Analysis Of Structural Ambiguity In English Adjunct Chunking

Chunking is a main preliminary task in the field of Machine Translation(MT).However,because of the complexity of syntactic structure,there exists a lot of ambiguity in the automatic identification of chunks.The adjunct chunks with flexible position and multiple functions usually cause a lot of misunderstandings and ambiguity in structure and semantic meaning.The structural ambiguity,especially the identification of adjunct chunks,is a difficult problem and the relevant researches on this field are very seldom.Therefore,the analysis of structural ambiguity in English adjunct chunking plays a guiding role for the machine identification field to break its bottleneck condition and improve the quality of machine identification to some extent.The reasons causing the structural ambiguity in English adjunct chunking are waiting to be explored.Hence,this thesis compares the results of machine identification(MI)of English adjuncts with the results of human identification(HI)based on the syntactic analysis of Halliday's Systemic Functional Grammar.It aims to study the main reasons of structural ambiguity in English adjunct chunking and then puts forward some suggestions to reduce and even eliminate the structural ambiguity in MI.In the process of the research,firstly,for the machine identification,three thousand sentences from a self-built corpus in the domain of business English involving 4190 adjunct chunks are chosen to be automatically identified.Then,the results of machine identification are analyzed.It's been found the structural ambiguity mainly results from four categories of adjunct structure which are prepositional phrases causing the most errors,and then nominal groups,adverbial groups and verbal groups.Then this thesis compares the results of HI and MI to further analyze the difficulty of machine identification and main reasons for structural ambiguity.For human identification,20 sentences that involve most errors of machine identification are adopted to ask students to conduct the chunking.The first step is to conduct a pre-test concerning the students' identification of English adjuncts in order to make sure the experiment is valid and can be carried out continuously.Next,200 students from Dalian University of Technology are asked to identify the adjuncts in the 20 sentences and then a questionnaire is proceeded to ask them how they figure out the adjuncts in the sentences.Then the author makes a comparison between MI and HI.The results show that students also make the most errors in the prepositional phrases and the errors of four categories of adjunct structure also appear in students' identification,but the error rate of HI is lower than MI.It's been found the placement of the adjunct and complexity of sentence structure are the main reasons to cause the structural ambiguity which affects the quality of machine identification and the adjunct learning and understanding of students.This thesis also makes some suggestions for the improvement of the machine performance and English language learning.