AI and Law
(Legal) argument mining
- Wyner 2010 - Approaches to text mining arguments from legal cases
- Atkinson 2017 - Towards Artificial Argumentation
Ontologies and topic modelling
- Panagis 2016- On top of topics leveraging topic modelling to study dynamic case law
- Ferre 2018 - Combining rule-based and embedding-based approaches to normalize textual entities with an ontology
Text similarity, recommendation, information retrieval
- Landthaler 2016 - Extending full text search for legal documents using word embeddings
- Nanda2019 - Unsupervised and supervised text similarity systems for automated identification of national implementing measures of European directives
- VanOpijnen 2014 - Op en in het web: Hoe toegankelijkheid van rechterlijke uitspraken kan worden verbeterd
- Mandal 2017 - Measuring similarity among legal court case documents
- Uses doc2vec
- Zhong (Unpublished) - Multi-representation law article recommendation framework incorporating external knowledge of law articles
Case factors, classifying cases, predicting case outcomes
- Terrible paper: Aubaid-Text classification using Word Embedding in Rule-based Methodologies
- WIP BrueninghausGenerating_legal_arguments_and_predictions_from_case_texts.md
- Westermann 2019 - Using Factors to Predict and Analyze Landlord Tenant Decisions
- Falakmasir 2017 - Utilizing Vector Space Models for Identifying Legal Factors from Text
Argument mining
See index argumentation mining .
Since I'm currently interested in modelling attack relations between arguments/conclusions, it may be interesting for me to check out Dung's abstract argumentation framework because 1) it does not consider internal structure of arguments, 2) it is apparently very good at modelling attack relations.
There is also argument mining work that targets a specific argument component, e.g. the claim. Have a look at that as well, this may be sufficient for me! See Lippi & Torrini (2015). This would also make constructing the training data easier, since it would only require positive labels for claims.
Use cases
- Prediction case outcomes
- Finding similar cases (e.g. in common law)
- E.g. cases on similar topics
- Discussing similar laws
- same or similar case factors
- Finding relevant case factors
- Argument extraction
- Distinguish legal from non-legal sentences or arguments
- Given a directive (e.g. EU directive), automatically find laws that implement the directive
- Given a particular case, find related legislative documents.
TODO
Case-based reasoning:
- J. L. Kolodner. An Introduction to Case-Based Reasoning. Artificial Intelligence Review, 6:3–34, 1992
Argument mining
- Statistical + rule-based [A] Robaldo L, Caselli T, Russo I, Grella M (2011) From Italian text to timeml document via dependency parsing. In: Computational linguistics and intelligent text processing—12th international conference, CICLing 2011, Tokyo, Japan, 2011, pp 177–187
- Medvedeva 2020 - Using machine learning to predict decisions of the European Court of Human Rights
- Savelkla 2021 - Transfer of predictive models across legal domains
- [C] Craandijk 2020 - Deep Learning for Abstract Argumentation Semantics
These systems "model how lawyers in common-law jurisdictions make use of past decisions when arguing a case" (29):
- HYPO by Kevin Ashley
- CATO by Vincent Aleven
Maybe relevant for figuring out how argument mining helps in finding relevant cases:
- Daniels, J.J., Rissland, E.L.: Finding legally relevant passages in case opinions. In: ICAIL 1997: Proceedings of the 6th International Conference on Artificial intelligence and Law, pp. 39–46. ACM, New York (1997)
- Brüninghaus, S., Ashley, K.D.: Generating legal arguments and predictions from case texts. In: ICAIL 2005, pp. 65–74. ACM Press, New York (2005)
Word embeddings used by Panagin 2016 en Landthaler 2016
Explainable word embeddings; connecting vectors and concepts
- Qureshi, EVE: Explainable Vector Based Embedding Technique Using Wikipedia, https://arxiv.org/abs/1702.06891; https://link.springer.com/article/10.1007/s10844-018-0511-x
-
Jang 2019 - Selection and Interpretation of Embedding Subspace for Query Classification;
- Uses SHAP to identify embedding subspaces that are strongly connected to particular queries
- Similar strategy for recommending relevant documents would be: which "concepts" in this document make it relevant to the query (which in this case is the reference document)
-
Stephen Mcgregor, Kat Agres, Matthew Purver, and Geraint Wiggins@qmul Ac Uk. 2015. From Distributional Semantics to Conceptual Spaces: A Novel Compu- tational Method for Concept Creation Geraint A. Wiggins. Journal of Artificial General Intelligence 6, 1 (2015), 55–86. https://doi.org/10.1515/jagi-2015-0004
- Jang 2019 about this paper: "For instance, the authors of [33] proposed a model to bridge the gap between distributional semantics (vector space) and conceptual space. The model dynamically selects specific dimensions associated with seed terms, which form a sub-space of terms defining the related concept."
3, 37, 51: finding more conceptually coherent dimensions in embeddings by using sparse coding
Enforcing sparse embeddings
Data
- Data van Nederlandse jurisprudentie
- European Court of Human Rights (ECHR)
- Used a lot because it's highly structured
- Also used by Mochales and Moens (2011) but annotations aren't public afaik
- If annotations would be public, I could try to transfer them from English to Dutch if there is overlap with the very modest amount of documents that are available both in English and Dutch. See here
- See an overview of corpora for argument mining in Lippi et al. (2016), specifically page 15.
- Table of argument mining corpora
Links
- Computation for argument mining
- Framework for computational persuasion
- Online TOAST structured argumentation framework