Phonological neighborhoods and phonetic similarity in spoken word recognition

Abstract

English neighborhood literature has demonstrated that neighborhood density affects the auditory word recognition. However, an unresolved question is exactly how neighborhood density should be calculated. In this paper the definition of a lexical neighborhood is explored in Japanese. Data for the analyses were collected from Japanese neighborhood experiments using the same 700 test words and a lexicon that consisted of only nouns from the NTT psycholinguistic database [Amano and Konodo, 1999]. Three different neighborhood calculations were used to analyze the data. The first calculation was based on Greenberg–Jenkins phoneme substitution, deletion, and insertion rules. The second calculation included prosodic information as another dimension in the neighborhood calculation in order to reflect the finding that prosodic information has a vital role in Japanese word recognition. The third calculation was based on the auditory properties of the words in the lexicon. Neighborhood density was measured by comparing the similarity of cochleagrams of the 66<th>000 audio files. The results of the analyses demonstrated that phonological similarity within the lexicon seems to be calculated based on higher-level abstract representations rather than on a lower-level acoustic-auditory representation in any of the experiments. The implications from the results to the current word recognition theories will also be discussed. [Work supported by NIH.]

Relationship between perceptual accuracy and information measures: A cross-linguistic study

“…As in Korean, the PI for null codas (in syllables without codas) was lower than the PI for the other possible codas, suggesting that a syllable without coda is fairly common in Japanese. This observation is compatible with the finding on the prevalence of CV syllables in Yoneyama (2000). …”

“…That /k/ has a lower PI than null onset suggests that onset /k/ is more common than syllables without onsets. The pattern is consistent with the finding in Japanese by Yoneyama (2000), in which /k/ was the most frequent sound.…”

“…Compared to English and Korean, PI for null onset (represented as ∅) is relatively high compared to other onset consonants, which suggests that Japanese syllables generally occur with onsets. To a certain extent, the finding is consistent with Yoneyama (2000), which showed that Japanese syllables generally begin with a consonant.…”

“…It is different from the observations from English or Korean, in which null onset always showed the highest FL of all other onsets. This pattern again can be accounted by the prevalence of CV syllables in Japanese (Yoneyama 2000). The highest FL measure shown by the null coda is also in line with this observation.…”

“…Such a high FL for /k/ implies that onset /k/ differentiates the greatest number of syllables in Japanese, which is compatible with the frequency count data of Japanese, where /k/ was found to be more frequent than /t/ in terms of segment frequency (Yoneyama 2000).…”