Mafuyu Kitahara

Domain-Relative Faithfulness and the OCP: Rendaku Revisited

The Obligatory Contour Principle (henceforth the OCP) was first proposed at the advent of autosegmental phonology ( Leben 1973; Goldsmith 1976) for a restrictive representation of autosegments. It was modified along with the development of phonological theories in the 1980s and has been adapted to the analyses of a wide variety of languages and phenomena (see McCarthy 1986 for a concise review and Yip 1988 for a typological consideration with respect to the OCP). The frequently cited definition of the OCP is as follows (McCarthy 1986):

Perception of American English alveolar stops and flaps by Japanese learners of English: Does allophonic variation matter?

A lexical decision experiment was conducted with Japanese learners of English with relatively low English proficiency to investigate whether second-language (L2) learners utilize allophonic variation when recognizing words in L2. The stimuli consisted of 36 isolated bisyllabic words containing word-medial /t/, half of which were flap-favored words (e.g., better, city) and the other half were [t]-favored words (e.g., faster, custom). All stimuli were recorded with two surface forms: /t/ as a flap (e.g., better with a flap) or as [t] (e.g., better with [t]). The stimuli were counterbalanced in the lists using a Latin Square design, so that participants only heard one of the two surface forms. The accuracy data indicated that flap-favored words pronounced with a flap (e.g, city with a flap) were recognized significantly less accurately than [t]-favored words with a flap (e.g., faster with a flap) and [t]-favored words with [t] (e.g., faster with [t]). These results suggest that Japanese learners prefer canon...

Production of a non-phonemic variant in a second language: Acoustic analysis of Japanese speakers’ production of American English flap

Second-language (L2) learners need to learn the sound system of an L2 so that they can distinguish L2 words. However, it is also instructive to learn non-phonemic, allophonic variations, particularly if learners want to sound native-like. The production of intervocalic /t d/ as an alveolar flap is a prime example of a non-phonemic variation that is salient in American English and presumably noticeable to many L2 learners. Yet, how well such non-phonemic variants are learned by L2 learners is a relatively under-explored subject. In the present study, Japanese learners’ production of alveolar flaps was investigated, to clarify how well learners can learn the phonetic environments in which flapping tends to occur, and how L2 experience affects their performance. Native Japanese speakers who had lived in North America for various lengths of time read a list of words and phrases that contained a potentially flappable stop, embedded in a carrier sentence. Preliminary results indicated that the rate of flapping varied considerably across different words and phrases and across speakers. Furthermore, acoustic parameters such as flap closure duration produced by some speakers showed intermediate values between native-like flaps and regular stops, suggesting that flapping is a gradient phenomenon. [Work supported by JSPS.]

Deriving functional load of phonemes from a prosodically extended neighborhood analysis

The functional load of phonemes is a long-standing, but not a main-stream notion in modern linguistics: that some pairs of phonemes distinguish more words than other pairs is intuitively plausible, but hard to quantify. Meanwhile, neighborhood effects in word recognition and production have been one of the central topics in psycholinguistics, leading to a wide variety of investigations. However, the Greenberg-Jenkins calculation, the most common definition of phonological neighborhood, deals only with deletion, addition, and substitution of phonemes, lacking any consideration of prosody. For example, homophones, which cannot be segmental neighbors and thus excluded in most neighborhood research, can be distinctive if lexical accent is specified. The role of onset/rhyme distinction in neighborhood calculation has been discussed, but morae, another basic unit of prosody, were not mentioned in the literature. We propose a novel method for calculating the functional load based on a prosodically extended neigh...

Functional load of segments and features

The present paper proposes a new measure of functional load for segments and features. In a nut shell, it is based on word frequencies and the number of minimal pairs in which the relevant segment/feature is crucial in distinction. For example, minimal pairs distinguished only by /t/ are most frequent in English while those distinguished by /k/ are most frequent in Japanese. As for functional load of features, single‐feature contrasts and multiple‐feature contrasts are incorporated in the calculation. In Japanese, [high] alone distinguishes the largest number of minimal pairs while [voice] distinguishes words most frequently in cooperation with other features. Word frequency and familiarity database for English and Japanese are used to observe the commonalities and differences in both languages with respect to the proposed measure of functional load. This line of analysis suggests a better account for a certain phonological process being more typical in one language but not in the other. Functional load c...

English stress timing and dynamical control of speech timing

Studies of English speech timing have concluded that, despite earlier claims, English is neither stress‐timed nor syllable‐timed. This study directly measured context effects on timing depending on prosodic neighborhood, and compared the contribution of any stress‐timing or syllable‐timing tendency while speech cycling. Ss repeated a phrase to a metronome at varying rates. This complex metronome included high‐pitched tones (‘beeps’) at roughly 1 s with a lower tone (‘boop’) occurring either 1/2 (2‐beat) or 2/3 (3‐beat) of the way between beeps. We hypothesize dynamical attractors for stressed syllables at simple harmonic fractions (1/2, 1/3, 2/3) of the phrase cycle. Various texts invite readings with two stresses (‘Bite the back’) or three stresses (‘Bite Bill’s back’), and some had perturbing unstressed syllables inserted between stressed ones (‘Biting Bill’s back’). Perturbing and target vowel onset times were fit with a multiple linear regression equation using time predictions assuming equal stress i...

Distinctive functional load of pitch‐accent in Japanese based on word familiarity

The present paper proposes to measure the ‘‘functional load of pitch‐accent’’ for Japanese words. It is calculated from (1) the ratio of the number of accentually contrastive homophone pairs to the number of noncontrastive pairs; and (2) the difference between word‐familiarity ratings for each pair. The functional load is larger when there are more contrastive homophones to noncontrastive homophones for the target word, and when the difference of word familiarity scores is smaller between each homophone pair. A large scale word‐familiarity database is used for the calculation of functional load and other statistical properties of pitch accent. Distribution of accent and opposition types at each word length are investigated with respect to familiarity scores. Results show that oppositions which include an unaccented form dominate in short simplex words, from low to high‐familiarity range. This suggests that the role of pitch accent in distinguishing homophones is biased to the presence–absence contrast, and not to the location per se of pitch accent. Preliminary results from a perception experiment suggest that top‐down information in the lexicon, such as functional load and distribution of accent and opposition types, interacts with the bottom‐up process in lexical access.

Effects of speech rhythm on timing accuracy of stressed syllable beats

It has been shown that English speakers have a strong tendency to locate stressed syllable onsets (that is, ‘‘beats’’ or ‘‘P centers’’) near harmonic fractions (like 1/2, 1/3, 2/3) of the repetition cycle of a repeated phrase [Cummins and Port, J. Phonetics (1999)]. These results were interpreted as evidence of a role for frequencies at multiples of the repetition cycle creating temporal attractors for stressed syllable onsets at harmonic fractions of that cycle. If harmonic oscillators account for temporal attractors, then, having an attractor at 2/3 should necessarily imply the existence of an attractor at 1/3, whereas an attractor at 1/2 would not similarly imply an attractor at 1/4 (since a third oscillator would be required). Thus, if a phrase like take a pack of cards is repeated with card onset at phase 1/2, there should be a weaker attractor controlling the timing accuracy for pack onset than if card onset is at 2/3. Four speakers were tested using various text materials and speaking rates. Prelim...

Vowel devoicing and the loss of lexical accent in Tokyo Japanese

The fundamental frequency of words with devoiced syllables was measured in order to see how the conflict between devoicing and the pitch accent is resolved in Tokyo Japanese. Previous studies found that the pitch after the devoiced accented syllable is elevated in order to compensate for the loss of the accent. Earlier results [M. Kitahara, J. Acoust. Soc. Am. 101, 3195 (1997)] show that the pitch elevation occurs only when there is no phrasal high tone before the devoiced region. The present study further eliminates the possibility of accent shifting by controlling the moraic structure of target words. They contain one or two devoiceable syllables and a moraic nasal in front of the devoiceable region. Four speakers of Tokyo Japanese pronounced target words embedded in a carrier phrase. If the accent shifts away from the devoiced syllable, pitch elevation should not occur because the accent is realized somewhere else. Preliminary analysis of the data shows that the accent on the devoiced syllable is not s...

Pitch accent, phrase tones, and vowel devoicing in Tokyo Japanese

In Tokyo Japanese, short high vowels devoice between voiceless consonants, which conflicts with the realization of pitch accent. The fundamental frequency around the devoiced region in a corpus of 12 speakers was measured to investigate if and how speakers compensate for the loss of information normally conveyed by the pitch accent. Previous studies revealed that elevation and abrupt pitch fall immediately after the devoiced region is characteristic of utterances containing a devoiced accented syllable. However, preliminary analyses of these data show the occurrence of the characteristic pitch pattern after a devoiced region is dependent on the pitch height before the devoiced region. If two syllables are consecutively devoiced and the first one bears an accent, the pitch elevation only occurs when the antepenultimate syllable has a low phrase tone. This suggests a nonlocal pitch adjustment exists dependent on the occurrence of devoicing and the location of phrase tones. This also implies that vowel devoi...