Study on sentence relations in the automatic detection of argumentation in legal cases

From AcaWiki
Jump to: navigation, search

Citation: Raquel Mochales-Palau, Marie-Francine Moens (2007) Study on sentence relations in the automatic detection of argumentation in legal cases. JURIX 2007 (RSS)
Internet Archive Scholar (search for fulltext): Study on sentence relations in the automatic detection of argumentation in legal cases
Download: http://portal.acm.org/citation.cfm?id=1565624
Tagged: Computer Science (RSS) argument mining (RSS), argument detection (RSS), legal argumentation (RSS), natural language processing (RSS)

Summary

This builds on the work of Automatic detection of arguments in legal texts; whereas that paper used argumentative texts from multiple domains (including newspapers and social media, despite the title), this work is restricted to the legal domain. Besides detecting argumentative and non-argumentative sentences, premises and conclusions are also detected. Additional features are added to analyze the importance of relations between sentences.

A corpus is annotated then machine learning is used with selected features.

Procedure

29 admissibility reports and 25 legal cases randomly selected by European Court of Human Rights August 2006 & December 2006. These contain facts, complaints, the law, and final conclusions from judges, expressed in long and complex sentences.

These were manually analyzed by two lawyers to indicate whether they contained arguments. There were 12,904 sentences (10,133 non-argumentative and 2,771 argumentative), which included 2,355 premises and 416 conclusions.

Average accuracy of the maximum entropy model is 82%, using only the information from the current analyzed sentence. (Previous experiments used a naive Bayes model; the increased amount of information in this case meant they could not satisfy the independence assumptions of the naive Bayes classifier). They also experimented with using information in adjacent sentences.

In future work they plan to look at the clause level, instead of the sentence level.

Features

These are increased from those used in Automatic detection of arguments in legal texts

New features

Previous sentences

  • unigrams in previous sentences
  • bigrams in previous sentences
  • word couples in previous sentence
  • adverbs in previous sentence (e.g. "unfortunately")
  • verbs in previous sentences (ignoring "to be", "to do", "to have")
  • modal auxiliary in previous sentences (modals are more likely in argumentative sentences)
  • text statistics in previous sentences (same as above)
  • punctuation in previous sentences
  • presence/absence 286 argumentative keywords (see Knott &  Dale. 1992) in previous sentences
  • negative/positive previous sentences (not, don't, won't, ...)
  • first/last words in previous sentences (connectors)
  • same words in previous & current sentences

Next sentences

  • unigrams in next sentences
  • bigrams in next sentences
  • word couples in next sentence
  • adverbs in next sentence (e.g. "unfortunately")
  • verbs in next sentences (ignoring "to be", "to do", "to have")
  • modal auxiliary in next sentences (modals are more likely in argumentative sentences)
  • text statistics in next sentences (same as above)
  • punctuation in next sentences
  • presence/absence 286 argumentative keywords (see Knott &  Dale. 1992) in next sentences
  • same words in current & next sentences
  • negative/positive next sentences (not, don't, won't, ...)
  • first/last words in next sentences (conectors)
  • same words in current & next sentences

Related Work/Selected References

Work used

Linguistic Perspective on Argumentation

  • I. M. Schlesinger, T. Keren Portnoy, T. Parush. 2001. The structure of arguments. Human Cognitive  Processing, 7. Amsterdam, John Benjamins Pub.Co.

Argument Detection

A.C.  Restificar,  S.S.  Ali  and  S.W.  McRoy.  1999.  ARGUER:  Using  argument  schemas  for  argument detection and rebuttal in dialogs. In UM99:Proceedings of the Seventh International Conference  on User Modeling, 315­317. Banff, Canada. - "studies argumentative speech dialogues and propose  a system (ARGUER) that identifies if the user's utterance is attacking or supporting the  system last utterance."

Theoretical and Practical Relevance

One limitation is that this is non-dialogical, defeasible argumentation, where one issue is being addressed by a given speaker. Further research on dialogical argumentation is needed.

Retrieved from "https://acawiki.org/index.php?title=Study_on_sentence_relations_in_the_automatic_detection_of_argumentation_in_legal_cases&oldid=7546"