摘要
本研究以兩類研究方法（電腦模擬和認知行為研究）探討語法及語意線索對中文句之句意理解的影響；研究結果並將檢驗「SVO理論」、「標準語句形式理論」及「競爭理論」等對句意理解的預測，以進行三理論間的比較。在認知行為實驗中，本研究操弄了語法標記（「把」、「被」）和詞序等語法線索，以及詞彙間的動物對比性和語意合理性等語意線索，然後觀察受試者對實驗刺激句中「動作者」（agent）的選擇。在模擬實驗中，本研究根據過去以及本研究的認知行為實驗結果，建構類神經網路模式，模擬各種語法和語意線索對受試者選擇「動作者」的影響。在三個認知行為實驗中，本研究以大學生和小學生為受試者，詳細的操弄了刺激句包含的各種語意及語法線索，並以文字和語音等兩種方式呈現刺激。綜合來看，三個實驗結果發現，對不同型態的句子，受試者會採用不同的線索判斷句意；對NVN類型的句子，受試者傾向以詞序推斷句子「動作者」，但對NNV語句則傾向以語意線索解釋句意，對VNN語句則沒有倚賴特定線索的傾向。在模擬實驗中，雖然檢視各線索的權重顯示，受試者會根據語意線索挑選「動作者」，然而進一步分析類神經網路之隱藏層的激發型態發現，中文使用者會將不含語法標記的語句劃分為NVN和非NVN等兩種類型，但對含有「把」或「被」字的句子則不會做如上的劃分。整合本研究之電腦模擬與認知行為實驗的結果顯示，中文語句的形式會影響中文使用者使用何種線索判斷句意；詞序是影響NVN句子句意理解的主要線索，語意則對非NVN句子的句意理解有較大的影響。本研究的結果支持「標準語句形式理論」。
並列摘要
In Chinese, a noun is not singular, nor plural, and a verb has no tense or person. Although Chinese is an isolating language, word order is not a strict cue to sentence meaning. The utilities of word order, syntactic markers and other linguistic devices in sentence comprehension have been tested, and the generally finding was that there was a weak word order effect, which was predicted by the competition model. However, inconsistent results where subjects showed strong tendency to pick “kettle” as agent in sentence like “Kettle lick pig” were reported in a number of research, and discounted because the subjects were Chinese and English bilinguals. The word order preference was said a transferring effect from English. In two experiments and one simulation, the role of word order, lexical semantics, and syntactic markers in interpreting simple Chinese sentences (consisting of two nouns and a verb) was tested. Experiment 1 found that college students who had learnt English for at least 7 years tended to choose the first noun as subject when the word order of a stimulus sentence was in canonical form, i.e., NVN, irrespective to its semantic properties. However, for non-canonical sentences, the participants based their interpretations on lexical rather than syntactic cues. The same result pattern was obtained in Experiment 2 when third grade elementary students who did not speak English were tested. A back-propagated network was trained to fit the behavioral data obtained in the present and previous studies. The syntactic and semantic features of a sentence were represented in inputs, and the output was set equal to the probability the first noun was chosen as agent. Analysis of the link weights between input and hidden nodes revealed that lexical semantics had more influence than word order on the net’s performance. However, analysis of the stimuli represented in the trained net showed that different types of sentences were processed differently. The internal activation of the hidden nodes exhibited different patterns for canonical and non-canonical sentences. Word order was important to comprehend canonical sentences, and lexical semantics were important for non-canonical ones. These results which indicated that Chinese speakers relying on different cues to interpret different types of sentences were not predicted by the competition model, but were in agreement with the canonical form theory proposed by Slobin and Bever.
參考文獻
延伸閱讀
- Liu, M. (1997). 從行動動詞到連接成份：中文“就”語法化的言談模式分析. Journal of Chinese Linguistics, 25(2), 258-289. https://www.airitilibrary.com/Article/Detail?DocID=P20181204001-199706-201901310005-201901310005-258-289
- Tu, C. (2014). 使用語意詞彙網路及語法相依性分析於中文文本蘊涵關係之研究 [master's thesis, Tamkang University]. Airiti Library. https://doi.org/10.6846/TKU.2014.00887
- Chen, J. Y. (2011). 應用中文句法權重於潛在語意分析技術於中文智能教學系統之對話計分研究 [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU.2011.02712
- Chang, W. F. (2021). 利用預訓練語言遮罩模型探索語意相近用詞於換句話說生成之研究 [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU202101638
- Yeh, C. (2018). 以事件相關電位看句法處理側化、半腦間互動關係以及語言能力的相互關係 [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU201804152