Research On The Application Of Linguistic Knowledge In Dependency Parsing

Parsing is a key component of natural language processing.The dependency parsing has advantages include that it's simple in form,easy to label,decoded efficiency and more closed to semantic relations.In recent years,dependency parsing has become a hot topic in the field of natural language processing.However,there are some problems such as sparse indicator features and low accuracy of long-distance dependency analysis.In order to solve the above mentioned problems,this paper attempts to and utilize linguistic knowledge include hierarchical analysis,syntactic composition analysis,verbal grammar characteristics and verb-object collocation in the process of dependency parsing.The main work includes the following:Firstly,aiming at the problem of low accuracy of long-distance dependency parsing,this paper proposes a method based on hierarchical component analysis.In this method,the components of the sentence are identified before the dependency parsing,which mainly includes the core components and the non-core components.Dependency parsing is performed within components and among components.In this way,the method can avoid generating the long-distance dependency relationships.In addition,different models are trained for different components.In CoNLL2009 corpus,the UAS value increased by 2.53 percentage points,and the LAS value increased by 2.82 percentage points.Secondly,deal with the results of dependency parsing by using linguistic knowledge.In this paper,the linguistic knowledge is used to formulate the error recognition rule.Using the rule can identify the verbs with missed dependency arcs of verb-object relation and the recognition accuracy is almost up to 93.80%.In this paper,a constraint-based decoding method is proposed to find the sub-nodes of verbs with missing verb-object relation arcs.This process uses the verb-object collocation bank building by training corpus.After post-processing,the UAS value is improved by 0.21 percentage points,and the UAS value of the verb-object relation is improved by 2.14 percentage points.Thirdly,the two methods proposed are integrated to form a dependency parser.In thefinal system,the UAS value increased by 2.7 percentage points and the LAS value increased by 3.24 percentage points.The experimental results show that using the linguistic knowledge can improve the dependency parsing effectively.