Automatic Charge Identification from Facts: A Few Sentence-Level Charge Annotations is All You Need

Proceedings of the 28th International Conference on Computational Linguistics, pages 1011–1022 Barcelona, Spain (Online), December 8-13, 2020 1011 Automatic Charge Identification from Facts: A Few Sentence-Level Charge Annotations is All You Need Shounak Paul IIT Kharagpur shounakpaul95@kgpian.iitkgp.ac.in Pawan Goyal IIT Kharagpur Saptarshi Ghosh IIT Kharagpur Abstract Automatic Charge Identification (ACI) is the task of identifying the relevant legal charges given the facts of a situation and the statutory laws that define these charges, and is a crucial aspect of the judicial process. Prior works focus on learning charge-side representations by modeling relationships between the charges, but not much effort has been made in improving fact-side representations. We observe that only a small fraction of sentences in the facts actually indicates the charges. We show that by using a very small subset (< 3%) of fact descriptions annotated with sentence-level charges, we can achieve an improvement across a range of different ACI models, as compared to modeling just the main document-level task on a much larger dataset. Additionally, we propose a novel model that utilizes sentence-level charge labels as an auxiliary task, coupled with the main task of document-level charge identification in a multi-task learning framework. The proposed model comprehensively outperforms a large number of recent baseline models for ACI. The improvement in performance is particularly noticeable for the rare charges which are known to be especially challenging to identify. 1 Introduction In countries following a Civil Law system, there exist written statutory laws defining ‘charges’ (e.g., theft, murder, sexual offence), which one must consider before establishing a particular charge in a given situation. Charge identification is a tedious and time-consuming task requiring legal expertise, since the written laws are often described abstractly to encompass the wide-ranging scenarios in which a charge can apply. Thus automating the charge identification task can benefit the police and law practitioners. The task of Automatic Charge Identification (ACI, also referred to as ‘charge prediction’) aims to de- termine the possible charges that might have occurred in a given situation, using automated methods (Hu et al., 2018; Wang et al., 2018; Wang et al., 2019). A method for ACI takes as input (i) a document con- taining a textual description of a situation, called ‘fact-document’ or ‘facts’, and, (ii) textual descriptions of the charges/charges, available as written laws (statutes) in the legal statues of a country. The method outputs the set of charges/charges committed in the given situation. Most efforts in this task have been devoted towards distinguishing between rare and confusing charges, or learning better representations of the charge descriptions. For instance, Hu et al. (2018) designed 10 discriminative attributes to disambiguate confusing charge pairs, while Wang et al. (2018) and Wang et al. (2019) exploit pairwise co-occurrence and hierarchies of legal statutes respectively, to learn better representations for the statute descriptions. However, there has not been much effort in learning better fact-side representations. Fact descriptions are usually long and contain a lot of background information – e.g., the relationship between the parties involved, past …

Content truncated to 500 words