A Study On Phonetic Features And Speech Recognition Models Of Pragmatic Functions Of Wh-words In Mandarin Chinese

Question,a product of the interaction between man and nature,is a common form of human language expression.Wh-words belong to prominent and powerful categories.Their formal means have strong expansibility and can express different categories related to their prototypical categories.Wh-words have developed non-interrogative uses related to basic uses: question function and referential function.They often express the speaker’s different verbal communication functions,such as statement,exclamation,and prohibition,etc.This complexity makes intelligent recognition of artificial speech very difficult,and intelligent recognition accuracy is not high.Therefore,an in-depth study of the pronunciation of sentences containing wh-words is helpful to the understanding of the information of sentence types,discourse structure and pragmatic meanings,and helpful to the perception and processing of discourse,especially for machines to recognize human language.By combing the crucial achievements of wh-words and speech and automatic speech recognition at home and abroad,this study used experimental phonetics and computer neural network methods to investigate the phonetic features of wh-words and their whole utterances to construct a recognition model.It reflects the relationship between the pragmatic functions of wh-words and whole utterances and acoustic features,and reveals the working mechanism of the human brain in language processing.The specific content includes the following three parts:The first part mainly investigates the usage distribution of wh-word shen2-me0.Wh-word shen2-me0 plays a unique role in the wh-word system.It is widely used and the most semantic types that can substitute for expression.It is the most basic wh-word in Chinese(Dai,2001).Based on the previous research on the wh-word shen2-me0,this part expounds the language phenomena of wh-word shen2-me0 in terms of syntax,semantics and pragmatics under three syntactic structure types of interrogative sentences,assertive sentences and exclamations,guided by the dichotomy: interrogative and non-interrogative.Through investigation,we find that “from full doubt to half doubt to no doubt”(Shao,2014)runs through the usage of wh-words.Speakers use wh-words to express their cognitive state of the world and their communicative intention: asking for information from others,informing others of information and expressing their feelings.The amount of requested information gradually decreases from the question to doubt,and the speaker’s subjectivity becomes increasing intense.At the same time,among the functional types that we examined,we have summarized scholars’ disagreements on wh-word shen2-me0’s uses,such as the interpretation of pragmatic functions when wh-word shen2-me0 and the modal particle “ma0” co-exist or not.However,these sentences,such as“ni3 chi1 le0 shen2-me0 dong1xi1 ma0?(What foods did you eat? / Did you eat anything?)and “ni3 chi1 le0 shen2-me0 dong1xi1?(What foods did you eat? / Did you eat anything?)”,are seldom misunderstood in the process of verbal communication.It can be seen that the “physical carrier”of the speaker’s utterance meaning,namely,the voice,has the function of“expressing meaning differently” and the listener can also perceive the speaker’s “true feelings” from the speaker’s voice.The second part is the production experiment.Based on the first part of wh-word shen2-me0 about the distribution of usage and existing problems,three experiments that three objects with the same syntactic structure but different functions were conducted.The first experiment is mainly aimed for the acoustic expression of the three distinct pragmatic functions(statement,interrogative and exclamation)where wh-word shen2-me0 is at the beginning and middle of utterances.The second experiment mainly focuses on the acoustic features of the pragmatic functions of wh-word shen2-me0 and its utterances generated by three different interrogative degrees in the interrogative category.In the third experiment,the modal particle “ma0” is added to the structure of the second experiment.Its main goal is to find the acoustic representation of the co-occurrence of wh-word shen2-me0 and modal particles in the expression of the pragmatic functions of wh-word.At the same time,the three target sentences in the second experiment and the third experiment respectively expressed the speaker’s utterance intention of “asking for real information”,“for uncertainty meaning” and “strongly questioning and opposing”.Through investigation of these acoustic features of wh-word shen2-me0 and its utterances,this study sums up their prosodic typicality and distinctive features when expressing different pragmatic functions,as follows:Ⅰ.Acoustic features of wh-word with different pragmatic functionsBy visual observation and the statistical analysis of pitch and duration of wh-word shen2-me0 and the whole utterances in the sum of1280 utterances spoken by ten Beijing participants(five male and five female)in Putonghua,we conclude some important distinctive phonetic features of different pragmatic functions in intonation components.In the local features,that is,wh-word,topline,baseline,range,scale and duration show obvious characteristics in the function of expressing question,statement and exclamation.In the question function,wh-word shen2-me0 have the widest pitch range,the highest point on the topline,the highest tone scale and the shortest duration.In the statement function,wh-word shen2-me0’s pitch range is the narrowest,the point on the topline is the smallest,tone scale is the lowest,and duration is the second;in the exclamation function,wh-word’s f0 shows an intermediate state and duration is the longest.In summary,the intonation parameters of wh-word shen2-me0 show a gradient distribution on the f0 and on duration: the question function > the exclamatory function> the statement function(f0)and the exclamatory function > the statement function > the question function(duration).From the perspective of overall utterances,we analyzed the suprasegmental performance of three different pragmatic functions: the boundary tone,sentence stress,whole sentence scale,whole pitch trend,whole sentence slope and whole sentence duration.In the question function the boundary tone shows a high profile(H%),wh-word shen2-me0 does not carry sentence stress,and the phenonmenon of pitch dropping occurs in the noun after wh-word shen2-me0,which infers that sentence stress falls on the noun after wh-word shen2-me0;meanwhile,the whole utterance has the highest scale,and the whole utterance expresses question meaning,the pitch trend is downward,but it tends to be flat.In the statement function,the boundary tone is low-key(L%);the noun after wh-word shen2-me0 carries the sentence stress;the whole sentence scale is the lowest,and the pitch trend drops sharply.In the exclamation function the boundary tone is low(L%);the sentence stress is also located at the noun after wh-word shen2-me0.The key of the whole utterance is in the middle by comparison with two other functions and the pitch trend is declination.Although the boundary tone of statement function and exclamation function are both low-key(L%),there is a distinction between them,that is,the boundary tone of the exclamation function is slightly higher than that of statement function.The results of observation about the sentence stress are consistent with Liu’s conclusion(2016)“in sentences with wh-words expressing declarative mood,interrogative word will never get sentence stress.”Moreover,we found that the sentence stress occurred at wh-word shen2-me0 in question mood.Although wh-word itself is prominent in the question function,it is not the focus from global acoustic features.The whole sentence expresses the question meaning as a whole,and the rising of the final tone of utterance expresses the speaker’s question and verification mentality towards the entire event.Intonation is more in line with the echo question.This can be concluded that wh-word shen2-me0’s environment in pragmatic functions is expressed as the speaker’s question,syntactically as the echo question sentence pattern,semantically as [+reference] does not acquire sentence stress.Although wh-word shen2-me0 shows similarities in the aspect of sentence stress in expressing three different pragmatic functions,there are still apparent differences and gradients in other intonation parameters.Ⅱ.Phonetic features of wh-word with different pragmatic functions under the same question moodThe expression of interrogative mood in modern Chinese has two ways: wh-word without modal particles and wh-word with modal particles.Therefore,the analysis compares the respective acoustic features within different pragmatic functions under the same question mood,and compares the acoustic features across different syntactic structures but with the same pragmatic functions.Through the investigation of 960 interrogative utterances with wh-word and 960 interrogative utterances with co-occurrence of wh-word and the modal auxiliary word,we found that different pragmatic functions under question mood have the following common features:1.When identifying different pragmatic functions expressed in the same mood,duration of wh-word itself and duration of the whole utterance is particularly prominent,which is consistent with the gradient:the speaker’s uncertainty > the speaker’s prohibition > asking for real information.2.As for the sentence stress,it tends to be consistent under two types of syntactic structures: in expressing of asking for real information,whether wh-word carries sentence stress is more complex,and it is not completely expressed as sentence stress;on the contrary,in most cases,the noun after wh-word carries sentence stress.In the function of uncertainty,although nouns behind wh-word carry sentence stress to express the speaker’s verification of events/things.In the opposition function,although the noun after wh-word carries the sentence stress,because the unreasonable event occurs again and stimulates the speaker’s dissatisfaction,so the pitch range of adverbs becomes wider than the other two functions reflecting accent phenomenon.In this pragmatic function,the adverb together with the noun after wh-word constructs the speaker’s negative or resistant emotion together with wh-word,and utterance shows broad focus.Under the influence of focus stress,while wh-word shen2-me0 is at the final initial,the lowering trend of global f0 is not apparent and tends to be flat or even rising.3.In terms of the boundary tone,the three different pragmatic functions show significant differences in the boundary tone at the final utterance: the boundary tone of uncertainty’s meaning is featured as high(H%);the boundary tone of prohibition’s meaning is featured as low(L%).However,although the boundary tone of the meaning of asking for real information is low(L%),the boundary tone is low by contrast with uncertainty’s meaning of the speaker,while the boundary tone is high by contrast with prohibition’s meaning of the speaker.The boundary tone reflects the intonation function of the last syllable at the final utterance.At the same time,the presence or absence of modal particles at the end of utterance will not change the boundary tone of three different pragmatic functional meanings,namely,uncertainty’s meaning of the speaker >asking for real information > prohibition’s meaning of the speaker.This result is the same as that of Xiong & Lin(2003)about the function of modal particle “ma0” that has not the differentiation function of mood,which instead depends on intonation.That illustrates incongruity between prosodic structure and syntactic structure.At the same time,it further illustrates the close connection between pragmatic function and acoustic features: the speaker’s pragmatic meaning is projected to sound,which reflects the speaker’s pragmatic meaning(Bolinger,1986,1989;Pierrehumbert & Hirschberg,1990;Halliday,1967;Ladd,2008;Burdin& Tyler,2018).The third part is the construction of the recognition model.The neural network and BP algorithm are adopted to analyze the acoustic data in four steps: network initialization-hidden layer to output layer calculation-output layer calculation-error calculation,and finally the topology structure is 18’6’2.The model results are stored in an XML file.In the model construction,to prevent over-training and over-fitting,the proportion of acoustic data is divided into 5: 3: 2.For speech recognition with different pragmatic functions,the recognition effect at the word level is relatively poor,while the recognition’s results of whole sentence speech data are significantly about 95%.At the same time,the contribution of maxf0,minf0,scale,pitch range and duration in distinguishing different pragmatic functions in the model construction is highly consistent with the phonetic feature analysis obtained in the production experiment.On the one hand,the recognition results of neural network model can prove the validity of acoustic data in the production experiments and the essential role the suprasegmentals play in perception.On the other hand,it shows that the neural network model can explain the different pragmatic meanings of the interrogative word shen2-me0 in utterances utilizing phonetic features and realize the identification of the polysemy of the interrogative words,which proves that the model is adequate.Taken together,we can find that the different pragmatic functions of wh-word are closely related to sound and have typical characteristics at the suprasegmental level,which provides a reference value for the language of ontology research,intelligent recognition of artificial speech,conversion of speech into text,and teaching Chinese as a foreign language.Meanwhile,the construction of the recognition model can be verified in future human-computer interaction research and further developed to solve more complicated problems.