| Null Instantiation is the core frame element which is neither expressed as a dependent of the predicator nor can it be found through gap filling in FrameNet. Definite Null Instantiation is those who has implicit express in sentence and can establish links between local semantic arguments, which also referred as DNI. Gap Filling aim to find an explicit expression for DNI in context. So the gap filling of definite null instantiation aim to find a potential explicit filler for missing frame element, which play a critical role to deep semantic analysis and deducing kernel dependency graph in natural language processing.In FrameNet, gap filling of definite null instantiation is often regarded as a special task of Semantic Role Labeling. Traditional semantic role labeling only focus on roles which have explicit expression while ignore the implicit arguments, which leads to lose a lot of important semantic information at last. In addition, it was proved that the missing semantics play a crucial role in the research of relation between locally semantic arguments. Therefore, finding those missing semantic roles has positive significance not only in promoting the text understanding, but also lead to substantial increase in performance of many Natural Language Processing applications.Our corpus provided from the NI task of SemEval-2010 Task 10, this paper research the gap filling of definite null instantiation in the corpus which has annotated the type of null instantiation already based on machine learning method. The main research achievement contains:(1) This paper define a rule to choose candidate words for DNI based on counting the part-of-speech of DNI fillers in train data, and determine the search space of candidate set based on experimental comparison. At last, this paper combine statistics and rules to choose the best candidate words set which has largest coverage area but smallest scale.(2) Based on classification, this paper introduces semantic features into gap filling of definite null instantiation, chooses twelve features of semantic and syntactic, and verifies the effectiveness of semantic features in gap filling of definite null instantiation.(3) Apply the best candidate words set and features to gap filling of definite null instantiation in test data of NI Task, achieve the automatic semantic label of definite null instantiation at last.The fruits of this paper further enrich methods of gap filling of definite null instantiation, and promote this task to semantic level. Experiments show that combining syntactic and semantic is helpful to gap filling of definite null instantiation, this paper also provide a new technical method for deep discourse semantic analysis in natural language processing. |
