Research On Temporal Relation Recognition And Inference Between Chinese Events

Under the environment of the explosive growth of information resources, information extraction has become an important method which can obtain valuable information from vast amounts of resources. Recognizing temporal relation between events is a subsequent study of event extraction, which aims to identify the temporal relation between two event mentions. It plays an important role in many Natural Language Processing(NLP) applications, such as question answering, information extraction and text summarization, etc.This dissertation focuses on following aspects: the construction of Chinese event temporal relation corpus, temporal relation recognition and inference between events. The main contents are as follows:(1) Annotating and constructing a temporal relation corpus of Chinese eventsCorpus construction is a fundamental research work in NLP. Due to the lack of Chinese event temporal relation corpus, this dissertation constructs a dense temporal relation corpus of Chinese events based on the ACE 2005 Chinese corpus and the temporal relations defined in the Time ML.(2) Supervised temporal relation recognition between Chinese eventsThis dissertation regards temporal relation recognition as a classification problem, and employs supervised method to resolve it. Several effective features, such as trigger semantics, special words, event arguments, event causality, event co-reference relation, etc. are introduced to our model. The experimental results illustrate that those features are effective in identifying temporal relation between Chinese event mentions.(3) Global inference model on temporal relation recognition between Chinese eventsA global inference model is applied to address the limitation of the above supervised methods. Temporal relation recognition is regarded as an Integer Linear Program(ILP) problem and many constraints, such as reflexivity, transitivity, event co-reference, comparison of time expressions, temporal conjunctions, pairs of event types, are applied to this model. The experimental results show that our global inference model outperforms the local strong rule-based inference method and local classifiers.This dissertation does an exploratory study on corpus construction and relation recognition for temporal relation recognition between Chinese events. Although the proposed methods are relatively simple, it will be beneficial for further research work in this filed and related fileds.