Automatic Identification of Rhetorical Roles using Conditional Random Fields for Legal Document Summarization

Automatic Identification of Rhetorical Roles using Conditional Random Fields for Legal Document Summarization M. Saravanan Department of CS & E IIT Madras, Chennai-36 msdess@yahoo.com B. Ravindran Department of CS & E IIT Madras, Chennai-36 ravi@cse.iitm.ac.in S. Raman Department of CS & E IIT Madras, Chennai-36 ramansubra@gmail.com Abstract In this paper, we propose a machine learning approach to rhetorical role identification from legal documents. In our approach, we annotate roles in sample documents with the help of legal experts and take them as training data. Conditional random field model has been trained with the data to perform rhetorical role identification with reinforcement of rich feature sets. The understanding of structure of a legal document and the application of mathematical model can brings out an effective summary in the final stage. Other important new findings in this work include that the training of a model for one sub-domain can be extended to another sub-domains with very limited augmenta- tion of feature sets. Moreover, we can significantly improve extraction-based summarization results by modifying the ranking of sentences with the importance of specific roles. 1 Introduction With the availability of large number of colossal legal documents in electronic format, there is a rising need for effective information retrieval tools to assist in organizing, processing and retrieving this information and presenting them in a suitable user-friendly format. To that end, text summariza- tion is an important step for many of these larger information management goals. In recent years, much attention has been focused on the problem of understanding the structure and textual units in legal judgments (Farzindar & Lapalme, 2004). In this case, performing automatic segmentation of a document to understand the rhetorical roles turns out to be an important research issue. For instance, Farzindar (2004) proposed a text summarization method to manipulate factual and heuristic knowledge from legal documents. Hachey and Grover (2005) explored machine learning approach to rhetorical status classification by performing fact extraction and sentence extraction for automatic summarization of texts in the legal domain. They formalized the problem to extract most important units based on the identification of thematic structure of the document and determina- tion of argumentative roles of the textual units in the judgment. They mainly used linguistic features to identify the thematic structures. In this paper, we discuss methods for automatic identification of rhetorical roles in legal judgments based on rules and on machine learning techniques. Using manually annotated sample documents on three different legal sub-domains (rent control, income tax and sales tax), we train an undirected graphical model to segment the documents along different rhetorical structures. To represent the documents for this work, we mainly used features like cue words, state transition, named entity, position and other local and global features. The segmented texts with identified roles play a crucial part in re-ordering the ranking in the final extraction-based summary. The important sentences are extracted based on the term distribution model given in [Saravanan et al, 2006]. In order to develop a generic approach to perform segmentation, …

Content truncated to 500 words