Analyzing discourse functions with acoustic features and phone embeddings: non-lexical items in Taiwan Mandarin
1017Authors
nan
Abstract
AbstractNon-lexical items are expressive devices used in conversations that are not words but are nevertheless meaningful. These items play crucial roles, such as signaling turn-taking or marking stances in interactions. However, as the non-lexical items do not stably correspond to written or phonological forms, past studies tend to focus on studying their acoustic properties, such as pitches and durations. In this paper, we investigate the discourse functions of non-lexical items through their acoustic properties and the phone embeddings extracted from a deep learning model. Firstly, we create a non-lexical item dataset based on the interpellation video clips from Taiwan’s Legislative Yuan. Then, we manually identify the non-lexical items and their discourse functions in the videos. Next, we analyze the acoustic properties of those items through statistical modeling and building classifiers based on phone embeddings extracted from a phone recognition model. We show that (1) the discourse functions have significant effects on the acoustic features; and (2) the classifiers built on phone embeddings perform better than the ones on conventional acoustic properties. These results suggest that phone embeddings may reflect the phonetic variations crucial in differentiating the discourse functions of non-lexical items.
Keywords
Full Content
Analyzing discourse functions with acoustic features and phone embeddings: non-lexical items in Taiwan Mandarin Pin-Er Chen National Taiwan University cckk2913@gmail.com Yu-Hsiang Tseng National Taiwan University seantyh@gmail.com Chi-Wei Wang National Taiwan University r09142007@ntu.edu.tw Fang-Chi Yeh National Tsing Hua University fangchiyeh2000@gmail.com Shu-Kai Hsieh National Taiwan University shukaihsieh@ntu.edu.tw Abstract Non-lexical items are expressive devices used in conversations that are not words but are nevertheless meaningful. These items play crucial roles, such as signaling turn-taking or marking stances in interac- tions. However, as the non-lexical items do not stably correspond to written or phono- logical forms, past studies tend to focus on studying their acoustic properties, such as pitches and durations. In this paper, we investigate the discourse functions of non- lexical items through their acoustic proper- ties and the phone embeddings extracted from a deep learning model. Firstly, we create a non-lexical item dataset based on the interpellation video clips from Taiwan’s Legislative Yuan. Then, we manually iden- tify the non-lexical items and their dis- course functions in the videos. Next, we an- alyze the acoustic properties of those items through statistical modeling and building classifiers based on phone embeddings ex- tracted from a phone recognition model. We show that (1) the discourse functions have significant effects on the acoustic features; and (2) the classifiers built on phone embeddings perform better than the ones on conventional acoustic properties. These results suggest that phone embed- dings may reflect the phonetic variations crucial in differentiating the discourse func- tions of non-lexical items. Keywords: non-lexical item, discourse function, acoustic property, acoustic represen- tation, pragmatics 1 Introduction People’s everyday interactions include sounds that are not verbal words in the traditional sense. These sounds, such as sighs, sniffs, and grunts, are used in indexing the turn-taking in dialogues, marking stance, showing affections, and expressing roles and meanings in conversa- tions (Dingemanse, 2020). Examples of these non-lexical items are un-huh in English as a marker showing understanding and attentive- ness, while the single syllable uh and um act as fillers and disfluency markers (Ward, 2006; Buschmeier et al., 2011). While these non-lexical items are important linguistically, they pose an interesting chal- lenge to linguistic inquiry. Non-lexical items do not belong to a major word class, and some do not conform to the language’s phonologi- cal requirements (Keevallik and Ogden, 2020). Moreover, while the phonetic properties of non-lexical items could be generally described, they are nevertheless “phonetically underspeci- fied.”(Keating, 1988) For example, in the study of “moan” in board game interactions, Hof- stetter (2020) found “moans” involve phonetic properties related to open vowels, irrespective of their frontness, backness, or roundedness. The study suggests that a non-lexical item can not be represented as a single phonetic sym- bol; instead, it may refer to the vowel space for which we do not have a general phonetic symbol. Some studies, therefore, analyze these items in terms of their acoustic properties: the components’ sound (Ward, 2006), the funda- mental frequencies, durations, and intensities. (Shan, 2021; Ballier and Chlébowski, 2021). In contrast to the conventional …
Content truncated to 500 words
Labeled Text
AbstractNon-lexical items are expressive devices used in conversations that are not words but are nevertheless meaningful. These items play crucial roles, such as signaling turn-taking or marking stances in interactions. However, as the non-lexical items do not stably correspond to written or phonological forms, past studies tend to focus on studying their acoustic properties, such as pitches and durations. In this paper, we investigate the discourse functions of non-lexical items through their acoustic properties and the phone embeddings extracted from a deep learning model. Firstly, we create a non-lexical item dataset based on the interpellation video clips from Taiwan’s Legislative Yuan. Then, we manually identify the non-lexical items and their discourse functions in the videos. Next, we analyze the acoustic properties of those items through statistical modeling and building classifiers based on phone embeddings extracted from a phone recognition model. We show that (1) the discourse functions have significant effects on the acoustic features; and (2) the classifiers built on phone embeddings perform better than the ones on conventional acoustic properties. These results suggest that phone embeddings may reflect the phonetic variations crucial in differentiating the discourse functions of non-lexical items.
Publication Details
- Paper ID:
- 1017
- Year:
- 2022
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan