Incorporating Domain Knowledge for Extractive Summarization of Legal Case Documents

by   Paheli Bhattacharya, et al.

Automatic summarization of legal case documents is an important and practical challenge. Apart from many domain-independent text summarization algorithms that can be used for this purpose, several algorithms have been developed specifically for summarizing legal case documents. However, most of the existing algorithms do not systematically incorporate domain knowledge that specifies what information should ideally be present in a legal case document summary. To address this gap, we propose an unsupervised summarization algorithm DELSumm which is designed to systematically incorporate guidelines from legal experts into an optimization setup. We conduct detailed experiments over case documents from the Indian Supreme Court. The experiments show that our proposed unsupervised method outperforms several strong baselines in terms of ROUGE scores, including both general summarization algorithms and legal-specific ones. In fact, though our proposed algorithm is unsupervised, it outperforms several supervised summarization models that are trained over thousands of document-summary pairs.



There are no comments yet.


page 1

page 2

page 3

page 4


Automatic Text Summarization of Legal Cases: A Hybrid Approach

Manual Summarization of large bodies of text involves a lot of human eff...

Robust Deep Reinforcement Learning for Extractive Legal Summarization

Automatic summarization of legal texts is an important and still a chall...

LawSum: A weakly supervised approach for Indian Legal Document Summarization

Unlike the courts in western countries, public records of Indian judicia...

Long Document Summarization in a Low Resource Setting using Pretrained Language Models

Abstractive summarization is the task of compressing a long document int...

Legal Search in Case Law and Statute Law

In this work we describe a method to identify document pairwise relevanc...

Plain English Summarization of Contracts

Unilateral contracts, such as terms of service, play a substantial role ...

Investigating Entropy for Extractive Document Summarization

Automatic text summarization aims to cut down readers time and cognitive...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1. Introduction

In a Common Law system (followed in most countries including USA, UK, India, Australia), law practitioners have to go through hundreds of legal case documents to understand how the Court behaved in different legal situations. Case documents usually span several tens to hundreds of pages and contain complex text, which makes comprehending them difficult even for legal experts. Hence, summaries of case documents often prove beneficial. Existing legal IR systems employ legal attorneys or para-legals to write case summaries, which is an expensive process. Hence, automatic summarization of case documents is a practical and well-known challenge (Polsley et al., 2016; Zhong et al., 2019; Farzindar and Lapalme, 2004; Saravanan et al., 2006; Liu and Chen, 2019).

Document summarization is broadly of two types – (i) extractive, where important sentences are extracted from the source document and included in the summary  (Luhn, 1958; Erkan and Radev, 2004; Liu and Chen, 2019), and (ii) abstractive, where the model attempts to generate a summary using suitable vocabulary  (Liu and Lapata, 2019; Nallapati et al., 2017). This paper focuses on extractive summarization since it is more common for legal case documents.

Law practitioners in various countries have a set of guidelines for how to summarize court cases, e.g., see (how-to-brief-case-cuny, 2017, ; Giles,Jessica, 2015) for guidelines on summarizing case documents of the USA and the UK. All these guidelines basically advise to consider different rhetorical/thematic segments that are usually present in case documents (e.g., facts of the case, legal issues being discussed, final judgment, etc.) and then to include certain important parts of each segment into the summary. We got similar advice from law experts in India – two senior law students and a faculty from the Rajiv Gandhi School of Intellectual Property Law, India – regarding Indian court case documents (see details in Section 5.1). Apart from focusing on summarizing the case document as a whole, it is also necessary that each of these segments is summarized well and has representations in the final summary. Another advantage of including representations of each segment in the summary is that, often law practitioners wish to read only the summary of particular segments (e.g., only the facts of a case, or the legal issues involved in a case).

There are several summarization algorithms specifically meant for legal court case documents (Polsley et al., 2016; Zhong et al., 2019; Farzindar and Lapalme, 2004; Saravanan et al., 2006; Liu and Chen, 2019) (details in Section 2). However, most of these algorithms do not take into account the guidelines from law practitioners. Though a few legal domain-specific algorithms recognize the presence of various rhetorical segments in case documents (Farzindar and Lapalme, 2004; Saravanan et al., 2006), they also do not consider expert guidelines in deciding what to include in the summary from which segment. As a result, these algorithms cannot represent some of the important segments well in the summary. For instance, a recent work (Bhattacharya et al., 2019a) showed that most existing algorithms do not represent the final judgement well in the summary, though this segment is considered very important by law practitioners.

The above limitations of existing methods motivate us to develop a summarization algorithm that will consider various rhetorical segments in a case document, and then decide which parts from each segment to include in the summary, based on guidelines from law practitioners. Supervised methods which are expected to learn these guidelines automatically would require a large number of expert written summaries for lengthy case judgements, which may not be available for all jurisdictions. Hence we go for an unsupervised summarization algorithm in this work.

We propose DELSumm (Domain-adaptive Extractive Legal Summ

arizer ), an unsupervised extractive summarization algorithm for legal case documents. We formulate the task of summarizing legal case documents as maximizing an Integer Linear Programming (ILP) objective function that aims to maximize the inclusion of the most informative sentences in the summary, and has balanced representations from all the thematic segments, while reducing redundancy. We demonstrate how DELSumm can be tuned to operationalize summarization guidelines obtained from law experts over Indian Supreme Court case documents. Comparison with several baseline methods – including five legal domain-specific methods and two state-of-the-art deep learning models trained over thousands of document-summary pairs – suggest that our proposed approach outperforms most existing methods, especially in summarizing the different rhetorical segments present in a case document. To summarize, the contributions of this paper are:

(1) We propose DELSumm that incorporates domain knowledge provided by legal experts for summarizing lengthy case documents.111Implementation publicly available at

(2) We perform extensive experiments on documents from the Supreme Court of India and compare our proposed method with eleven baseline approaches. We find that DELSumm outperforms a large number of legal-specific as well as general summarisation methods, including supervised neural models trained over thousands of document-summary pairs. Especially, DELSumm achieves much better summarization of the individual rhetorical segments in a case document, compared to most prior methods.

(3) We also show that our proposed approach is robust to inaccurate rhetorical labels generated algorithmically. There is a negligible drop in performance when DELSumm uses rhetorical sentence labels generated by a rhetorical segmentation method (which is more a practical setup), instead of using expert-annotated labels.

To the best of our knowledge, this is the first systematic attempt to computationally model and incorporate legal domain knowledge for summarization of legal case documents. Through comparison with 11 methods, including 2 state-of-the-art deep learning methods, we show that an unsupervised algorithm developed by intelligently including domain expertise, can surpass the performance of supervised learning models even when the latter are trained over large training data (which is anyway expensive to obtain in an expert-driven domain such as Law).

2. Related Work

Extractive text summarization aims to detect important

sentences from the full document and include them in the summary. Existing methods for extractive summarization that can be applied for legal document summarization, can be broadly classified into four classes: (i) Unsupervised domain-independent, (ii) Unsupervised domain-specific, (iii) Supervised domain-independent, and (iv) Supervised domain-specific. We briefly describe some popular methods from each of the classes in this section.

2.1. Unsupervised domain-independent methods

Popular, unsupervised extractive summarization algorithms identify important sentences either by using Frequency-based methods (e.g. Luhn (Luhn, 1958)) or Graph-based methods (e.g. LexRank (Erkan and Radev, 2004)). The summary is the top ranked sentences. There are algorithms from the family of matrix-factorization, such as LSA (Gong and Liu, 2001).

Summarization has also been treated as an Integer Linear Programming (ILP) optimization problem. Such methods have been applied for summarizing news documents (Banerjee et al., 2015) and social media posts (Rudra et al., 2018). Our proposed approach in this work is also based on ILP-based optimization.

2.2. Unsupervised domain-specific methods

There are several unsupervised methods for extractive summarization specifically designed for legal case documents. One of the earliest methods for unsupervised legal document summarization that takes into account the rhetorical structure of a case document was LetSum (Farzindar and Lapalme, 2004)

. They consider certain cue-phrases to assign rhetorical/semantic roles to a sentence in the source document. They specifically consider the roles – Introduction, Context, Juridical Analysis and Conclusion. Sentences are ranked based on their TF-IDF values for estimating their importance. The final summary is generated by taking 10% from the Introduction, 25% from Context, 60% from Juridical Analysis and 5% from the Conclusion segments.

While LetSum considers TF-IDF to rank sentences, Saravanan (Saravanan et al., 2006) use a K-Mixture Model to rank sentences for deciding which sentences to include in the summary. We refer to this work as KMM in the rest of the paper. Note that this work also identifies rhetorical roles of sentences (using a graphical model). However, this is a post-summarization step that is mainly used for displaying the summary in a structured way; the rhetorical roles are not used for generating the summary.

Another method CaseSummarizer (Polsley et al., 2016) finds the importance of a sentence based on several factors including its TF-IDF value, the number of dates present in the sentence, number of named entities and whether the sentence is at the start of a section. Sentences are then ranked in order to generate the summary comprising of the top-ranked sentences.

Zhong (Zhong et al., 2019) creates a template based summary for Board of Veteran Appeals (BVA) decisions from the U.S. Department of Veteran Affairs. The summary contains (i) one sentence from the procedural history (ii) one sentence from issue (iii) one sentence from the service history of the veteran (iv) variable number of Reasoning & Evidential Support sentences selected using Maximum Margin Relevance (v) one sentence from the conclusion. We refer to this method as MMR in this paper.

Limitations of the methods: CaseSummarizer does not assume the presence of any rhetorical role. The other methods – LetSum, KMM, and MMR – consider their presence but do not include them while estimating the importance of sentences for generating the summary. Specifically, LetSum and KMM use generic term-distributional models (TF-IDF and K-mixture models respectively) and MMR uses Maximum Margin Relevance. As evident, rhetorical roles do not play any role in finally selecting the sentences for inclusion in the summary.

We believe that it is more plausible to measure the importance of sentences in each rhetorical segment separately using domain knowledge. In fact, for each segment, the parameters measuring sentence importance can vary. As an example, consider the segment ‘Statute’. One can say that sentences from this segment that actually contain a reference to a Statute/Act name are important. In contrast, this reasoning will not hold for the segment ‘Facts’ and one may derive a different measure based on the presence of certain Part-of-Speech tags e.g., Nouns (names of people, location, etc.). Hence, in the present work, we especially include such domain knowledge along with segment representations in a more systematic way for generating summaries.

2.3. Supervised domain-independent methods

Supervised neural (Deep Learning-based) methods for extractive text summarization treat the task as a binary classification problem, where sentence representations are learnt using a hierarchical encoder. A survey of the existing approaches can be found in (Dong, 2018).

Two popular extractive methods are NeuralSum (Cheng and Lapata, 2016) and SummaRunner (Nallapati et al., 2017)

. These methods use RNN (Recurrent Neural Network) encoders to learn the sentence representations from scratch. Sentence selection into summary is based on – content, salience, novelty and absolute and relative position importance. These parameters are also learned in the end-to-end models.

Recently, pretrained encoders (especially transformer-based models such as BERT (Devlin et al., 2019)) have gained much popularity. These encoders have already been pretrained using a large amount of open domain data. Given a sentence, they can directly output its sentence representation. These models can be used in a supervised summarization by fine-tuning the last few layers on domain-specific data. BERTSUM (Liu and Lapata, 2019) is a BERT-based extractive summarization method. Unlike SummaRuNNer, here sentence selection into summary is based on trigram overlap with the currently generated summary.

Limitations of the neural methods: These Deep Learning architectures have been evaluated mostly in the news summarization domain, and news documents are much shorter and contain simpler language than legal documents. Recent works (Sotudeh et al., 2021; Xiao and Carenini, 2019) show that these methods do not perform well in summarizing scientific articles from PubMed and arxiv, which are longer sequences. However, it has not been explored how well these neural models would work in summarizing legal case documents; we explore this question for two popular neural summarization models in this paper.

2.4. Supervised domain-specific methods

Liu (Liu and Chen, 2019)

have recently developed a supervised learning method for extractive summarization of legal documents. Sentences are represented using handcrafted features like number of words in a sentence, position of the sentence in a document etc. Machine Learning (ML) classifiers (e.g., Decision Tree, MLP, LSTM) are then applied for obtaining the final summary. We refer to this method as

Gist (Liu and Chen, 2019).

Limitations of the method: Although this method was developed and applied for Chinese legal case documents, domain-specific attributes such as rhetorical labels were not considered.

2.5. Rhetorical roles in a legal case document

A legal case document can be structured into thematic segments, where each sentence can be labelled with a rhetorical role. Note that case documents often do not implicitly specify these rhetorical roles; there exist algorithms that assign rhetorical roles to the sentences (Bhattacharya et al., 2019b; Yamada et al., 2019; Saravanan et al., 2008). Different prior works have considered different sets of rhetorical roles (Farzindar and Lapalme, 2004; Saravanan et al., 2006; Zhong et al., 2019; Bhattacharya et al., 2019b). It was shown in our prior work (Bhattacharya et al., 2019a) that a mapping between the different sets of rhetorical roles is possible.

In this work, we consider a set of eight rhetorical roles suggested in our prior work (Bhattacharya et al., 2019b). Briefly, the rhetorical roles are as follows – (i) Facts: the events that led to filing the case, (ii) Issue: legal questions/points being discussed, (iii) Ruling by Lower Court: case documents from higher courts (e.g., Supreme Court) can contain decisions delivered at the lower courts (e.g. Tribunal), (iv) Precedent: citations to relevant prior cases, (v) Statute: citations to statutory laws that are applicable to the case (e.g., Dowry Prohibition Act, Indian Penal Code), (vi) Arguments delivered by the contending parties, (vii) Ratio: rationale based on which the judgment is given, and (viii) Final judgement of the present court.

In this work, we attempt to generate a summary that includes representations of all rhetorical segments in the source document (full text). We assume that the rhetorical labels of every sentence in the source document is already labeled either by legal experts or by applying the method proposed in (Bhattacharya et al., 2019b).

3. Dataset

For evaluation of summarization algorithms, we need a set of source documents (full text) and their gold standard summaries. Additionally, since we propose to use the rhetorical labels of sentences in a source document for generating the summary, we need rhetorical label annotations of the source documents. Additionally, since the supervised methods Gist, SumaRuNNer and BERTSUM require training over large number of document-summary pairs, we also need such a training set. In this section, we describe the training set and the evaluation set (that is actually used for performance evaluation of summarization algorithms).

Evaluation set: Our evaluation dataset consists of a set of Indian Supreme Court case documents, where each sentence is tagged with a rhetorical/semantic label by law experts (out of the rhetorical roles described in Section 2.5). This dataset is made available by our prior work (Bhattacharya et al., 2019b). We asked two senior law students (from the Rajiv Gandhi School of Intellectual Property Law, one of the most reputed law schools in India) to write summaries for each of these document. They preferred to write extractive summaries of approximately one-third of the length of the documents. We asked the experts to summarize each rhetorical segment separately (so that we can evaluate how well various models summarize the individual sections); only, they preferred to summarize the rhetorical segments ‘Ratio’ and ‘Precedent’ together.

All the summarization methods have been used to generate summaries for these documents. These summaries were then uniformly evaluated against the two gold standard summaries written by the law students, using the standard ROUGE scores (details given in later sections). We report the average ROUGE scores over the two sets of gold standard summaries.

Training set : For training supervised methods (Gist, SumaRuNNer and BERTSUM), we need an additional training dataset consisting of document-summary pairs. To this end, we crawled Indian Supreme Court case documents and their headnotes (short abstractive summaries) from which is an archive of Indian court cases. These document-summary pairs were used to train the supervised summarization models (details in later sections). We ensured that there was no overlap between this training set and the evaluation set of documents.

Note that the headnotes described above are not considered to be summaries of sufficiently good quality by our law experts, and this is why the experts preferred to write their own summaries as gold standard for the documents in the evaluation set. It can be argued that it is unfair to train the supervised summarization models using summaries that are known to be of poorer quality than the target summaries. However, the supervised summarization models require thousands of document-summary pairs for training, and it is practically impossible to obtain so many summaries of the target quality. Hence our only option is to use the available headnotes for training the supervised summarization models. This situation can be thought of as a trade-off between quantity and quality of training data – if a method requires large amounts of training data, then that data may be of poorer quality than the target quality.

4. Proposed Approach: DELSumm

In this section, we describe our proposed algorithm DELSumm . Our algorithm uses an optimization framework to incorporate legal domain knowledge (similar to what is stated in (how-to-brief-case-cuny, 2017, ; Giles,Jessica, 2015)) into an objective function with constraints. The objective function is then maximized using Integer Linear Programming (ILP). The symbols used to explain the algorithm are stated in Table 1.

Notation Meaning
Desired summary length (number of words)
Number of sentences in the document
Number of segments in the document
Number of distinct content words in the document
Index for sentence ()
Index for content word ()
Index for segment ()
Indicator variable for sentence (1 if sentence is to be included in summary, 0 otherwise)
Indicator variable for content word (1 if is to be included in summary, 0 otherwise)
Indicator variable for sentence citing a prior-case (1 if there is a citation, 0 otherwise)
Indicator variable for sentence citing a statute (1 if there is a citation, 0 otherwise)
Number of words in sentence
Informativeness of sentence
Set of content words in sentence
Position of sentence in the document
Score (a measure of importance) of content word
Set of sentences where content word is present
Weight (a measure of importance) of segment
Set of sentences belonging to segment
Minimum number of sentences to be selected from segment in the summary
Table 1. Notations used in the DELSumm algorithm

We consider that a case document has a set of rhetorical segments (e.g., the rhetorical segments stated in Section 2.5), and the summary is supposed to contain a representation of each segment (as indicated in (how-to-brief-case-cuny, 2017, ; Giles,Jessica, 2015)). The algorithm takes as input – (i) a case document where each sentence has a label signifying its rhetorical segment, and (ii) the desired number of words in the summary. The algorithm then outputs a summary of at most words containing a representation of each segment.

The optimization framework: We formulate the summarization problem using an optimization framework. The ILP formulation maximizes the following factors:
(i) Informativeness of a sentence: The informativeness of a sentence defines the importance of a sentence in terms of its information content. More informative sentences are likely to be included in the summary. We assume that domain experts will define how to estimate the informativeness of a sentence.
(ii) Content words: Content words signify domain-specific vocabulary (e.g., terms from a legal dictionary, or names of statutes, etc) and noun-phrases. The importance of each content word is given by . We assume that domain experts will define what words/terms should be considered as ‘content words’, and how important the content words are.

A summary of length words, consisting of the most informative sentences and content words, is achieved by maximizing the following objective function:


subject to constraints


The objective function in Eqn. 1 tries to maximize the inclusion of informative sentences (through the indicator variables) and the number of important content words (through the indicator variables). Here (respectively, ) is set to if the algorithm decides that sentence (respectively, content word ) should be included in the summary. Eqn. 2 constraints that the summary length is at most words (note from Table 1 that is the number of words in sentence ). Eqn. 3 implies that if a particular sentence is selected for inclusion in the summary (i.e., if ), then the all the content words contained in that sentence () are also selected. Eqn. 4 suggests that if a content word is selected for inclusion in the summary (i.e., if ), then at least one sentence where that content word in present is also selected.

Eqn. 5 ensures from each segment a minimum number of sentences should be selected in the summary. This is a key step that ensures representation of all rhetorical segments in the generated summary. We assume that suitable values of will be obtained from domain knowledge.

Any ILP solver can be used to solve the ILP; we specifically used the GUROBI optimizer ( Finally, those sentences for which is set to will be included in the summary generated by DELSumm .

Handling Redundancy: DELSumm implicitly handles redundancy in the summary through the variables that indicate inclusion of content words (refer to Eqn. 3). According to Eqn. 3, if a sentence is included in the summary, then the content words in that sentence are also selected. Consider that two sentences and are similar, so that including both in the summary would be redundant. The two sentences are expected to have the same content words. Hence, adding both and (which have the same content words) will not help in maximizing the objective function. Hence the ILP method is expected to refrain from adding both sentences, thus prohibiting redundancy in the summary.

Applying DELSumm to case documents of a specific jurisdiction: It can be noted that this section has given only a general description of DELSumm , where it has been assumed that many details will be derived from domain knowledge (e.g., the informativeness or importance of a sentence or content word, or values of ). In the next section, we specify how these details are derived from the guidelines given by domain experts from India.

5. Applying DELSumm on Indian case documents

In this section, we describe how DELSumm was adopted to summarize Indian case documents. We first describe the summarization guidelines stated by law experts from India, and then discuss how to adopt those guidelines into DELSumm . We then discuss comparative results of the proposed algorithm and the baselines on the India dataset.

5.1. Guidelines from Law experts

We consulted law experts – senior law students and faculty members from the Rajiv Gandhi School of Intellectual Property Law (a reputed law school in India) – to know how Indian case documents should be summarized. Based on the rhetorical segments described in Section 2.5, the law experts suggested the following guidelines.
(G1) In general, the summary should contain representations from all segments of a case document, except the segment ‘Ruling by Lower Court’ which may be omitted. The relative importance of segments in a summary should be: Final judgment Issue Fact (Statute, Precedent, Ratio) Argument.
(G2) The segments ‘Final judgement’ and ‘Issue’ are exceptionally important for the summary. The segments are usually very short in the documents, and so they can be included completely in the summary.
(G3) The important sentences in various rhetorical segments (Fact, Statute, Precedent, Ratio) should be decided as follows – (a) Fact: sentences that appear at the beginning; (b) Statute : sentences that contain citations to an Act; (c) Precedent : sentences that contain citation to a prior-cases; (d) Ratio : sentences appearing at the end of the document and sentences that contain citation to an act/law/prior-case.
(G4) The summary must contain sentences that give important details of the case, including the Acts and sections of Acts that were referred, names of people, places etc. that concern the case, and so on. Also sentences containing specific legal keywords are usually important.

5.2. Operationalizing the guidelines

Guideline Operationalization in DELSumm
(G1) Segment weights Highest assigned to Final judgement and Issue; weight decreasing exponentially / linearly for other segments:
(G2) Segment
(G3) Sentence
(G4) Content Words &
their weights
Table 2. Parameter settings in DELSumm for operationalizing the guidelines provided by law experts for Indian case documents (stated in Section 5.1). Parameter settings such as weights of various segments, how to compute importance of sentences in various segments, etc. are decided based on the segment . The symbols are explained in Table 1.

Table 2 shows how the above-mentioned guidelines from law experts in India have been operationalized in DELSumm .

Operationalizing G1: We experiment with two ways of assigning weights to segments – linearly decreasing and exponentially decreasing – based on guideline G1 stated above. We finally decided to go with the exponentially decreasing weights. Specifically, we assign weight to the ‘Final Judgement’ segment (that is judged to be most important by experts), followed by for ‘Issue’, for ‘Fact’, for ‘Statute’, ‘Ratio & ‘Precedent’ and finally for ‘Argument’.

Operationalizing G2: As per guideline G2, (the minimum number of sentences from a segment ) is set to ensure a minimum representation of every segment in the summary – the ‘final judgement’ and ‘issue’ rhetorical segments are to be included fully, and at least sentences from every other segment are to be included.

Operationalizing G3: The informativeness of a sentence defines the importance of a sentence in terms of its information content. More informative sentences are likely to be included in the summary. To find the informativeness of a sentence , we use the guideline G3 stated above. therefore depends on the rhetorical segment (Fact, Issue, etc.) that contains a particular sentence . For instance, G3(a) dictates that sentences appearing at the beginning of the rhetorical segment ‘Fact’ are important. We incorporate this guideline by weighing the sentences within ‘Fact’ by the inverse of its position in the document.

According to G3(b) and G3(c), sentences that contains mentions of Statutes/Act names (e.g., Section 302 of the Indian Penal Code, Article 15 of the Constitution, Dowry Prohibition Act 1961, etc.) and prior-cases/precedents are important. We incorporate this guideline through a Boolean/indicator variable which is if the sentence contains a mention of a Statute. We use regular expression patterns to detect Statute/Act name mentions. Similarly for detecting if a sentence contains a reference to a prior case, we use the Boolean variable . If a regular expression pattern of the form Party 1 vs. Party 2 is found in a sentence , is set to .

For understanding important Ratio sentences, we lookup guideline G3(d), which mandates that sentences containing references to either a Statute/Act name or a prior-case are important to be included in the summary. We therefore use the two Boolean/indicator variables and for detecting the presence of a Statute/Act name or a prior-case. If any one of them occurs in the sentence, we consider that sentence from segment Ratio to be informative.

Operationalizing G4: Based on guideline G4 stated above, we need to identify some important ‘content words’ whose presence would indicate the sentences that contain especially important details of the case. To this end, we consider three types of content words – mentions of Acts/sections of Acts, keywords from a legal dictionary, and noun phrases (since names of people, places, etc. are important). For identifying legal keywords, we use a legal dictionary from the website For identifying mentions of Acts and statutes, we use a comprehensive list of Acts in the Indian judiciary, obtained from Westlaw India ( As content word scores (), we assign a weight of to statute mentions, to legal phrases and to noun phrases; these scores are based on suggestions by the legal experts on the relative importance of sentences containing different types of content words.

6. Baselines and Experimental Setup

This section describes the baseline summarization methods that we consider for comparison with the proposed method. Also the experimental setup used to compare all methods is discussed.

6.1. Baseline summarization methods

As described in Section 2, there are four classes of methods that can be applied for extractive summarization of legal case documents. We consider some representative methods from each class as baselines.

Unsupervised domain-independent methods: We consider as baselines the following popular algorithms from this class: (1) Luhn (Luhn, 1958), (2) LexRank (Erkan and Radev, 2004), (3) Reduction (Jing, 2000), and (4) LSA (Gong and Liu, 2001) (see Section 2.1 for brief descriptions of these methods).

Unsupervised domain-specific methods: Section 2.2 gave brief descriptions of these methods. Among such methods, we consider the following four as baselines – (1) LetSum (Farzindar and Lapalme, 2004), (2) KMM (Saravanan et al., 2006), (3)  CaseSummarizer (Polsley et al., 2016) and (4) a simplified version of MMR (Zhong et al., 2019) – in the absence of identical datasets, we adopt only the Maximum Margin Relevance module and use it to summarize a document.

Supervised domain-specific methods: From this class of algorithms, we consider Gist (Liu and Chen, 2019)

, a recent method that applies general ML algorithms to the task. The best performance was observed by using Gradient Boosted Decision Tree as the ML classifier, which we report (see Section 

2.4 for details of the method).

Supervised domain-independent methods: Among these methods, we consider the neural method SummaRuNNer (Nallapati et al., 2017) (implementation available at

). Similar to Gist, they consider the task of extractive summarization as binary classification problem. The classifier returns a ranked list of sentences based on their prediction/confidence probabilities about their inclusion in the summary. We include sentences from this list in decreasing order of their predicted probabilities, until the desired summary length is reached.

We also apply a recent BERT-based summarization method BERTSUM (Liu and Lapata, 2019) (implementation available at In BERTSUM, the sentence selection function is a simple binary classification task (whether or not to include a sentence in the summary).222The original BERTSUM model uses a post-processing step called Trigram Blocking that excludes a candidate sentence if it has a significant amount of trigram overlap with the already generated summary (to minimize redundancy in the summary). However, we observed that this step leads to summaries that are too short, as also observed in (Sotudeh et al., 2021). Hence we ignore this step. Similar to SummaRuNNer, we use the ranked list of sentences based on the confidence probabilities of their inclusion in the summary. We include sentences one-by-one into the final summary, until the desired length is reached.

Implementations of the baselines: For Luhn, LexRank, Reduction and LSA, we use the implementations from the Python sumy module. The implementations of SummaRuNNEr and BERTSUM are also available publicly, at the URLs stated above. We implemented the legal domain-specific summarization methods – CaseSummarizer, LetSum, GraphicalModel, MMR and Gist – following the descriptions in the corresponding papers as closely as possible (the same parameter values were used as specified in the papers).

Converting abstractive summaries to extractive for training supervised summarization models: The supervised methods Gist, SumaRuNNer and BERTSUM require labelled data for training, where every sentence in the source document must be labeled as if this sentence is suitable for inclusion in the summary, and labeled as otherwise. The publicly available headnotes in our training datasets (see Section 3) are abstractive in nature and cannot be directly used to train such methods. Hence we convert the abstractive summaries to extractive for training these methods.

To this end, we adopt the technique mentioned in (Nallapati et al., 2017). Briefly, we assign label (suitable for inclusion in the summary) to those sentences from the source document (full text), that greedily maximizes the ROUGE 2-overlap with the abstractive summary. The rest of the sentences in the source document are assigned label .

6.2. Experimental setup

Now we discuss the setup for our experiments to compare the performances of various summarization models.

Target length of summaries: Summarization models need to be told the target length (in number of words) of the summaries that they need to generated. For every document in the India dataset, we have two gold standard summaries written by two domain experts. For a particular document, we consider the target summary length to be the average of the number of words in the summaries (for that document) written by the two experts.

For each document, every summarization model is made to generate a summary of length at most words, which ensures a fair comparison among the different models.

Evaluation of summary quality: We use the standard ROUGE metrics for evaluation of summary quality (computed using v. 1.0.0). For a particular document, we have gold standard summaries written by two legal experts (as was described in Section 3

). We compute the ROUGE scores of the algorithmic summary and each of the gold standard summary, and take the average of the two. We remove a common set of English stopwords from both the gold standard summaries and the algorithm-generated summaries before measuring ROUGE scores. We report ROUGE-2 Recall and F-score, and ROUGE-L Recall and F-scores. The scores are averaged over all

documents in the evaluation set.

7. Results and Analyses

This section compares the performances of the proposed DELSumm and all the baseline algorithms. We compare the performances broadly in two ways – (i) how good the summaries for the full documents are, and (ii) how well each individual rhetorical segment is summarized. Table 3 and Table 4 show the performances of various methods in summarizing the Indian Supreme Court case documents. In both tables, ROUGE scores are averaged over the documents in the India evaluation set.

7.1. Evaluation of the full summaries

Algorithm ROUGE 2 ROUGE-L
Unsupervised Domain Independent
(i) LexRank 0.3662 0.3719 0.5068 0.5392
(ii) LSA 0.3644 0.3646 0.5632 0.5483
(iii) Luhn 0.3907 0.3873 0.5424 0.5521
(iv) Reduction 0.3778 0.3814 0.5194 0.542
Unsupervised Domain Specific
(v) LetSum 0.403 0.4137 0.5898 0.5846*
(vi) KMM 0.354 0.3546 0.5407 0.5385
(vii) CaseSummarizer 0.3699 0.3765 0.4925 0.5349
(viii) MMR 0.3733 0.3729 0.6064 0.568
DELSumm (proposed) 0.4323 0.4217 0.6831 0.6017
Supervised Domain Independent
(ix) SummaRuNNer 0.4104* 0.4149* 0.5835 0.5821
(x) BERTSUM 0.4044 0.4048 0.56 0.5529
Supervised Domain Specific
(xi) Gist 0.3567 0.3593 0.5712 0.5501
Table 3. Performance of the summarization methods as per ROUGE metrics (R = recall, F = Fscore). All values are averaged over the documents in the evaluation set. The best value for each measure is in bold and the second best in underlined. Entries with asterisk(*) indicates DELSumm is not

statistically significant than baselines using 95% confidence interval by Student’s T-Test.

Table 3 shows the performances of summarizing the full documents in terms of ROUGE-2 and ROUGE-L Recall and F-scores. Considering all the approaches, DELSumm achieves the best performance across all the measures. The second-best metric values are achieved by LetSum (For ROUGE-L F-score), MMR (for ROUGE-L Recall), and SummaRuNNer (for ROUGE-2 Recall and F-score). To check if their performances are statistically significantly different from those achieved by DELSumm , we perform Student’s T-Test at 95% () with these three methods. Entries that are not statistically significant are marked with asterisk (*) in Table 3.

It can be noted that, out of all the domain-specific methods in Table 3 (both supervised and unsupervised), DELSumm performs the best for all the measures. While LetSum is the only prior work that considers rhetorical segments while generating the summary (see Section 2), none of these methods take into account domain-specific guidelines on what to include in the summary. DELSumm on the other hand models the legal knowledge about what to include in a summary, thus being able to perform better.

Importantly, DELSumm outperforms even supervised neural models such as SummaRuNNer and BERTSUM (though the difference in the performance of SummaRuNNer with respect to DELSumm is not statistically significant in terms of ROUGE-2 scores). Note that DELSumm is a completely unsupervised method while SummaRuNNer and BERTSUM are deep-learning based supervised methods trained over document-summary pairs. The fact that DELSumm still outperforms these supervised models is because DELSumm intelligently utilizes domain knowledge while generating the summaries.

7.2. Evaluation of segment-wise summarization

Unsupervised, Domain Independent
LexRank 0.0619 0.3469 0.4550 0.2661 0.3658 0.4284
LSA 0.0275 0.2529 0.5217 0.2268 0.3527 0.3705
Luhn 0.0358 0.2754 0.5408 0.2662 0.2927 0.3781
Reduction 0.0352 0.3153 0.5064 0.2579 0.3059 0.4390
Unsupervised, Domain Specific
LetSum 0.0423 0.3926 0.6246 0.3469 0.3853 0.2830
KMM 0.3254 0.2979 0.4124 0.3415 0.4450 0.416
CaseSummarizer 0.2474 0.3537 0.4500 0.2255 0.4461 0.4184
MMR 0.4378 0.3548 0.4442 0.2763 0.4647 0.3705
DELSumm 0.7929 0.6635 0.5539 0.4030 0.4305 0.4370
Supervised, Domain Independent
SummaRuNNer 0.4451 0.2990 0.5231 0.1636 0.5215 0.3090
BERTSUM 0.0662 0.3544 0.6376 0.2535 0.3121 0.3262
Supervised, Domain Specific
Gist 0.5844 0.3856 0.4621 0.2759 0.4537 0.2132
Table 4. Segment-wise performance (ROUGE-L F-scores) of the methods. All values are averaged over the documents in the evaluation set. Values in the column headings are the % of sentences in the full document that belong to a segment. Values highlighted in red-underlined. The best value for each segment in green-bold.

Overall ROUGE scores are not the best metrics for evaluating case document summaries. Law experts opine that, even methods that achieve high overall ROUGE scores may not represent every segment well in the summary (Bhattacharya et al., 2019a). A segment-wise performance evaluation is practically important since law practitioners often intend to read the summary of a particular segment, and not always the full summary. Hence we perform a segment-wise performance evaluation of the algorithms.

To this end, we proceed as follows. For each rhetorical segment (e.g., Facts, Issues), we extract the portions of an algorithmic summary and the gold standard summary that represent the given segment, using the gold standard rhetorical labels of individual sentences (which are present in our evaluation set, as detailed in Section 3). Then we compute the ROUGE scores on those specific text portions only. For instance, to compute the ROUGE score on the ‘Fact’ segment of a particular document, we only consider those sentences in the gold standard summary and the algorithmic summary, which have the rhetorical label ‘Fact’ in the said document. We report the average ROUGE score for a particular segment, averaged over all documents in the evaluation set.

Table 4 shows the segment-wise ROUGE-L F-scores (averaged over all documents in the evaluation set). The values are underlined and highlighted in red color, while the best value for each segment is highlighted in boldface and green color.

Similar to (Bhattacharya et al., 2019a), we also observe that many of the baseline methods could not represent the ‘Final judgement’ and ‘Issue’ well in their summaries. This flaw is especially critical since these two segments are the most important in the summary according to the law experts (see Section 5.1). In contrast, DELSumm achieves a very high performance on these two important segments. This difference in performance is possibly because these two segments are also the shortest segments (constitutes only 2.9% and 1.2% of the whole document, as stated in the first row of Table 4), and hence are missed by other methods which do not know of their domain-specific importance. This observation show the necessity of an informed algorithm that can incorporate domain knowledge from experts.

DELSumm also represents the ‘Statute’ segment better than all other methods. For the ‘Statute’ segment, the algorithm was formulated in such a way (through the variable) that it is able to incorporate sentences that contain mention of an Act/law. Other methods did not perform well in this aspect. The performance of DELSumm for the ‘Argument’ segment is second-best () after that of Reduction (). These values are very close, and the difference is not statistically significant.

Our method is unable to perform as well for the ‘Precedent Ratio’ and ‘Facts’ segments as some other methods. Note that, these segments accounts for maximum number of sentences in a document and also forms a large part of the summary (see Table 4, first row). Hence neural methods (e.g., BERTSUM) and methods relying of TF-IDF measures (e.g., LetSum) obtained relatively large amounts of training data for these segments, and hence performed well for these segments.

Finally, although LetSum is a segment-aware algorithm, its summarization mechanism is not strong enough to understand the importance of sentences in all segments; for instance, it performs very poorly for the ‘Final judgement’ segment. DL methods fail to perform well for the smaller segments (e.g., Final judgement, Issue) for which lesser amounts of training data is available.

Overall, it is to note that DELSumm achieves a much more balanced representation of segments in the summary, compared to all the baseline methods. Contrary to the baseline methods, DELSumm shows decent performance for all the segments. This has been taken care mainly through the constraint in Eqn. 5. Hence, even when the most important segments (e.g., Final judgement) are optimized in the summary, the less important ones (e.g., Arguments) are not missed out.

7.3. Discussion on the comparative results

The results stated above suggest that our proposed method performs better than many existing domain-independent as well as legal domain-specific algorithms. The primary reason for this superior performance of DELSumm is that it is able to incorporate legal domain knowledge much more efficiently through a theoretically grounded approach. By doing this, it is able to surpass the performance of deep learning and machine learning approaches such as SummaRuNNer, BERTSUM and Gist (that are trained over document-summary pairs). These results show that if legal domain knowledge can be used intelligently, even simple unsupervised summarization algorithms can compete with state-of-the-art supervised learning approaches.

It can be argued that document-summary pairs are too less to train a deep learning summarization model; additionally, as stated in Section 3, the summaries (headnotes) in the training set may not be of the same quality as those in the evaluation set. However, in domains such as legal, it is difficult and expensive to obtain rich quality training data in huge amounts. In such scenarios, DELSumm , being an unsupervised method, can work efficiently by incorporating domain knowledge.

Another advantage of DELSumm is that it is easily customizable – if the requirement is that a particular segment should have more representation in the summary than the others, the segment weights and representations can be tuned accordingly. If the sentence importance measures are different, that too can be adjusted. Other domain-specific and supervised learning methods do not exhibit such flexibility.

8. Using DELSumm with Algorithmic Rhetorical Labels

DELSumm requires sentences of the source document (full text) to be labelled with a rhetorical role prior to summarization. To this end, till now we have been using expert-annotated labels. However, in a real scenario, it is not possible to get expert-annotations for rhetorical role labelling for every document. A more practical approach is to first computationally generate these rhetorical labels through some algorithm that detects rhetorical roles of sentences in a legal document, and then consider the labelled document as input to DELSumm for summarization.

Note that algorithmically derived rhetorical labels are likely to be noisy (i.e., not 100% accurate). In this section, we analyse how the performance of DELSumm is affected by such noisy algorithmic labels when compared to gold standard labels. To this end, we perform the following experiment (which is inspired by the idea of -fold cross validation that is popular in Machine Learning literature).

Experimental Design: Assume that we have a dataset consisting of documents, each having sentence-wise rhetorical label annotations as well as gold-standard summaries (e.g., the Indian dataset we are using in this paper, with ). We divide the dataset into folds (e.g., ). In each fold , there are documents for evaluation, which we call as the evaluation set . The remaining documents are considered for training a model for rhetorical role labeling, termed as training set . We also have the expert annotated labels for the documents in fold , and their gold standard summaries. For each fold, we use for training the state-of-the-art rhetorical role identification model Hier-BiLSTM-CRF from our prior work (Bhattacharya et al., 2019b).333Implementation available at

Using the trained model for that fold, we infer the rhetorical labels of the documents in the evaluation set for that fold. Now, we have two sets of rhetorical labels for the sentences in the documents in – (i) gold standard labels given by law experts, and (ii) algorithmic labels assigned by the model (Bhattacharya et al., 2019b). We next summarize every document in with DELSumm twice, once using the gold standard sentence labels, and again using the algorithmic sentence labels. We perform three evaluations:

(i) Accuracy of rhetorical labelling: we check what fraction of the algorithmic labels match with the gold standard labels (over the documents in ). This measure gives an idea of the accuracy of the rhetorical labeling model (Bhattacharya et al., 2019b).

(ii) ROUGE scores with Gold Std. Labels: We compute the ROUGE scores between the gold standard summaries and the summaries generated by DELSumm considering the gold standard labels. We average the ROUGE scores over all documents in the set (this is very similar to what we did in Section 5).

(iii) ROUGE scores with Algorithmic Labels: We compute the ROUGE scores between the gold standard summaries and summaries generated by DELSumm with the algorithmic labels, i.e., labels inferred by the model trained over . Again we average the ROUGE scores over all documents in the set .

The differences between these two sets of ROUGE scores can be used to quantify the degradation in performance of DELSumm while using noisy algorithmic labels instead of gold standard labels.

Figure 1. Effect on the performance of DELSumm with algorithmic labels. There is minimal degradation of ROUGE scores when using algorithmic labels

Results: We perform the above experiment with the India dataset where , and we consider folds. Hence in each fold, we use a training set of documents and an evaluation set of documents. Figure 1 shows the results of the above experiment. The horizontal axis shows the rhetorical labeling accuracy across the folds; we see that the model (Bhattacharya et al., 2019b) achieves accuracies in the range , i.e., about 82%–85% of the algorithmic labels are correct. The figure shows the two types of ROUGE-L F-scores as described above – one set using the algorithmic labels (denoted by red diamonds), and the other set using the gold-standard labels (denoted by black circles) – for each of the folds. Each value is averaged over the documents in the evaluation set of the corresponding fold. We find that the two sets of ROUGE scores are very close. In other words, even when noisy/inaccurate labels are used in DELSumm , the drop in performance is very low, indicating the robustness of DELSumm .

Algorithm ROUGE-2 ROUGE-L
LetSum 0.4030 0.4137 0.5898 0.5846
MMR 0.3733 0.3729 0.6064 0.5680
SummaRuNNer 0.4104 0.4149 0.5835 0.5821
DELSumm (GSL) 0.4323 0.4217 0.6831 0.6017
DELSumm (AL) 0.4193 0.4075 0.6645 0.5897
Table 5. Performance of DELSumm with Gold Standard labels (GSL) and Algorithmic labels (AL) over the documents. Apart from the last row, all other rows are repeated from Table 3. While DELSumm (AL) shows slightly degraded performance than DELSumm (GSL), it still outperforms the other methods according to all ROUGE measures except ROUGE-2 F-score.

Finally, we compute the ROUGE scores averaged over all documents across all folds. Table 5 shows the average ROUGE scores achieved by DELSumm using the algorithmic labels (denoted ‘DELSumm AL’). For ease of comparison, we repeat some rows from Table 3 – the performance of DELSumm with gold standard rhetorical labels (denoted ‘DELSumm GSL’), and the performance of the closest competitors LetSum, MMR and SummaRuNNer. We observe that, even when DELSumm is used with automatically generated algorithmic labels (with about 15% noise), it still performs better than LetSum, MMR and SummaRuNNer (which were its closest competitors) according to all ROUGE measures except ROUGE-2 Fscore. This robustness of DELSumm is because, through the sentence informativeness and content words measures in the ILP formulation, DELSumm can capture useful information even if the rhetorical labels may be a little inaccurate.

9. Conclusion

We propose DELSumm , an unsupervised algorithm that systematically incorporates domain knowledge for extractive summarization of legal case documents. Extensive experiments and comparison with as many as eleven baselines, including deep learning-based approaches as well as domain-specific approaches, show the utility of our approach. The strengths our approach are: (i) DELSumm systematically encodes domain knowledge necessary for legal document summarization into a computational approach, (ii) although an unsupervised approach, it performs at par with supervised learning models trained over huge amounts of training data, (iii) it is able to provide a summary that has a balanced representation from all the rhetorical segments, which is highly lacking in prior approaches (iv) inaccuracy in labels do not degrade the performance of DELSumm much, thus showing its robustness and rich information identification capabilities, and (v) the method is flexible and generalizable to summarize documents from other jurisdictions; all that is needed are the expert guidelines for what to include the summary, and how to identify important content words. The objective function and constraints can be adjusted as per the requirements of different jurisdictions (e.g., giving more weight to certain segments). The implementation of DELSumm is publicly available at

In future, we plan to apply DELSumm to documents from other jurisdictions as well as to generate different types of summaries (e.g., for different stakeholders) and analyse the performance.

Acknowledgements: The authors thank the Law experts from the Rajiv Gandhi School of Intellectual Property Law, India who helped in developing the gold standard data and provided the guidelines for summarization. The research is partially supported by the TCG Centres for Research and Education in Science and Technology (CREST) through the project titled ‘Smart Legal Consultant: AI-based Legal Analytics’. This work is also supported in part by the European Union’s Horizon 2020 research and innovation programme under grant agreement No 832921. P. Bhattacharya is supported by a Fellowship from Tata Consultancy Services.


  • S. Banerjee, P. Mitra, and K. Sugiyama (2015) Multi-document abstractive summarization using ILP based multi-sentence compression. In

    Proc. International Conference on Artificial Intelligence

    Cited by: §2.1.
  • P. Bhattacharya, K. Hiware, S. Rajgaria, N. Pochhi, K. Ghosh, and S. Ghosh (2019a) A comparative study of summarization algorithms applied to legal case judgments. In Proc. European Conference on Information Retrieval, Cited by: §1, §2.5, §7.2, §7.2.
  • P. Bhattacharya, S. Paul, K. Ghosh, S. Ghosh, and A. Wyner (2019b) Identification of rhetorical roles of sentences in indian legal judgments. Proc. Legal knowledge and information systems (JURIX). Cited by: §2.5, §2.5, §2.5, §3, §8, §8, §8, §8.
  • Giles,Jessica (2015) Writing case notes and case comments. Note: Cited by: §1, §4, §4.
  • J. Cheng and M. Lapata (2016) Neural summarization by extracting sentences and words. In Proc. Annual Meeting of the Association for Computational Linguistics, Cited by: §2.3.
  • J. Devlin, M. Chang, K. Lee, and K. Toutanova (2019) BERT: pre-training of deep bidirectional transformers for language understanding. In Proc. NAACL-HLT, Cited by: §2.3.
  • Y. Dong (2018)

    A survey on neural network-based summarization methods

    CoRR abs/1804.04589. External Links: Link, 1804.04589 Cited by: §2.3.
  • G. Erkan and D. R. Radev (2004) LexRank: graph-based lexical centrality as salience in text summarization. J. Artif. Int. Res. 22 (1). Cited by: §1, §2.1, §6.1.
  • A. Farzindar and G. Lapalme (2004) Letsum, an automatic legal text summarizing system. Proc. Legal knowledge and information systems (JURIX). Cited by: §1, §1, §2.2, §2.5, §6.1.
  • Y. Gong and X. Liu (2001) Generic text summarization using relevance measure and latent semantic analysis. In Proc. International conference on Research and development in information retrieval (SIGIR), Cited by: §2.1, §6.1.
  • how-to-brief-case-cuny (2017) How to brief a case. Note: Cited by: §1, §4, §4.
  • [12] () Introduction to case briefing. Note: Cited by: §1, §4, §4.
  • H. Jing (2000) Sentence reduction for automatic text summarization. In

    Proc. Applied Natural Language Processing Conference

    Cited by: §6.1.
  • C. Liu and K. Chen (2019) Extracting the gist of chinese judgments of the supreme court. In Proc. International Conference on Artificial Intelligence and Law (ICAIL), Cited by: §1, §1, §1, §2.4, §6.1.
  • Y. Liu and M. Lapata (2019) Text summarization with pretrained encoders. In Proc. EMNLP-IJCNLP, Cited by: §1, §2.3, §6.1.
  • H.P. Luhn (1958) The automatic creation of literature abstracts. IBM Journal of Research Development,2(2). Cited by: §1, §2.1, §6.1.
  • R. Nallapati, F. Zhai, and B. Zhou (2017) SummaRuNNer: a recurrent neural network based sequence model for extractive summarization of documents. In Proc. AAAI Conference on Artificial Intelligence, Cited by: §1, §2.3, §6.1, §6.1.
  • S. Polsley, P. Jhunjhunwala, and R. Huang (2016) CaseSummarizer: a system for automated summarization of legal texts. In Proc. Iinternational conference on Computational Linguistics (COLING) System Demonstrations, Cited by: §1, §1, §2.2, §6.1.
  • K. Rudra, P. Goyal, N. Ganguly, P. Mitra, and M. Imran (2018) Identifying sub-events and summarizing disaster-related information from microblogs. In Proc. International Conference on Research & Development in Information Retrieval (SIGIR), Cited by: §2.1.
  • M. Saravanan, B. Ravindran, and S. Raman (2006) Improving legal document summarization using graphical models. In Legal knowledge and information systems, JURIX, Cited by: §1, §1, §2.2, §2.5, §6.1.
  • M. Saravanan, B. Ravindran, and S. Raman (2008) Automatic Identification of Rhetorical Roles using Conditional Random Fields for Legal Document Summarization. In Proc. International Joint Conference on Natural Language Processing, Cited by: §2.5.
  • S. Sotudeh, A. Cohan, and N. Goharian (2021) On generating extended summaries of long documents. In The AAAI-21 Workshop on Scientific Document Understanding (SDU 2021), Cited by: §2.3, footnote 2.
  • W. Xiao and G. Carenini (2019) Extractive summarization of long documents by combining global and local context. In Proc. EMNLP-IJCNLP, Cited by: §2.3.
  • H. Yamada, S. Teufel, and T. Tokunaga (2019) Neural network based rhetorical status classification for japanese judgment documents.. In Proc. Legal knowledge and information systems (JURIX), Cited by: §2.5.
  • L. Zhong, Z. Zhong, Z. Zhao, S. Wang, K. D. Ashley, and M. Grabmair (2019) Automatic summarization of legal decisions using iterative masking of predictive sentences. In Proc. International Conference on Artificial Intelligence and Law (ICAIL), Cited by: §1, §1, §2.2, §2.5, §6.1.