A Study on Phonological Coherence and Holistic Processing of Chunks in Mandarin Spontaneous Speech

漢語自發話語中語塊的語音連貫和整體加工研究

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
  • Hua Chen (External person) (External Supervisor)
  • Bin LI (Supervisor)
Award date7 Sept 2022

Abstract

Chunks are multi-word sequences with independent meaning and function, or formulaic based on the intuition of native speakers. They are ubiquitous in our daily expressions and of various categories, including collocations, phrasal structures, conversational conventions, etc., and profoundly influence language use and acquisition.

Chunks are suggested to be stored as a single unit in the mental lexicon, holistically restored and retrieved due to their frequent use and fixed internal structure. Previous studies have identified the processing advantages of chunks through psycholinguistic experiments of eye-movements and self-paced reading but did not provide direct evidence of the existence of holistic storage and processing. Meanwhile, researchers suggest that the psycholinguistic state of a speaker can be revealed by prosodic cues in continuous speech and propose phonological coherence of chunks, i.e., chunks are usually articulated within a single intonation contour and resistant to inner pauses and dysfluencies. Phonological coherence as a criterion to detect the holistic nature of multi-word sequences has been proved robust in the co-occurrence of intonation units and chunks in English spontaneous speech. However, previous studies on phonological coherence usually lacked consideration of chunk categories and did not consider the influence of speech styles on the realization of speech prosody and pause placement. Moreover, research on the prosodic realization of chunks is scarce in Mandarin.

Therefore, based on the hypothesis of holistic storage and retrieval and phonological coherence, this thesis aims to explore the processing issues of Mandarin chunks through their prosodic realization in spontaneous speech. It attempts to answer two major research questions:

1. Are chunks phonologically coherent in Mandarin spontaneous speech?
2. Are chunks holistically processed in Mandarin spontaneous speech?

Firstly, the study examined the relationship between chunks and intonation phrases and explored factors that affected their alignments. It further examined the influence of chunk category and speech genre on the co-occurrence of chunks and intonation phrases.

Secondly, the study investigated the co-occurrence of chunks and processing units, defined as the utterance sequences divided by pauses, and examined the pause placements before and in the middle of chunks. The study further explored the influence of chunk production on the co-occurrence of processing units and intonation units and discussed the effect of speech genre on the psycholinguistic processing of chunks.

The data for the current study selected Mandarin monologues in formal and informal settings with an effective length of 1 hour and 20 minutes for each. Two independent groups of judges were invited to extract the target units. Each group was naive about the research objectives, and the judges for chunk identification were unaware of the acoustic information of the data. Six Mandarin natives were first trained on chunk definition, chunk taxonomies, and identification criteria and then independently identified chunks by the transcribed texts without any indication of speech segmentation. Chunks were selected according to the agreement by at least four of the judges. Acoustic annotations of intonation phrases and intermediate phrases were done independently by a professional phonetician specialized in Mandarin phonetics. Pauses were annotated by the author of the thesis.

The findings on phonological coherence showed that, firstly, chunks were generally inclined to be placed within an intonation phrase, indicating that Mandarin chunks are phonological coherent. However, the ratio of total boundary alignment between chunks and intonation phrases was relatively low. The misalignments between chunks and intonation phrases were caused by structural and functional factors. Secondly, major chunk categories (collocations, frame constructions, and institutionalized expressions) co-occurred with intonation phrases in significantly different ways. Frame constructions were more likely to cross intonation phrase boundaries. In contrast, institutionalized expressions showed higher tendencies to align with intonation phrase boundaries. Subordinate chunk categories also showed significant differences in co-occurring with intonation phrases. Semi-fixed chunks, such as phrasal constraints, were more likely to cross intonation boundaries. This indicates that the chunk properties, including fixedness, grammatical level, continuity, and meaning opaqueness, significantly influenced the intonational encoding of chunks. Moreover, chunks tended to be situated in both formal and informal speech genres, and speech genre significantly influenced the co-occurrence of chunks and intonation phrases, as informal speech contained more total alignment cases. In contrast, formal speech had more chunks crossing the intonation phrase boundaries. This shows the critical effect of speech formality and discoursal topics on the phonological coherence of chunks. Meanwhile, major chunk categories exhibited significantly different co-occurrence patterns in the two speech genres. By examining subordinate chunk categories, only fixed collocations, direct collocations, phrasal constraints, and pragmatic markers showed a significant difference in co-occurrence with intonation phrases between genres. This indicates different ways of intonation encoding of these chunk categories under different speech genres and the universality of phonological coherence of the rest of the subordinate chunk categories across speech genres.

The study on holistic processing has found that, firstly, chunks were mainly contained within processing units. However, the total co-occurrence of chunks and processing units was relatively low, not supporting the claims that chunks are processing units in spontaneous speech production. Major chunk categories showed a significant difference in co-occurring with processing units, indicating different psycholinguistic processes among major chunk categories. Institutionalized expressions were mainly situated within processing units, and frame constructions showed higher possibilities of crossing the processing unit boundaries. In addition, subordinate chunk categories were found to co-occur with processing units in different patterns, indicating the influence of chunk properties of continuity and meaning transparency on the psycholinguistic processing of chunks. From the perspective of pause placement, major chunk categories differed significantly in pause placement before chunks, as frame constructions were more likely to receive grammatical pauses, indicating that different chunk categories undergo different psycholinguistic processes. However, major chunk categories showed no significant difference in the possibility of pre-chunk hesitation, demonstrating that the cognitive fluency of chunks is universal across chunk categories. Meanwhile, major chunk categories were significantly different in pause placement within chunks. Collocations and frame constructions received internal hesitations at different degrees, while no internal dysfluencies institutionalized expressions were observed. This again justifies the different psycholinguistic processes among chunk categories. Secondly, compared to intermediate phrases, when chunk production was involved, intonation phrases contained more total co-occurrence cases with processing units. This justifies intonation phrases as a more reliable domain of psycholinguistic processing in spontaneous speech production. In addition, chunk categories significantly influenced the co-occurrence of processing units and units of intonation, as the production of frame constructions led to more misalignment cases. These patterns were found in both levels of the intonation hierarchy. By examining hesitation placement, pre-chunk hesitations mainly occurred at the intonational boundaries, while internal chunk hesitations were more found within intonation phrases and intermediate phrases by contrast. This indicates that speakers tend to maintain phonological coherence in chunk production when encountering processing difficulties, which proves the psycholinguistic reality of chunks’ holisticity. Thirdly, chunks were inclined to be situated within one processing unit in the formal and informal speech data. Speech genre significantly influenced the co-occurrence of chunks and processing units, as chunks showed a higher rate of total co-occurrence with processing units in informal speech. By contrast, formal speech data exhibited more chunks crossing the processing unit boundaries. From the perspective of chunk categories, major chunk categories showed a higher rate of total co-occurrence with processing units in informal speech. In addition, only fixed collocations, phrasal constraints, and pragmatic markers were found to co-occur with processing units in significantly different patterns under different genres. Additionally, the formal and informal speech genre significantly differed in hesitation types, as formal speech had more filled pauses. This suggests that speech formality and discourse topics greatly influence the psycholinguistic processing of chunks and support the universality of holistic processing of chunks across speech genres.

To conclude, the current study has proved and explained the phonological coherence of Mandarin chunks in spontaneous speech from the perspectives of chunk categories and speech genre. Meanwhile, it provided empirical evidence of holistic processing from the approach combining prosodic and pause placement analysis. The study has furthered the hypothesis of phonological coherence by discussing the relationship among chunks, units of intonation, and processing units, which contributes to our understanding of mental processing patterns of speech production. The comprehensive approach of combining intonation analysis and pause analysis proves a valid alternative to probe into the mental processing of speech production. The findings of this study provide implications to theories on chunks and the syntactic-prosody interface and contribute to implications for the design of Mandarin instructions and teaching, as well as the study of natural language processing.
語塊是一個具有獨立意義或功能的基於本族語者語言直覺被程式化的多詞序列。它们泛存在於我們日常用語之中,涵蓋了搭配、短語框架、會話用語等諸多類別,並且深刻的影響著我們的語言使用和語言學習。

以往對漢語語塊的研究聚焦於對語塊的分類和心理加工表徵的探討。語塊由於較高的使用頻次或是內部結構的凝固性,被認為是像一個單詞一樣在我們的頭腦中被整體儲存和加工。相關研究通過眼動、自定步速閱讀等心理學研究方式證實了語塊單位的加工優勢,然而卻很難直接證明語塊整體性的心理特質。此間,有學者提出可以通過韻律線索來探究說話人的心理加工狀態。他們指出語塊具有語音連貫的特性,即語塊經常通過一個語調單位進行產出,內部極少有停頓和不流暢現象出現。語音連貫作為探究多詞序列整體性的重要線索在英語自發話語中得到了一定程度的證明,然而以往對語塊語音連貫的檢驗忽略了對語塊類別的考慮,同時也缺少考量語體對韻律和停頓分布的影響。漢語語塊同樣被認為具有整體儲存整體加工的特質,然而文獻中鮮有對漢語語塊語音連貫進行探討。

據此,本研究基於語塊的整存整取的假設和語音連貫的概念,從語塊自然語流實際產出的韻律情況來討論語塊的加工問題,並提出了以下兩個主要問題:

1.漢語自然語流中語塊是否具有語音連貫的特性?
2.漢語自然語流中語塊是否被整體加工?

首先,本研究檢驗了語塊和語調單位的共現關係,探究了導致語塊跨越語調短語邊界的因素。同時,本研究進一步考察了語塊類別和語體兩個因素對語塊和語調短語共現的影響。

其次,本研究考察了語塊和加工單位的共現情況以及停頓在語塊產出過程中的分布,加工單位被界定為由停頓和遲疑標記劃分的話語序列。本研究繼而探究了語塊產出對加工單位和語調單位共現的影響,以及語體對語塊心理加工的影響。

本研究選取正式(電視社會評論)和非正式(電影節脫口秀主持詞)的漢語自發話語各1小時20分鐘,本研究採用三組相互獨立的參與者來對研究目標進行提取,確保語調單位標注和語塊識別的參與者都不瞭解研究目的,同時保證語塊識別者不接觸語料的語音信息。研究採用本族語者識別的方式進行語塊識別,6名漢語本族語者首先接受關於語塊的定義、語塊的分類以及識別標準的培訓,隨後根據沒有話語切分提示的文本獨立完成識別任務。四名及以上參與者共同認可的多詞單位被作為此研究的目標語塊。語音語料由專業的漢語語音學研究者獨立完成,並根據語調邊界的語音指徵,將語調單位劃分為語調短語和中間短語。遲疑標記則根據語料信息由研究者獨立完成。

關於語音連貫的研究發現,首先從整體來看,語塊更傾向於處於語調短語 內部,這體現出語塊具有語音連貫的特性,但是語塊和語調短語邊界完全對齊的 概率相對較低。語塊和語調短語不一致的情況可能由語塊的結構以及功能因素導 致。其次,主要語塊類別(搭配、框架構式以及機構化表達類語塊)和語調短語 的共現存在顯著性差異,框架構式更容易跨越語調短語邊界而機構化表達和語調 短語的共現程度相對較高。這體現出語塊的固定性、連續性以及語義的不透明性 等特質對語塊的語調編碼有著顯著的影響。次級語塊類別同樣和語調短語的共現 具有顯著差異,半固定的語塊如限制性短語更容易跨越語調邊界。這進一步證明 瞭語塊的不同特性對於其語音連貫的影響。此外,語塊在正式和非正式語體中都 傾向於處在一個語調短語中,而語體對語塊和語調短語的共現有著顯著性的影響,非正式語體包含有更多完全對齊的情況而正式語體包含有更多的語塊跨語調短 語邊界的情況。這表明話語的正式程度以及話題對於語塊的語音連貫有著重要的 影響。同時,不同語體中主要語塊類別和語調單位的共現存在顯著差異。在所檢 驗的次級語塊類別中只有固定搭配、直接搭配、限制性短語和語用標記語和語調 短語的共現存在語體間的差異,這表明語塊的語調編碼與語體有著密切關係,同 時一定程度上支持語塊語音連貫在語體間具有普遍性。

關於整體加工的研究發現:第一,語塊主要處於加工單位內部,但是語塊和加工單位完全共現的程度並不高,這可能不能支持將語塊作為話語產出的加工單位的論斷。主要語塊類別和加工單位的共現有著顯著的不同,這表明不同語塊類別具有不同的心理加工過程。機構化表達都處於一個加工單位內部,而框架構式語塊跨越加工單位邊界的概率更高。此外,次級語塊類別同樣與加工單位有著不同的共現形式,體現出連續性和語義透明性等特質對語塊心理加工的影響。從停頓分布的角度來看,主要語塊類別在語塊前的停頓分布有著顯著性差異,框架 構式語塊會接受更多的語法停頓。這顯示出不同語塊類別的具有不同的加工過程。然而,主要語塊類別前出現遲疑的概率相近,同樣的規律也體現在次級語塊類別 當中,體現出語塊認知流利度普遍存在於各個語塊類別當中。同時,主要語塊類 別在其內部的停頓分布方面有著顯著性差異。搭配和框架構式語塊內部具有不同 程度的遲疑,而機構化表達中沒有觀察到話語不流暢現象。這再次證明不同語塊 類別產出過程中的心理加工方式不同。第二,相較於中間短語,當涉及語塊產出 時語調短語和加工單位更吻合,證明語調短語是自然話語產出中的一個更為可靠 的加工範疇。另外,語塊類別顯著影響了加工單位和語調短語的共現,包含框架 構式語塊的加工單位和語調短語共現程度較低。這樣的分布在兩個語調層級中都 存在。從遲疑分布來看,語塊前的遲疑主要發生在語調單位的邊界,而更多語塊 內部的遲疑發生在了語調短語和中間短語內部。這表明說話者在遇到加工困難時 傾向於保持語塊產出的語音連貫性,證明瞭語塊的整體加工的心理現實。第三,語塊在正式和非正式語體中都傾向處於一個加工單位內部。語體對語塊和加工單 位的共現有著顯著影響,在非正式話語中語塊和加工單位的共現程度更高,而正 式話語中語塊跨語調邊界的比例更高。從語塊類別來看,非正式話語中主要語塊 類別和加工單位的完全共現程度更高。此外,只有固定搭配、限制性短語和語用 標記語類語塊在不同語體中和加工單位的共現存在語體間的顯著差異。此外,正 式和非正式話語中語塊內部的遲疑类型具有顯著差異,正式話語包含更多的填充 停頓。這表明話語正式程度和話題對語塊的心理加工有著密切的影響,同時一定 程度上支持語塊整體加工在不同語體中具有普遍性。

綜上所述,本研究從語塊類別和話語類型角度證明和解釋了自然話語產出中漢語語塊的語音連貫特質,並採用語調分析和停頓分析相結合的方法為語塊的整體加工提供了實證依據。本研究進一步將語音連貫的假設拓展到對語塊、語調單位和加工單位之間關係的討論,增進了對話語產出加工模式的瞭解。語調分析和停頓分析相結合的方法為探討話語的心理加工提供了有效的研究視角。本研究的發現為語塊以及句法韻律界面的相關理論提供了啓示,並有助於漢語教學設計和自然語言處理的相關研究。