Abstract

Changes in fundamental frequency result from two fundamentally different processes: phonological or linguistic determinants of f0 (macroprosody), which are used consciously by a speaker to achieve various intonation patterns or lexical tones, and phonetic determinants of f0 (microprosody), which are an automatic, unintended consequence of coarticulation with neighbouring or concurrent speech sounds. Vowel-intrinsic f0 refers to a phonetic phenomenon by which f0 decreases with vowel openness, while vowel-intrinsic pitch describes the increase in perceived pitch with vowel openness. This paradoxical relationship is believed to be necessary so that vowel-intrinsic (phonetic) f0 does not affect linguistic (phonological) f0 use in tone and intonation languages. However, there is evidence that perceptual normalisation is at best only partial. Thus arises the question as to how the listener processes the "leftover" intrinsic f0. We predict that, in tone languages, any portion of intrinsic (phonetic) f0 not fully compensated for in perception must be processed as phonological f0 and thus interpreted as tone. Therefore, the aim of this dissertation is to establish the degree to which intrinsic f0 interferes with phonological f0 in the form of lexical tone with broader implications for tone processing, sound change and speech technology. In a first set of experiments, we investigated intrinsic f0 and intrinsic pitch on monophthongs in Hong Kong Cantonese. In terms of intrinsic f0, results revealed an f0-raising effect of close vowels on f0 of the high (55) level tone, but not the low (22) or mid (33) level tones in production. However, machine classification of f0 based on training with tone information only showed an effect of vowel openness on the correct classification of the low (22) and mid (33) level tones and not the high (55) level tone. This latter effect was reflected in perception, where we found a pitch-lowering effect of close vowels on the low (22) vs. mid (33) category boundary but not on the mid (33) vs. high (55) category boundary. Data from speaker-listeners who participated in both experiments revealed that intrinsic f0 and intrinsic pitch were matched for the lower tone contrast, but that intrinsic pitch lagged behind intrinsic f0 for the higher tone contrast. We concluded that this relationship reflects compensation for coarticulation where a phonological contrast would otherwise be at risk of misperception. In the second set of experiments, we extended a previous study by investigating whether perception of intrinsic pitch is dynamic in Hong Kong Cantonese and falls or rises depending on diphthong pattern. Contrary to our results for monophthongs, this was not the case. In order to establish whether this was part of a mismatch between production and perception, we additionally investigated intrinsic f0 in production of Cantonese diphthongs. However, there was no falling or rising intrinsic f0, which explained the absence of an intrinsic pitch effect in perception. Furthermore, both the diphthong production and perception studies revealed somewhat paradoxical results involving an effect in the opposite direction to our hypothesis in one experimental condition. We provide a number of possible explanations for these results, both theoretical and methodological. Overall, we were able to confirm the presence of intrinsic f0 and intrinsic pitch in a complex tone system. However, the effects were not consistent across all conditions. Intrinsic pitch applied only to the condition in which intrinsic f0 has the most potential for disrupting tone contrasts, which we interpret as evidence of compensation for coarticulation. The results are discussed in terms of their implications for tone processing, the phonetics-phonology interface and speech technology.