Using Semantic Networks to Identify Temporal Expressions from Semantic Roles

International Conference RANLP 2009 - Borovets, Bulgaria, pages 219–224 Using Semantic Networks to Identify Temporal Expressions from Semantic Roles Hector Llorens, Borja Navarro, Estela Saquete Natural Language Processing Research Group University of Alicante, Spain {hllorens,borja,stela}@dlsi.ua.es Abstract Nowadays, the temporal aspects of natural lan- guage are receiving a great research interest. TimeML has been adopted as a standard for tem- poral information annotation by a large number of researchers. Available TimeML resources are very limited in size and in diversity of languages. This paper analyzes a combination of semantic roles and semantic networks information for im- proving this situation. An automatic approach using semantic networks to convert temporal se- mantic roles into TimeML TIMEX3 elements is presented. This approach has been quantita- tively evaluated for English and Spanish. The results point out that the presented approach can help in a semi-automatic creation of TimeML re- sources for the evaluated languages and could be also valid for other European languages. Keywords TimeML, TimeBank, TE identification, Semantic Roles 1 Introduction In recent years, the research interest on automatic treatment of temporal information of natural language (NL) text has experienced an important growth [8]. One of the main reasons for that are the benefits that temporal information brings to Question answer- ing (QA), summarization and many other natural lan- guage processing (NLP) areas [17]. Specialized work- shops and conferences [12, 15], and evaluation forums [20, 21] reflect the importance of this field. Further- more, the development of language independent sys- tems has become an important issue among NLP com- munity. This has been reflected in many conferences such as CLEF1, as well as in works specific to Tempo- ral Expression (TE) recognition field [24, 10]. In this paper, we present an approach to temporal expression identification from a multilingual point of view. There are different ways to represent temporal infor- mation in NL. One of them is TimeML [13], which has been recently adopted as de facto standard annotation scheme by a large number of researchers [17]. A different way to represent time is defined in Se- mantic role labeling (SRL). SRL consists of determin- ing basic event structures in a sentence, detecting se- mantic relations among entities and events. The tem- 1 European Cross-Language Evaluation Forum poral information of the events is represented by the temporal semantic role. SRL field has achieved impor- tant results in the last years [5]. Currently, the major problem of TimeML lies on the lack of resources, specially the lack of corpora for lan- guages other than English. Specifically, this work is fo- cused on the benefits that available semantic networks and semantic roles corpora can introduce to TE identi- fication. To achieve the proposed objective, we present an automatic system that identifies TimeML TEs from semantic roles using semantic networks as validation method. Furthermore, this system is designed to han- dle the task multilingually, provided that there are se- mantic networks and semantic roles resources available for the target language. To measure the performance and the possibilities …

Content truncated to 500 words