Phonology Circle will meet on Wednesdays this semester from 5–7pm in 32-D831 unless otherwise noted. The Phonology Circle is a weekly forum for the presentation of current research in phonology and phonetics. If you want to receive the email announcements, or have any other comments about Phonology Circle, please email Michael Kenstowicz.
Youngah Do & Michael Kenstowicz
Kyungsang Korean Accent Patterns: Lexical Drift, Loanwords, Novel Words
South Kyungsang is a pitch accent language with three lexically contrastive tonal patterns for monosyllabic nouns and four patterns for di- and tri-syllables. Our Phonology Circle presentation of last fall (10/26) focused on deviations from the etymologically expected accent class based on a word’s attested accent in Middle Korean. We showed that words are attracted to the statistically more frequent class and that this lexical drift tends to affect less frequent words (a common trait of analogical change Phillips 1984). Our data also indicated that sonorant codas, syllable weight, and the presence of fortis or aspirated consonants bias a word towards a particular accent class. In this presentation we show that these factors also emerge in a novel word experiment. We also present results of elicitations from 13 speakers designed to track the correlation between the well-known T > s analogy in the infection of nouns and the substitution of the High-High tonal pattern in place of the etymologically expected High-Low. We invite suggestions on the proper statistical measures to evaluate and present our results.
Peter Graff and Emad Taliep
English Speakers Track Absolute Frequency in Consonant Co-occurrence
Phonotactic wellformedness has been shown to affect behavior in a variety of linguistic tasks such as word-likeness judgments (e.g., Coleman and Pierrehumbert 1997). However, different researchers have often employed rather different measures for phonotactic wellformedness. This is particularly evident in the literature on consonant co-occurrence where two distinct measures of the wellformedness of constellations of consonants separated by vowels (e.g., p_t in pat) have been hypothesized. The first measure is the joint probability (i.e., absolute frequency) of the constellation P(C1_C2). Studies have found, e.g., that the joint probability of labial-coronal constellations (e.g. pat) is greater than the joint probability of coronal-labial ones (e.g. tap) in many different languages (MacNeilage et. al 1999, Carrissimo-Bertola 2010) and children as young as 7 months have been shown to be sensitive to this phonotactic (Nazzi et al. 2009, Gonzalez Gomez and Nazzi 2012). The second measure of wellformedness often employed is the observed- overexpected ratio (O/E) also known as pointwise mutual information (PMI). The difference between PMI and joint probability is that PMI relativizes the joint probability of a constellation to the independent probabilities of its parts such that PMI(C1;C2) = log(P(C1_C2)/(P(C1)·P(C2))), i.e. log(O/E). PMI has been hypothesized as a measure of wellformedness in studies of OCP-Place type phenomena (e.g. Frisch et al. 2004, Coetzee and Pater 2008) and PMI-based generalizations have been shown to affect wordlikeness judgments in languages like Arabic (Frisch and Zawaydeh, 2001). In this study, we compare the patterning of PMI and joint probability in the English lexicon and show that English CVCs present an ideal test-case for disambiguating the two measures. We present results from an online wordlikeness judgment study (N=50) and find that speakers’ judgments are more accurately predicted by joint probability (i.e., absolute frequency) than PMI.
Adam Albright
Discovering and modeling cumulative markedness interactions with loglinear models
It is often observed that “phonology can’t count”. This principle rules out, among other things, languages in which a marked structure is tolerated once or twice within a word, but not three or more times. In this talk, I discuss a set of restrictions in Lakhota (Siouan) which are very similar to such a ‘threshold’ effect: roots often contain a single marked structure (fricative, aspirated or ejective stop, consonant cluster), but roots containing multiple marked structures are rarer than one would expect, based on the independent frequencies of those structures. This observation leads two questions: is the degree of underattestation significant, and if so how should it be accommodated in a grammatical model? I show that both questions can be addressed using log-linear (maximum entropy) models of constraint interaction. First, I present results of a series of statistical models of the Lakhota lexicon, attempting to predict the relative type frequency of root shapes based on their phonological properties. The results show that models with interaction terms, in which multiple simultaneous violations may be penalized more than expected based on the individual violations, do significantly better at predicting lexical counts. Furthermore, the effect is strongest for combinations of structures that are independently most strongly penalized. Thus, it appears that cumulative effects are real, and some form of ‘counting’ is indeed warranted. I argue that these statistical models are too powerful, however: they could, in principle, impose strong penalties on combinations that are independently penalized only weakly (or not at all), or they could even reverse the direction of the preference so that languages tolerate a marked structure only in the presence of another marked structure. I argue that we can avoid these predictions with a simpler model, in which markedness constraints interact with MParse (Prince and Smolensky 1993/2004). I present the results of a learning simulation, showing that the observed cumulative effects can be predicted using a small set of markedness constraints on simple structures. Finally, I consider some typological predictions of the proposed model.
Sverre Strausland Johnsen
Vowel Weakening in Old West Saxon
The predecessor of the Modern English weak verbs with a past tense in -ed is the Old English second weak conjugation. In West Saxon, the main dialect of Old English, the past tense forms of these verbs exhibit both the vowel ‘a’ and ‘o’: ‘andswarade ~ andswarode, wundad ~ wundod’, corresponding to Modern English ‘answered’ and ‘wounded’. The explanation given in the grammars of Old English is that the ‘o’, which goes back to an older ‘u’, stems from the verb forms where an original *u followed in the ending. I raise an alternative hypothesis by which the ‘o’ (< ‘u’) originated in medial syllables through vowel reduction. A statistical analysis of the verb forms in the largest Old West Saxon manuscript shows that ‘o’ is significantly more common in medial syllables than in final syllables, but that there is no correlation between the distribution of ‘a’ and ‘o’ and where an original *u followed in the ending. The explanation in the grammars is therefore not supported. I suggest that the vowel has been reduced in medial syllables because vowels in medial syllables are shorter than in final syllables.
Suyeon Yun
Epenthesis Positioning in Loan Adaptation: Phonetics, Phonology, Typology
In loan adaptation, vowel epenthesis frequently occurs as a repair, when a cluster of a source language is phonotactically illegal in the borrowing language. The most notable previous finding has been that the position of epenthetic vowels differs depending on the type of cluster; sonority- rising clusters, especially stop-sonorant (TR), are more likely to be split by an epenthetic vowel than sonority-falling clusters, especially sibilant-stop (ST), e.g., ‘plastic’ > [bilastik] (internal epenthesis) vs. ‘study’ > [istadi] (external epenthesis) (Egyptian Arabic; Broselow 1992).
This study investigates epenthesis patterns in all possible types of clusters, both in word-initial and in word-final positions, from a cross-linguistic survey of loanwords. From the results, I propose new generalizations about the preferred site of epenthesis: (i) if a cluster contains an obstruent, a vowel is epenthesized after the obstruent, e.g., ‘camp’ > [kh?mph?] (Korean); (ii) if a cluster contains a sonorant, a vowel is epenthesized before the sonorant, e.g., rubl (Russian) > [rub?l] (Kirgiz).
By focusing only on initial clusters (Gouskova 2003, Steriade 2006) or on ST and TR clusters (Broselow 1992, Fleischhacker 2001, 2005), previous work has failed to identify the current broad generalizations and cannot uniformly explain the cases where the epenthesis patterns are different word-initially and word-finally, e.g., mnemonicheskij (Russ.) > [ymnemonicheskij] ‘mnemonic’ (Kirghiz) with external epenthesis vs. gimn (Russian) > [gimun] ‘hymn’ (Kirghiz) with internal epenthesis. My hypothesis is that the typology results from perceptual similarity between a consonant and its epenthesized form. Specifically, an obstruent is perceptually more similar to an obstruent-vowel sequence than to a vowel-obstruent, and a sonorant is perceptually more similar to a vowel-sonorant sequence than to a sonorant-vowel. I will show relevant phonetic bases and experimental results supporting this hypothesis. Based on this, the typology will be analyzed based on the P-map hypothesis (Steriade 2001/2009).
Sameer ud Dowla Khan (Brown University)
What echo reduplication reveals about correspondence and similarity
Unlike canonical reduplication, echo reduplication involves obligatory differences between the base and reduplicant, either in the form of subtraction or fixed segmentism, e.g. Bengali /goli/ > /oli goli/ ‘alleys, etc.’ and /kashi/ > /kashi tashi/ ‘cough, etc.’, respectively. I show that the unique properties of echo reduplication primarily stem from the multiple competing (anti-)correspondence relations at work, including IO-, BR- and IR-correspondence constraints, an anticorrespondence constraint, and morphemic constraints, all of which can be ranked relative to markedness constraints.Echo reduplication is also investigated as a productive alternation sensitive to phonological similarity. Results of a production experiment on Bengali reveal that BR-homophony avoidance is gradient as opposed to categorical. Bases that begin with consonants more similar to the /t/ are less likely to be echo-reduplicated with the default fixed segment /t/, and more likely to prefer one of the backup labial segments /m, f, p, u/. This homophony avoidance requires a gradient notion of phonological similarity, which can be closely modeled using a probabilistic metric that assigns different weights to different phonological features of the consonants being compared. Possible sources for feature weights are discussed, and will lead to future extensions of the current study.
Gretchen Kern
Perceptual Similarity in Sonority Contours: Evidence from Early Irish Rhyming Patterns
This is a practice talk for the 7th Celtic Linguistics Conference in June.
This paper aims to explain the basis of Early Irish rhyme and lend support to the concept of phonological similarity in sonority contours. Previous evidence for sonority contour-based similarity has come from patterns of epenthesis in consonant clusters (Flemming, 2008; Steriade, 2006) arguing that the perceptual distance between C1C2 and C1VC2 is smaller when there is a steep rise in sonority between C1 and C2 because that rise is more similar to the one between C1 and an epenthetic vowel. Adapting this, I argue that in Early (Old & Middle) Irish rhyme, the relevant dimension of similarity was the sonority profile: the difference in sonority between the most and least sonorous points. This can be used to explain some puzzling facts about Early Irish rhyme, namely the division of rhymeable consonants into classes, the behavior of consonant clusters, and the behavior of rhymes of two syllables.
