Recognition of Polish Temporal Expressions

Proceedings of Recent Advances in Natural Language Processing, pages 282–290, Hissar, Bulgaria, Sep 7–9 2015. Recognition of Polish Temporal Expressions Jan Koco´n† G4.19 Research Group Wrocław University of Technology jan.kocon@pwr.edu.pl Michał Marci´nczuk G4.19 Research Group Wrocław University of Technology michal.marcinczuk@pwr.edu.pl Abstract In this article we present the result of the recent research in the recognition of Polish temporal expressions. The tem- poral information extracted from the text plays major role in many information ex- traction systems, like question answering, event recognition or discourse analysis. We prepared a broad description of Pol- ish temporal expressions, called PLIMEX. It is based on the state-of-the-art solutions for English, mostly TimeML specification. This solution can be used for the extraction of events and their attributes, in order to anchor events in time and to reason about the persistence of events. We prepared the annotation guidelines and we annotated all documents in Polish Corpus of Wrocław University of Technology (KPWr) using our specification. Here we describe results achieved by Liner2 machine learning sys- tem, adapted to recognise Polish temporal expressions. 1 Introduction Recognition of temporal expressions and events became an active area of the research and plays a significant role in many natural language engi- neering systems. It is one of the major tasks in in- formation extraction, which aim is to extract spe- cific elements from unstructured data. In this re- search we focus on tracking changes over time in text written in natural language. Further reason- ing about changes requires the information about temporally grounded events. Textual references to time tell us how long something lasts, when something happens or how often occurs. People are usually conscious of their location in time — in most cases we know what is the current year, month and date and we use this information to capture the meaning of ex- pressions like “yesterday”, “tomorrow”, “five days ago”, “16th of November”. Even in texts written in formal language (like newspaper articles), the global meaning of the given temporal expressions can be deduced by the analysis of the whole con- text of the document (often with metadata, such as document creation time). We can treat the global meaning of a temporal expression as a point in a timeline (e.g. “5th of December 2005”), a range (not always anchored to a specific point in a time- line, e.g. “two weeks”) or even as a set of points in a timeline (e.g. “each Tuesday”). To determine the exact date, human (or machine learning sys- tem) often must know the full temporal context. These examples do not cover the complexity of the temporal expressions understanding. Some- times a temporal expression is not the reference to the real world, but describes a fictional event. Sometimes a part of the text describes past or fu- ture, but it is not explicitly stated, but in other part of the document there are some clues to find out what tense is given. Also determining the tem- poral function of an expression can be a serious …

Content truncated to 500 words