Phonetics and Phonology Workshop
sbowman October 15th, 2012
The Stanford Phonetics and Phonology Workshop (long known as ‘P-interest’) meets for presentations and paper discussions every Friday 12–1pm in Margaret Jacks Hall 126 (map; getting to Stanford), and monthly at a rotating cast of bars for P-int night conversation. All are welcome to come and participate.
All of our activities are announced over our mailing list, which you can join by filling out your information here.
2013–2014 organizers: Sam Bowman and Jeremy Calder
Faculty advisor: Arto Anttila
2013–2014 Schedule of Events
Fall Quarter 2013
9/27: Organizational meeting
10/4: Jaye Padgett (UCSC): Domain generalization in artificial language learning (work done with Scott Myers, UT Austin)
Many languages have restrictions on word-final segments, such as a requirement that any word-final obstruent be voiceless. There is a phonetic basis for final devoicing at the ends of utterances, but not the ends of words. Historical linguists have long noted this mismatch, and have attributed it to an analogical generalization of such restrictions from utterance-final to word-final position. To test whether language learners actually generalize in this way, two artificial language learning experiments were conducted. Participants heard nonsense sentences in which there was a restriction on utterance-final obstruents, but in which no information was available about word-final, utterance-medial obstruents. They were then tested on utterances that included obstruents in both positions. They learned the pattern and generalized it to word-final utterance-medial position, confirming that learners are biased toward word-based distributional patterns. The results also bear on licensing by cue, naturalness in learnability, and resistance to alternations (output-output correspondence).
10/11: Stephanie Shih (Stanford, Berkeley, &c.): Unstable surface correspondence as the source of local conspiracies (work done with Sharon Inkelas, Berkeley)
In Agreement by Correspondence theory (ABC; Hansson 2001; Rose and Walker 2004; a.o.), phonological patterns such as harmony and dissimilation arise from the interaction of corresponding surface segments. In harmony, corresponding segments become more similar in order to satisfy featural identity within a correspondence set. In dissimilation, the cost of satisfying identity is too high, and segments become less similar to escape the costly correspondence relationship (Bennett 2013). Harmony and disharmony, therefore, are repairs for resolving the same conspiracy of what we term unstable surface correspondence, in which two structures are similar enough to interact but too uncomfortably similar to co-exist within a certain distance.
In this paper, we argue that viewing local effects of assimilation and dissimilation as consequences of unstable surface correspondence offers an improved perspective on classic nasal-consonant (NC: e.g., *NC̥) patterns that have previously been regulated in Optimality Theory by context-specific markedness constraints (cf. Padgett 1993; Pater 1999/2004). Shifting the burden of grammatical analysis from (potentially arbitrary) contextual markedness to similarity-based surface correspondence illuminates the critical questions of which types of correspondences are the most unstable and which repairs are most likely to resolve them. This is an improvement over previous assumptions that local assimilation should be handled with one theory (autosegmental spreading), and long-distance interactions with another (ABC) (e.g., Rose and Walker 2004; Gallagher 2008; a.o.). The presupposition that local and long-distance effects are different obscures important parallels: recent work (Wayment 2009; Jurgec 2013) has shown that the similarity bias in segments participating in local assimilation resembles similarity thresholds for long-distance correspondences. Our proposal builds on these observations in showing that the underlying motive—unstable correspondence—drives the same repairs for both long-distance and local phonological patterns.
10/15: Tuesday P-int night at Rose and Crown, 7:30p
10/18: Sarah Bakst (Berkeley): A phonetic basis for the patterning of [χ]
The sonority hierarchy determines a segment’s sonority by its natural class, with obstruents registering low on the scale, followed by nasals, liquids, glides, and finally high-sonority vowels. The definition of sonority and existence thereof remain in dispute, but most definitions relate to syllable structure and phonotactics. The phonetic definition in Wright (2004) ranks segments based on the robustness of formant transitions. Other definitions rely on phonological patterning; Clements (1990) relates sonority to the ability of a segment to be a syllable peak.
Some phonetic realizations of the French rhotic are problematic for the sonority hierarchy. When the rhotic occurs in onsets following a voiceless stop, it is realized as a voiceless uvular fricative [χ]. French rhotics, regardless of the phonetic realization, pattern as high-sonority liquids and are one of only a few French segments allowed to occur between a consonant and a vowel; because of its rhotic status, [χ] is the only fricative in French that may occur in this position. There are two possibilities for the analysis of this segment: either the French sonority hierarchy is phonological and abstract, or there is some phonetic property of the voiceless uvular fricative, such as the ability to bear more robust perceptual cues to preceding segments, that allows this peculiar patterning. The present experiment tests whether [χ] is better than another fricative found in French, [f], at providing cues of a preceding stop in stop-fricative clusters.
10/25: Florian Lionnet (Berkeley): Phonological teamwork: An Agreement by Correspondence account of multiple-trigger assimilation (PDF abstract)
11/1: Brianna Kaufman (UCSC)
This talk will address the pattern of emphatic reduplication in Turkish e.g. kara “black” to kapkara “very black.” The picture is complicated by the fact that the fixed segment that intervenes between the reduplicant and the base alternates between four segments. Which of the four fixed segments surfaces is conditioned in part by co-occurrence restrictions. The standard approaches to reduplication within DM are unable to derive the correct outputs because they do not reference the right level of prosody, and because they fail to account for attested phonological alternations and variations. The proposal at hand will expand a blended model of DM and OT (á la Haugen 2008; 2011) in order to account for the very complicated case of fixed segmentism in Turkish.
11/8: No Meeting (Sam, Stephanie, and Ben gone for Phonology@UMass)
11/15: Greg Finley (Berkeley): Detection of phonetic features in nonspeech
In this talk I present experimental evidence that listeners can perceive an articulatory feature, lip rounding, from nonspeech auditory stimuli. Two experimental conditions, A and B, are discussed. Experiment A tested compensation for coarticulation in SV syllables (where S is a sibilant fricative). Listeners compensated for rounding on the vowel even for certain types of nonspeech vowels, including sine-wave speech and single-formant speech (an artificial glottal source band-pass filtered by a single formant) when the formant was close to the F2 of a back rounded vowel. In Experiment B, listeners preferentially associated frequency-modulated pure tones (rising or falling beeps) with video of rounded or unrounded speech sounds depending on frequency range and on direction of modulation: lower range and downward modulation were associated with rounded vowels in CV syllables, and upward modulation with rounded glides. These results show that phonetic information can be gleaned from simple auditory objects (i.e., a gestalt speech percept, or even the illusion of speech, is not necessary), and they suggest that more strongly categorical phonetic percepts can be composed of these objects.
11/20: Wednesday P-int night at The Rose and Crown at 7:30 pm
11/22: Matt Faytak (Berkeley)
Vowels produced with concomitant frication are observed in a wide range of languages and suggest a few interesting complications to phonological theory. After surveying the cross-linguistic similarities and differences that hold within the class of spirantized vowels, I put forward a series of phonetically natural sound changes to motivate their odd phonological behavior and explain their distribution. I additionally highlight the need for further research on languages with spirantized vowels, speakers of which are conveniently available on most American university campuses.
12/6: Melinda Fricke (Penn State, Berkeley): Phonetic reduction and the lexicon: exploring effects of positional neighborhood density on articulatory duration
In retrieval-based accounts of phonetic variation (Bell et al., 2009; Gahl et al., 2012), the ease with which a word can be retrieved from the lexicon has been hypothesized to affect its phonetic realization in connected speech: more accessible words tend to be produced with more reduced pronunciations, all else being equal. In this talk, I present analyses of data from a word learning experiment with preschoolers and from single word and spontaneous speech produced by adults indicating that phonological overlap between words in the lexicon has a small but significant effect on the ease with which individual segments are encoded for production. I hypothesize that previously observed effects of phonological neighborhood density on phonetic duration are in fact the result of fluctuations in the speed of phonological encoding, and that retrieval-based accounts of phonetic variation can more accurately be localized at the segmental (rather than lexical) level.
12/9 (Monday): Lev Blumenfeld (Carleton): Metrical easiness and typicality as a window into grouping structure of verse
The so-called Russian Method in metrics (Bely 1929; Hayes 2013) seeks to investigate the structure of verse by comparing the distribution of prosodic and other properties in verse and prose. In this talk I offer a new application of the Russian method in quantifying what Hanson & Kiparsky (1996) have informally called Metrical Interest. I investigate two related measures of verse rhythm and their role in metrical grammars. The first is Metrical Easiness, which is the degree of strictness of metrical correspondence constraints, related to the probability that random strings of prose are metrical. I show that while Easiness plays a role in meters, that role is indirect. Secondly, I quantify a related notion of Prosodic Typicality, or the natural language frequency of a prosodic structure instantiated by a line of verse. In a case study of English and Russian sonnets, I argue that Typicality is controlled by the metrical grammar, in that low typicality is associated with closure in a grouping structure, and that typicality reveals otherwise inaccessible aspects of the metrical organization of poems.
12/13: Olek Glowka (Stanford): Prosodic variation in Polish Noun Phrases: against recursion below the phonological phrase
Polish NPs modified by postnominal adjectives exhibit pervasive prosodic variation, with main prominence found either on the lexically stressed syllable of the head noun or on that of the modifier. The distribution of prosodic variants has been proposed to reflect the speaker’s interpretation of the modifier, amenable to a classificatory or an ascriptive reading (Mańczak 1952, Sussex 1976). I present behavioral and acoustic evidence to argue that the variants are encoded as distinct prosodic constituents, a compound and a phrase respectively. The findings challenge the characterization of compounds as recursive prosodic words and lend support to an independent prosodic domain.
Winter Quarter 2014
1/10: Organizational meeting in the chair’s office (460-127)
1/24: Kodi Weatherholtz (OSU)
1/17: TBA (may conflict with CSLI gradient grammar workshop)
1/31: Kevin McGowan (Stanford)
3/7: Tentative: Larry Hyman
3/14: All department event: SemFest
2012–2013 Schedule of Events
Spring Quarter 2013
4/5: Sam Bowman (Stanford): Two arguments for vowel harmony by trigger competition (CLS/mfm practice talk) following an organizational meeting.
I present two phenomena in front-back vowel harmony which are difficult to account for in standard theories, and argue that with some necessary elaborations, Trigger Competition (TC, Kimper, 2011) is best suited to account for both. TC is a new harmony framework based on a positive constraint (imperative) set in Serial Harmonic Grammar, and allows for agreement between non-adjacent segments. The constraint considers both the distance between trigger and target and the nature of the trigger in assigning rewards, allowing for a fairly sophisticated approach to non-participating segments.
Hungarian vowel harmony shows a pattern of optionality (Benus, Gafos, and Goldstein, 2003) in its handling of phonetically front transparent vowels in harmonically back contexts: Back suffixes are used after single transparent vowels, either front or back suffixes after the semi-transparent vowel /e/ or after pairs of transparent vowels, and front suffixes after transparent vowel–/e/ sequences. Under TC, this emerges readily: Distance and trigger strength conspire to produce these additive effects.
In Seto, the transparent vowels /i/ and /e/ can appear in back vowel contexts without interacting with harmony. Remarkably, back equivalents to these vowels, /ɨ/ and /ɤ/, also appear in the inventory. Both conventional approaches to transparency in local harmony systems—neutralization and underspecification—require that neutral vowels be un-paired, but TC has no such requirement: If any constraint prevents a vowel from alternating, it will be neutral, and if it is a weak trigger, it will be transparent.
4/12: Andrea Davis (Arizona): When is Phonetic Variation Helpful for Learning Word Forms?
Phonetic variation between speakers promotes generalization when learning new words (Richtsmeier et al., 2009; Rost & McMurray, 2009, 2010). But is variation always helpful for generalization? It could be the case that whether variation is beneficial for generalization depends on a variety of factors, including prior experience with the language, the developmental stage of the learner, whether or not the new words are similar in form to other words, or whether the test is on perception vs production of the new words. The proposed work focuses on two of these factors. Do learners with more experience with a language still benefit from phonetic variation, when learning new words? Additionally, is there a difference between perception and production, in whether experienced learners continue to benefit from phonetic variation?
4/19: QP Fest: Stanford speakers TBA
4/22: Monday P-int night at Rose and Crown
4/26: Stephanie Shih (Stanford): Function versus content word prosodification: evidence from phonetic reducibility (Davis Grammatical Word Workshop practice talk)
The division between lexical content words and grammatical function words has been motivated in part by differences in stress and prosody. The traditional view maintains that content words have lexically-programmed stress whereas monosyllabic function words are lexically unstressed and appear on the surface in both strong (unreduced) or weak (reduced) forms. Despite this commonly categorical divide, natural language corpus studies based on intonational prominence have suggested that function words themselves are not a homogeneous class when it comes to their prosodification (e.g., Altenberg 1987; Hirschberg 1993; Bell et al. 2003). In this talk, I follow this latter view: with evidence from phonetic reduction in a corpus of conversational American English, I show that the extent to which function words appear in strong and weak forms varies by subclasses, with some function words behaving like lexically-stressed content words and others exhibiting more variable prosodic realizations. I focus specifically on the prosodification of a function word as weak or strong as conditioned by the neighboring context of weak and strong syllables. Crucially, content words and function word subclasses will differ in their sensitivity to rhythmic environment.
5/3: Paper Discussion: Florian Schiel et al.: Rhythm and Formant Features for Automatic Alcohol Detection
5/17: Emily Cibelli (UC Berkeley): Early processing pathways of words and pseudowords: Evidence from electrocorticography
Pseudowords – phonotactically-legal novel forms like “blick” and “piteretion” – are common tools employed in studies of lexical processing. They are often compared to words, under the assumption that these novel forms isolate sub-lexical levels of processing; however, there is some debate about whether words and pseudowords utilize shared or distinct pathways at early stages of processing. Critically, the answer to this question affects the interpretation what is being isolated in word-pseudoword comparisons.
In recent years, this issue has been examined by a wealth of neuroimaging studies, with the goal of identifying lexical and sub-lexical processing pathways at the neural level. This work contributes to that growing body of literature by presenting data from a word-pseudoword listening task using electrocorticography (ECoG), a technique which records neural activity directly from the cerebral cortex. ECoG data is relatively new to the neurolinguistics literature, but has a high spatial and temporal resolution, an advantage in tracing processing pathways. Results are discussed in light of Hickok and Poeppel’s (2007) dual-stream model of language processing. The data suggests that at early processing stages, words and pseudowords share a pathway in regions of the brain identified as being involved with phonetic, phonological, and lexical processing.
5/24: no meeting
5/30: Thursday P-int night at Rose & Crown
5/31: Allan Schwade (UCSC) The Non-Grammatical Gender of Words
Walker and Hay (2011) demonstrated that English listeners are faster and more accurate at identifying auditorily presented words in a lexical decision task when words associated with a certain age-group were spoken by speakers from that age-group, supporting exemplar models that claim tokens are tagged for attributes of the talker (Johnson, 1997; Pierrehumbert, 2001). The study to be presented expands on the work of Walker and Hay by showing that English speakers’ reaction times for orthographically-presented words associated with a non-grammatical gender are primed by images of men and women, albeit in unexpected ways. The results raise interesting questions regarding the ability of people to report the sociological attributes associated with words, and the robustness of sociological priming effects across different modalities.
6/7: Chris Donlay (UCSB)
Winter Quarter 2013
1/11: Matt Faytak (Berkeley): Obstruent Vowels in Kom
Vowels are attested with a wide range of secondary articulations not involving the tongue body, such as nasalization or pharyngealization. The Grassfields Bantu language Kom (ISO 639-3 bkm, Bantoid, Cameroon) appears to distinguish between vowels with and without an additional coronal and labiodental constriction; similar vowels are attested in other languages of the region (Fransen 1995, Connell 2007).
After a discussion of the phonetics of Kom’s vowel system, I argue that the set of [+high] vowels in Kom consists of four canonical high vowels /i y u ɯ/ and two additional “obstruent vowel” phonemes, which I denote as /z v/, which are consistently realized with significant alveolar and labiodental constriction, respectively. I further argue that diphthong formation, in which the obstruent vowels freely participate, supports this analysis. The Kom vowel system as I analyze it is a major departure from prior work on Kom (Shultz 1993, 1997) and an unusual addition to attested arrangements of vowel systems in the acoustic space delimited by the vocal tract.
1/16: P-Int night at Rose and Crown
1/18: Paper discussion: Paster, in Press: Rethinking the ‘duplication problem’
1/25: Joint meeting with UCSC Phlunch: Stephanie Shih (Stanford).
2/1: Departmental event, no meeting.
2/8: Departmental event, no meeting.
2/15: Alex Djalali (Stanford): A constructive solution to the ranking problem in Partial Order Optimality Theory
I give a solution to the ranking problem in Partial Order Optimality Theory (PoOT), which can be stated as follows: Allowing for free variation, given a finite set of input/output pairs, i.e., a dataset, that a speaker knows to be part of some language, how can learn the set of all PoOT grammars under some constraint set compatible with that dataset?
For an arbitrary dataset, we provide set-theoretic means for constructing the set of all PoOT grammars compatible with that dataset. Specifically, we determine the set of all strict orders of constraints that are compatible with dataset. As every strict total order is in fact a strict order, our solution is applicable in both PoOT and classical optimality theory (COT), showing that the ranking problem in COT is a special instance of a more general one in PoOT.
2/22: Departmental event, no meeting.
2/27: P-int night at Rose and Crown
3/1: Jared Bernstein (Pearson): TBA
3/8: Ed King and Seung Kyung Kim (Stanford): PRAAT tutorial
3/13: P-int night at Rose and Crown
3/15: Dead week: No meeting
Fall Quarter 2012
9/28: Organizational meeting
10/2: P-int night: Madam Tam’s at 8p
10/5: Jonah Katz (Berkeley): Rhyme Patterns Reiterate Phonological Typology
10/12: Joint meeting with the CrISP workshop
10/19: Paper discussion: Benus, Gafos and Goldstein, 2003: Phonetics and Phonology of Transparent Vowels in Hungarian
10/26: Jason Riggle (UChicago): Evaluating models of variation via grammar sampling
Modeling variation is challenging because the combination of linguistic and nonlinguistic factors can make it difficult to determine when a proposed grammatical model is fitting and when it is overfitting observed data. Nonetheless, modeling variation is also appealing because it incorporates an additional (and rich) stream of information for the creation and evaluation of linguistic models. In this talk I focus on constraint-based phonological grammars that generate patterns of variation by sampling from rankings of constraints. I show first, that several published claims which assert that sampling from rankings cannot generate fine-grained distributions are based on dubious (and tacit) assumptions about the sampling operation. I then present a range of empirical phenomena where sampling models can fit the data quite well and, in fact, maybe too well. I conclude with a discussion of how overfitting can be assessed for such models and, relatedly, I ask what role noise in the training data plays when fitting–and perhaps overfitting–patterns of variation.
11/2: Paul Kiparsky (Stanford): How stress became pitch accent in Scandinavian: evidence from Fenno-Swedish
The Swedish and Norwegian contrast between pitch accent 1 and 2 is standardly treated by associating an inherent lexical pitch with accent 2 words. Phonetically these words are distinguished from accent 1 words in central Scandinavian by having two pitch peaks, and in southern Scandinavian by the delayed timing of their single pitch peak. I show that the corresponding accent distinction in the Swedish of Tenala (Finland) requires a very different analysis. In this dialect, pitch accent is fully predictable from stress: accent 2 occurs in all and only those words that have two or more feet — just the distribution hypothesized by Riad (2003, 2005) for Proto-Nordic. Älvdalen Swedish and Gudbrandsdalen Norwegian can tentatively be assigned to the same type. The reason why pitch accent remained a redundant feature in these conservative dialects is that they retain the Nordic stress and quantity features that condition it, perhaps through the influence of the co-territorial Saami and Finnish languages. It was only the elimination of stressed light syllables in the rest of continental Scandinavian that first made the accent 1 vs. accent 2 contrast phonemic. This further suggests a novel contact-based explanation for the fact that pitch accent arose only in Scandinavian, although its precursor prosodic system was pan-Germanic.
11/9: Junko Ito (UC Santa Cruz): Matching Prosodic Constituents
The theory of the syntax-prosody interface has taken a significant step in advance with the development of Match Theory (Selkirk 2009, 2011, etc.), whose central idea is that syntax-prosody mapping constraints are of a very simple kind, essentially demanding the direct replication of core constituents in syntax (words/phrases/clauses) by corresponding prosodic constituents (prosodic words/phonological phrases/intonational phrases). Match constraints are constituency-based, hence both edges are required to be matched. Prosody as it emerges from the syntax-phonology map often does not exactly correspond to syntax at all, but Match Theory interprets this as due not to the mapping constraints themselves, which would require an imperfect and distorted match, but rather to the fact that the mapping constraints, even though they demand a perfect match, are often dominated by other constraints that govern prosodic form (binarity, antilapse, etc.) and result in syntax-prosody disalignment for this reason. Match Theory, in its focus on constituency and not privileged (left or right) boundaries, is in this central aspect very different from its antecedent and competitor, the End-based Alignment Theory, which builds syntax-prosody disalignment into the mapping constraints themselves.
This talk will take up some extensions of Match Theory to prosody-prosody matching (e.g., requiring prosodic words to be coextensive with feet), and consider how it fares with respect to Generalized Alignment Theory and Generalized Template Theory (McCarthy and Prince 1994, 95, 99, etc.). In exploring the connection between Match constraints and what might be considered “Emergence of the Unmarked” effects, we will look at two case studies where such prosody-prosody match constraints play a role: (1) Serbo-Croatian vowel shortening (Zec 1999) and (2) the distribution of the Danish stød (glottal accent), following an analysis along the lines of Kiparsky 1995/2006 on Livonian stød.
11/16: Tania Rojas-Esponda (Stanford):
Sundanese possesses two plural allomorphs, ar and al. They are particularly important as they can be used to pluralize adjectives, nouns and also verbs. Cohn, Holton and McCarthy present analyses for the ar/al alternation, starting from ar as the assumed underlying form. Roughly, they claim that the r in the plural affix ar dissimilates to l if there is another r in the word, except when the other r is directly neighboring (separated from the affixal r only by a vowel). McCarthy deals with this exception by introducing the markedness constraints *lVrV and *rVlV. These two constraints are natural enough given the OCP, which says that similar but distinct consonants (such as l and r) should not occur close to each other. However, McCarthy also claims there is a crucial asymmetrical treatment of the two sequences lVrV and rVlV when they occur underlyingly. I will talk about whether the data support this claim.
The analyses of Cohn, Holton and McCarthy are based on the same, and rather limited, set of examples. In my talk I use a larger set of data extracted from a Sundanese version of the bible that has on the order of a million words as well as data from an online Sundanese dictionary to test some of the generalizations made by these authors.
11/23: no meeting (holiday)
11/30: Olga Dmitrieva (Berkeley, PhD Stanford) and Giulio Caviglia (Purdue):
Convex regions and phonological frequency: Extending the weighted constraints approach.
The currently dominant framework for modeling phonological phenomena, Optimality Theory, provides tools for capturing categorical phonological typology (factorial typology) and can be extended to account for gradient and stochastic phenomena, as well as their frequency. Recently, a competing approach, based on weighted rather than ranked constraints, Harmonic Grammar, has been gaining popularity. In this talk we explore the mathematical bases of Optimality Theory and Harmonic Grammar, the underlying connection between the two and their differences.
We also address the limitations of Optimality Theory compared to weighted constraints approach and the fact that Harmonic Grammar, the currently most developed implementation of the weighted constraints approach does not fully explore the potential of this method.
We then propose a natural extension of the weighted constraints approach, which allows for the development of whole phonological typologies, equivalent to those produced by the factorial typology in standard OT. This method also provides an estimate of the relative frequencies of the possible language types and output types, based on the relative volumes of the convex regions in the weight space.
12/4: P-int night: TBA
12/7: LSA practice talks: Jeremy Calder (Stanford); Sam Bowman (Stanford)