Final Report Summary - RPPV (Representation and Processing of Pitch Variation in Tonal Languages)
In connected speech, spoken words may vary considerably from their forms in isolation. Much of the variation is regular. For example, the [n] in green bag is often assimilated to [m] due to the following segment [b], yielding something similar to “greem bag”. There have been heated debates on whether and how speakers store such regular segmental variants in their long-term memory (i.e. representation in the mental lexicon) and how they process them in speech production and perception. To shed light on this debate, this project aims to examine the representation and processing of pitch variation, a supra-segmental phenomenon which presents even more challenges to our general understanding of variation in speech given its complexity. Our main empirical focus is pitch variation in Chinese dialects, where pitch variation can originate from a wide range of sources varying from lexical tonal contrasts to pragmatic meanings.
Within the project, we have addressed the central research question through the looking glass of a range of tonal variation patterns in Chinese dialects. Take Beijing Mandarin as an example. The dialect has four lexical tones, among which Tone 3 (T3) has two realizations (‘allotones’). In most contexts, T3 characters (each of which corresponds to a single syllable) are realized as a low tone. However, when followed by another T3 character, the first T3 surfaces with a rising contour (i.e. the sandhi allotone). Moreover, the rising contour of this sandhi allotone is very similar to that of the lexical Tone 2 (T2), making the sandhi rising and the lexical rising ambiguous. Different dialects show different patterns of tonal variation. The following specific questions have been addressed in detail: 1) What is the range of different types of tonal variation? 2) How are the tonal variants represented in the mental lexicon? 3) When we speak, how do we retrieve words with tonal variants from the mental lexicon? 4) What sort of mental representations are activated when we read printed words with varied pronunciations? And 5) what is the unit of planning before speakers produce small utterances containing phonological variants?
We have contrived/adopted a range of experimental paradigms to tap into the questions, including detailed acoustic analyses of corpus data, the form preparation paradigm, the picture word interference paradigm, and the prepared speech paradigm, some of which are used commonly in psychological research. A series of experiments were conducted, with linguistic acoustic analysis, psycholinguistic behavioural measurements, as well as neuropsychological brain responses. In doing so, we are able to go beyond simply documenting the range of variation and the type of representation that is activated during speech processing. Instead, we can now start to ask how various types of variation can differentially affect cognitive processing of speech sounds. Furthermore, we are able to start tapping into the question of when the processing of different types of variation occurs and where in the brain.
Our results suggest that tonal variants (e.g. the canonical T3 and its rising variant) must be stored in the mental lexicon probably under the same category of lexical tone. During speech production, it is the context-specific representation that is activated, suggesting that when we speak, the brain takes into account information about the context and uses it immediately to select and activate the appropriate form. When words with phonological variants are presented visually, both are activated, suggesting that, when reading silently, the brain activates not only the form appropriate to the specific context, but also the other variant not appropriate for the context. When speakers plan upon short utterances, the unit of planning seems to be contingent upon the number of syllables rather than the way the words can be grouped according to their morpho-syntactic composition (which seems to be a more dominant pattern in West-Germanic languages). Last but not least, discourse context affects the way utterances are planned and articulated.
We have extended our empirical language base from Beijing Mandarin to other dialects (e.g. Jinan and Tianjin Mandarin, Cantonese, Shanghai Wu dialect, Jiaxing Wu dialect, Lianjiang Eastern Min dialect and Taiwanese Southern Min dialect) and languages (e.g. Germanic Dutch and Slavic Russian). These languages present very interesting supra-segmental systems, understanding of which has proved to contribute further to our central research program. These studies have also made it clear that prosodic systems can be illusive and that it requires deep understanding of the language systems for us to design good experimental work. Several contributions are of particular note here. First, we noted that the younger generations of dialectal speakers in China, given the influence of the national language – Standard Chinese, are essentially bilingual speakers of different tonal systems (sometimes in addition to segmental differences across dialects). Second, much of the existing descriptions on tonal variation in the literature has been based upon impressionistic auditory descriptions by fieldworkers (often without any additional objective acoustic data). Data collected from our project have served to verify and clarify mysteries resulting from these impressionistic (and sometimes erroneous) descriptions and provided access to solid empirical bases that will become central to the proper understanding of the variation phenomena.
Our findings challenge traditional linguistic analyses of phonological variants, which usually analyze variants as composed of underlying and surface forms where the surface forms can be computed from the underlying representations via “phonological rules”. Our data also suggest that the way words are retrieved from the lexicon in speech production and activated in speech perception differs across languages due to the different characteristics of the lexicons and the linguistic systems. Similarly, speech planning may not only have different default units across languages, but also show flexibility within a language. Together, these results challenge the one-model-for-all-languages approach in psycholinguistic research and emphasize that proper models of speech production/perception must take into consideration how different linguistic systems influence the way words are stored and processed in the human brain.
Within the project, we have addressed the central research question through the looking glass of a range of tonal variation patterns in Chinese dialects. Take Beijing Mandarin as an example. The dialect has four lexical tones, among which Tone 3 (T3) has two realizations (‘allotones’). In most contexts, T3 characters (each of which corresponds to a single syllable) are realized as a low tone. However, when followed by another T3 character, the first T3 surfaces with a rising contour (i.e. the sandhi allotone). Moreover, the rising contour of this sandhi allotone is very similar to that of the lexical Tone 2 (T2), making the sandhi rising and the lexical rising ambiguous. Different dialects show different patterns of tonal variation. The following specific questions have been addressed in detail: 1) What is the range of different types of tonal variation? 2) How are the tonal variants represented in the mental lexicon? 3) When we speak, how do we retrieve words with tonal variants from the mental lexicon? 4) What sort of mental representations are activated when we read printed words with varied pronunciations? And 5) what is the unit of planning before speakers produce small utterances containing phonological variants?
We have contrived/adopted a range of experimental paradigms to tap into the questions, including detailed acoustic analyses of corpus data, the form preparation paradigm, the picture word interference paradigm, and the prepared speech paradigm, some of which are used commonly in psychological research. A series of experiments were conducted, with linguistic acoustic analysis, psycholinguistic behavioural measurements, as well as neuropsychological brain responses. In doing so, we are able to go beyond simply documenting the range of variation and the type of representation that is activated during speech processing. Instead, we can now start to ask how various types of variation can differentially affect cognitive processing of speech sounds. Furthermore, we are able to start tapping into the question of when the processing of different types of variation occurs and where in the brain.
Our results suggest that tonal variants (e.g. the canonical T3 and its rising variant) must be stored in the mental lexicon probably under the same category of lexical tone. During speech production, it is the context-specific representation that is activated, suggesting that when we speak, the brain takes into account information about the context and uses it immediately to select and activate the appropriate form. When words with phonological variants are presented visually, both are activated, suggesting that, when reading silently, the brain activates not only the form appropriate to the specific context, but also the other variant not appropriate for the context. When speakers plan upon short utterances, the unit of planning seems to be contingent upon the number of syllables rather than the way the words can be grouped according to their morpho-syntactic composition (which seems to be a more dominant pattern in West-Germanic languages). Last but not least, discourse context affects the way utterances are planned and articulated.
We have extended our empirical language base from Beijing Mandarin to other dialects (e.g. Jinan and Tianjin Mandarin, Cantonese, Shanghai Wu dialect, Jiaxing Wu dialect, Lianjiang Eastern Min dialect and Taiwanese Southern Min dialect) and languages (e.g. Germanic Dutch and Slavic Russian). These languages present very interesting supra-segmental systems, understanding of which has proved to contribute further to our central research program. These studies have also made it clear that prosodic systems can be illusive and that it requires deep understanding of the language systems for us to design good experimental work. Several contributions are of particular note here. First, we noted that the younger generations of dialectal speakers in China, given the influence of the national language – Standard Chinese, are essentially bilingual speakers of different tonal systems (sometimes in addition to segmental differences across dialects). Second, much of the existing descriptions on tonal variation in the literature has been based upon impressionistic auditory descriptions by fieldworkers (often without any additional objective acoustic data). Data collected from our project have served to verify and clarify mysteries resulting from these impressionistic (and sometimes erroneous) descriptions and provided access to solid empirical bases that will become central to the proper understanding of the variation phenomena.
Our findings challenge traditional linguistic analyses of phonological variants, which usually analyze variants as composed of underlying and surface forms where the surface forms can be computed from the underlying representations via “phonological rules”. Our data also suggest that the way words are retrieved from the lexicon in speech production and activated in speech perception differs across languages due to the different characteristics of the lexicons and the linguistic systems. Similarly, speech planning may not only have different default units across languages, but also show flexibility within a language. Together, these results challenge the one-model-for-all-languages approach in psycholinguistic research and emphasize that proper models of speech production/perception must take into consideration how different linguistic systems influence the way words are stored and processed in the human brain.