talk, I will present some preliminary examinations of tools and techniques used in time series analysis and representation in comparison to more traditional methods, and discuss how machine learning and statistical advances might further influence acoustic analysis.

September 16

Rachel Weissler (University of Oregon): What is incorporated in emotional prosody perception? Evidence from race perception studies and analysis of acoustic cues

This research is centered upon how American English-speaking listeners cognitively interact with Black and White Voices. We investigated how individuals make judgements about the race and emotion of speakers. Participants listened to isolated words from an African American English (AAE) speaker and a Standardized American English (SdAE) speaker in happy, neutral, and angry prosodies, and were asked to indicate the perceived race and emotion of the speaker. Speech stimuli were analyzed for variation in pitch, creaky voice, and intensity, three acoustic factors used to distinguish emotion. Results of the perception study showed that SdAE was rated whitest in the happy condition, whereas AAE was rated blackest in neutral and angry conditions. Interestingly, the acoustic measurements of the two speakers evidenced that they use pitch, creak duration, and intensity in similar ways (according to mean and range). The results of the perception study indicate that listeners must be relying on cues beyond emotional acoustic ones to make their decisions for race and emotion of speaker. We argue that the pervasiveness of the Angry Black Woman trope in the U.S. is a stereotype that may have influenced participants' choices. As this is a first foray into raciolinguistic ideologies and emotion perception, we suggest that incorporating stereotypes into interpretation of emotion perception is crucial, as it may be a stronger driver of determining emotion from the speech than acoustic cues.

September 23

Scott Borgeson (Michigan State University): Long-distance compensatory lengthening

Compensatory lengthening (CL) is the phenomenon wherein one sound in a word is deleted or shortened, and another grows longer to make up for it. In mora theory (Hayes 1989), it amounts to the transfer of a mora from one segment to another. Traditionally, the two segments involved have always been adjacent to one another, or at the very least in adjacent syllables, but in this talk, I show (with evidence from Slovak and Estonian) that they can in fact be separated by multiple syllable boundaries.

Currently, no theoretical machinery exists that can distinguish between long-distance CL (LDCL) of this sort and the more widely-attested local CL, and as a result any language that displays local CL is also predicted to tolerate LDCL, contrary to fact. To fill this gap, I propose an expanded definition of the constraint LIN that applies across all tiers in the prosodic hierarchy. This will prohibit the inversion of precedence relations between moras and segments, effectively punishing moras the further they move from their input positions and thus requiring CL to be as local as possible.
The addition of this constraint accomplishes two things. First, it renders LDCL more marked than local CL, and thus guarantees that LDCL should be rarer cross-linguistically, and disfavored even in the languages that do tolerate it. Second, it nevertheless allows for the existence of LDCL in some cases—specifically, if LIN is dominated by some markedness constraint, and if local CL violates that constraint but LDCL does not, then LDCL will be selected instead. For example, CL in Estonian may not create new long vowels or geminates because of the constraints *VV and *GEM. If local CL can take place without doing so, it is always selected, but if local CL violates this prohibition and LDCL does not, then LDCL is chosen instead.

September 30

Noah Hermalin (UC Berkeley): An Introduction to Phonographic Writing Systems

This talk is intended to be a general introduction to phonographic writing systems, which are writing systems for which graphic units primarily map to phonological or phonetic information. The first portion of the talk will go over some basic writing system terminology, then discuss the typological categories which are commonly used to describe phonographic writing systems, including syllabaries, alphabets, abjads, and abugidas/alphasyllabaries. From there, we'll go into more detail on the range of extant (and possible) phonographic writing systems, with an eye for questions such as: what information is more or less likely to be explicitly encoded in different (types of) phonographic writing systems; what strategies do different writing systems use to convey similar information; what challenges do extant writing systems pose for common typological categories of writing systems; and what relevance do phonographic writing systems have for phonetics and phonology research. Time-permitting, the talk will close with a brief discussion of some ongoing work regarding how one can quantify how phonographic a writing system is.

October 7

Allegra Robertson (UC Berkeley): Rough around the edges: Representing root-edge laryngeal features in Yánesha’

In Yánesha’ (Arawakan), the phonetic, phonotactic, and prosodic traits of laryngeals indicate that they are suprasegmental features associated with vowel segments, resulting in laryngealized vowels /Vʰ/ and /Vˀ/ (Duff-Tripp, 1997; Robertson, 2021). The non-segmental status of laryngeals is at odds with most Arawakan languages (Michael et al., 2015), but their unusual characteristics do not end there. Although laryngeals are contrastive and lexically consistent, they emerge and disappear at morpheme boundaries in seemingly unexpected ways. Furthermore, noun possession data imply that, in addition to lexical and (occasional) phonological factors, morphosyntactic factors affect laryngeals. Starting from an agnostic and purely intuitive space, this talk seeks to clarify and formalize the complex behavior of laryngeals in Yánesha’, using original data from 2022 fieldwork. In this preliminary study, I explore the relative advantages of three frameworks to capture Yánesha’ laryngeal behavior: Autosegmentalism (e.g. Goldsmith, 1976), Q-theory (e.g. Inkelas & Shih, 2014), and Cophonology Theory (e.g. Inkelas, Orgun & Zoll 1997). I provisionally conclude that two of the three frameworks can account for this phenomenon, but with differing implications for the constraints at play.  

October 14

AMP practice talks

October 21

Canceled (AMP)

October 28

Michael Obiri-Yeboah (Georgetown): Grammatical Tone Interactions in Complex Verbs in TAM Constructions in Gua
Research in tonal properties has revealed interesting roles that tone plays using pitch to show both lexical and grammatical properties of language. Rolle (2018) provides crosslinguistic patterns and analytical tools for accounting for grammatical tones - marking of grammatical structures by a function of different tonal patterns. In this talk, I discuss grammatical tone (GT) in Gua and show that GT in Gua can be analyzed as tone melodies, L, HL and LH. I show further that, aside from the tonal patterns in verb roots, bòlí ‘break!’, bòlì ‘breaks’, bólì ‘broke’, there are verbal prefixes that increase the complexities in the tense, aspect and mood (TAM) structures in the language. I provide formal phonological analysis that accounts for the interactions between GT and complex verbs in TAM structures in Gua. The analysis also enhances our understanding of morphophonological interactions in linguistic theory. 

November 4

Rachel Walker (UC Santa Cruz): Gestural Organization and Quantity in English Rhotic-final Rhymes

In phonological structure, the segment root node is classically the locus of temporal organization for subsegmental units, such as features, governing their sequencing and overlap (e.g. Clements 1985, Sagey 1986). Root nodes also classically figure in the calculation of weight-by-position, by which coda consonants are assigned a mora (Hayes 1989). In this talk, I discuss evidence that motivates encoding temporal relations directly among subsegmental elements, represented phonologically as gestures (Browman & Goldstein 1986, 1989). A case study of phonotactics in syllable rhymes of American English, supported by a real-time MRI study of speech articulation, provides evidence for a controlled sequence of articulations in coda liquids. This study finds support for phonological representations that include 1) sequencing of subsegments within a segment (within a liquid consonant), and 2) cross-segment partial overlap (between a liquid and preceding vowel). Further, the assignment of weight in the rhyme is sensitive to these configurations. To accommodate such scenarios, it is proposed that segments are represented as sets of gestures without a root node (Walker 2017, Smith 2018) with a requisite component of temporal coordination at the subsegmental level. A revised version of weight-by-position is proposed that operates over subsegmental temporal structure. By contrast, the scenarios motivated by the phonotactics of rhymes with coda liquids are problematic for a theory in which sequencing is controlled at the level of root nodes.

November 11

no meeting (Veteran's Day)

November 18

meeting cancelled due to graduate student strike

November 25

meeting cancelled due to graduate student strike

December 2

meeting cancelled due to graduate student strike

December 9

meeting cancelled due to graduate student strike