Research On Function Representation And Measurement Of Modern Chinese Function Words Based On Chinese AMR

Function words are crucial to automatic meaning analysis due to its grammatical functions in Chinese.However,there are still some problems,which make it difficult to use the information of function words in the field of Chinese information processing,especially in the automatic analysis of syntax and semantics,including:(1)The classification of function words in the field of Chinese information processing complies with traditional word class system that is applied in teaching methodology.Although function words are classified according to their functions,there are some disputes,which makes it more difficult to annotate function and utilize resources.(2)There is no proper formal expression to express function,and rules of function words are only written according to parts of speech of their collocation words.(3)Knowledge bases of function words is scarce,and the annotation information of function words in other resources is uneven,and it is difficult to give the probability of each function word under each function.The following researches have been carried out to try to express the syntactic and semantic information of function words and solve above issues.(1)This paper delimits the scope of function words according to their grammatical functions and determines that prepositions,conjunctions,auxiliary words and modal particles are included.Considering similarities and differences of these four categories,function words are classified into two categories: Modal particles and some auxiliary words are one category expressing concepts;prepositions,conjunctions,and some auxiliary words are the other corresponding to relations between concepts.Thus,ensuring the function standard,it gets rid of the controversy caused by traditional word class system.Besides,new classification can correspond to traditional one,which highlights function,and facilitates the subsequent resource construction and application.(2)This paper introduces Chinese AMR under the lack of formal representation.Function words representing concepts and relations are annotated in two ways.The former is annotated as a node and the latter is annotated on a directed arc,which makes annotation more reasonable.Secondly,it designs a new numbering scheme,which makes words of sentences align with concepts and relations and shows function directly.Besides,it has expanded semantic relations,and concepts can be added and deleted flexibly,which ensures that function can be expressed more completely.(3)This paper constructs a Chinese AMR corpus with 8,586 sentences and a knowledge base with 19,803 pieces of function word information due to scarce resources.It analyzes the distribution and specific functions.Firstly,function words are very common,whose words corresponding to relations between concepts are more frequently used.Functions words can be founded in 89.42% sentences,37.43% of which express concepts and 62.57% of which correspond to the relations between concepts.Secondly,there are other words between some function words and the words on their related nodes.Recognizing usages and functions of function words requires syntax and semantic information.80.27% auxiliary words on nodes and 44.76% modal words are next to the words on their parent nodes.82.55% auxiliary words on arcs,57.46% prepositions and 39.75% conjunctions are next to words on the corresponding nodes.Thirdly,collocations of word categories in related nodes are found.Parent nodes of 93.94% auxiliary words annotated on the node and 80.34% modal words are verbs.As regard to corresponding nodes and parent nodes of function words that annotated on arcs,64.10% prepositions are “noun-verb” collocation,while 54.67% of conjunctions are “verb-verb” collocation due to compound sentences.Fourthly,functions of function words vary a lot.It is found that auxiliary words representing concepts are mainly aspect auxiliary,and 91.91% aspect information in sentences are expressed by three auxiliary word types.Participation ratios and types of function words are different when they represent different mood,among which participation ratios of judgement and imperative moods are the highest.Auxiliary words that correspond to relations are mainly structural auxiliary words.And “de” is the most frequently used and can express the most functions.Prepositions have flexible functions,while conjunctions are more fixed.Finally,the paper integrates positions,related nodes,examples,and other information of function words,and constructs a knowledge base where the dynamic probability information of each function word can be obtained.This also lays a foundation for future research and application.