Abstract:

This thesis reports on the results of production and perception experiments which investigated the nature of lexical accent in Tokyo Japanese (simply Japanese hereafter). Word prosody of Japanese is often labeled as pitch accent, characterized by a steep fall in F0 from the accented mora to the following one. Words that have accent on the final-mora (final-accented words) and words that have no accent (unaccented words) apparently have the same tone sequence, yet they differ in that a particle has a surface low tone after final-accented words while it has a high tone after unaccented words. It has been debated whether the two accent patterns are identical when they occur in isolation, and whether the tone on the following particle is the only acoustic correlate of pitch accent.

In the present study, a computerized database was used to search for all bimoraic and disyllabic minimal pairs of final-accented and unaccented words in Japanese. Because word familiarity is known to influence word recognition and production, only words with a relatively high familiarity rating were used, resulting in 20 minimal pairs. Ten native Japanese speakers (five males and females) produced the 20 pairs in isolation and sentence-medially followed by a particle. The production study found that, when the two types of words were produced in isolation, they were not significantly different in either their F0 peak in the second mora or their F0 rise from the first to second mora. When the words were produced sentence-medially, there was a significant difference within words, both in F0 peak and F0 rise. There was also a greater fall in F0 into the following particle for final-accented words. In the perception experiment, recordings from the production study were used to create three sets of stimuli: (A)?final-accented and unaccented words produced in isolation, (B)?final-accented and unaccented words excised from a sentence, and (C)?words and the following particle excised from a sentence. The listeners (n=23) were not able to identify words under conditions (A) or (B). Thus even though the stimuli type B differed in F0 peak and F0 rise, these acoustic properties were not sufficient for listeners to identify the words. Furthermore, the accuracy was only about 80 percent even when there was a following particle (stimuli type C). However, a strong positive correlation was found between the accuracy and the size of F0 fall difference between the two accent patterns in each pair, suggesting that F0 fall played an important role in perceiving accent. Thus, while accent information was redundantly present in F0 peak, F0 rise, and F0 fall of words produced in sentence context, listeners appear to use only the F0 fall into the following particle for word identification.