The BIUTTE Research Platform for Transformation-based Textual Entailment Recognition

The BIUTEE Research Platform for Transformation-based Textual Entailment Recognition Asher Stern and Ido Dagan1 Recent progress in research of the Recognizing Textual Entailment (RTE) task shows a constantly-increasing level of complexity in this research field. A way to avoid having this complexity becoming a bar- rier for researchers, especially for new-comers in the field, is to provide a freely available RTE system with a high level of flexibility and ex- tensibility. In this paper, we introduce our RTE system, BiuTee2, and suggest it as an e↵ective research framework for RTE. In partic- ular, BiuTee follows the prominent transformation-based paradigm for RTE, and o↵ers an accessible platform for research within this ap- proach. We describe each of BiuTee’s components and point out the mechanisms and properties which directly support adaptations and integration of new components. In addition, we describe BiuTee’s vi- sual tracing tool, which provides notable assistance for researchers in refining and “debugging” their knowledge resources and inference components. 1 Introduction and Background Textual inference (the ability to automatically find conclusions that can be inferred from a natural language text) is a capability required for many tasks at the semantic level of Natural Language Processing (NLP). For example, a typical Information Extraction (IE) task may be to extract, from a natural language text, the employer-employee 1Bar-Ilan University, Ramat-Gan, Israel 2BiuTee: Bar Ilan University Textual Entailment Engine. It is freely available at http://www.cs.biu.ac.il/~nlp/downloads/biutee/ 1 LiLT Volume 9 Perspectives on Semantic Representations for Textual Inference. Copyright c ⃝ 2014, CSLI Publications. 2 / Asher Stern and Ido Dagan relationship. Such a task can be formalized as the task of identifying text fragments from which it can be concluded that “X is employed by Y” for some entities X and Y. Similarly, a typical Question Answering (QA) task might be the task of finding answers to the question “By whom X is employed”, for some entity X (a person, in this case). Ad- dressing both these examples requires a mechanism that recognizes that the text fragment “X is employed by Y” can be inferred from a given text. Recognizing Textual Entailment (RTE) unifies this concept, and aims to serve as a common paradigm for textual inference. By adopting the RTE paradigm, the e↵orts required to solve the problem of textual inference need not to be duplicated for multiple NLP tasks which re- quire this capability. Rather, the aim is to develop a generic solver for the RTE task, which can then be used as an inference component in task-specific systems. The formal definition of the Recognizing Textual Entailment task is as follows. Given two text fragments, one termed text and the other hy- pothesis, the task is to recognize whether the hypothesis can be inferred from the text (Dagan et al. 2006a). Since first introduced, several approaches have been proposed for this task, ranging from shallow lexical similarity methods (e.g., Clark and Harrison 2010; MacKinlay and Baldwin 2009), to complex linguistically- motivated methods, which incorporate extensive linguistic analysis (syntactic parsing, coreference resolution, semantic …

Content truncated to 500 words