Method and system for retrieval of prior court cases using witness testimonies
This disclosure relates generally to retrieval of prior court cases using witness testimonies. Conventional state-of-the-art methods use supervised techniques for answering basic questions in legal domain using numerous features and do not address interpretability of results and the performance and precision of retrieving prior court cases for these methods are less. Embodiments of the present disclosure obtains an embedded representation for an event structure of a user query and testimony sentences identified from prior court cases using a trained Bi-LSTM classifier and a set of linguistic rules. A similarity is estimated between the embedded representation for the event structure of the user query and the event structure of each testimony sentence from the prior court cases. Further a relevance score is assigned in accordance with the estimated similarity to retrieve the relevant prior court cases. The disclosed method is used to retrieve the relevant prior court cases using witness testimonies.
1. A processor implemented method for retrieving one or more prior court cases using witness testimonies, the method comprising:
receiving, via one or more hardware processors, a user query for retrieving one or more prior court cases, wherein the one or more prior court cases comprise a set of testimony sentences and a set of non-testimony sentences, wherein the user query and each testimony sentence of the set of testimony sentences comprise an event structure with a predicate and one or more arguments;
obtaining an embedded representation for the event structure of the user query, using a trained denoising auto-encoder executed via the one or more hardware processors, wherein the denoising auto-encoder is trained using one or more testimony sentences from the set of testimony sentences wherein the one or more testimony sentences are identified using a trained classifier and a set of linguistic rules, wherein the trained classifier is trained using a set of training data, wherein the set of training data comprises of
(i) the set of testimony sentences satisfying a set of predefined linguistic rules,
(ii) the set of non-testimony sentences satisfying a subset of the set of predefined linguistic rules and
(iii) a set of sentences which are neither identified as testimony nor non-testimony by the linguistic rules and
wherein the set of predefined linguistic rules satisfied by the set of testimony sentences are:
(i) presence of explicit or implicit witness mentions wherein the implicit mentions is any one of (a) pronouns (b) person-indicating common nouns (c) actual person names,
(ii) presence of at least one statement-indicating verb,
(iii) dependency subtree rooted at the at least one statement indicating verb should contain at least one of the following: a clausal complement or open clausal complement,
(iv) the statement verb should not have a child which negates itself, and
(v) the statement verb should have at least one witness mention within its subject or agent dependency subtree but should not have any legal role mention within its subject or agent dependency subtree;
estimating, via the one or more hardware processors, a similarity between the obtained embedded representation of the user query and an embedded representation of each of the one or more testimony sentences using a similarity function;
assigning a relevance score, via the one or more hardware processors, to the one or more prior court cases associated with the one or more testimony sentences in accordance with the estimated similarity; and
retrieving a predefined number of prior court cases out of the one or more prior court cases, wherein the predefined number is decided based on the assigned relevance score.
2. The processor implemented method as claimed in claim 1 further comprising training the denoising auto-encoder, the training comprising:
(i) encoding the event structure by masking either the predicate or one or more arguments of the event structure of each of the one or more testimony sentences;
(ii) reconstructing the embedded representation of the encoded event structure of each of the one or more testimony sentences; and
(iii) storing the embedded representation of each of the one or more testimony sentences in the database.
3. The processor implemented method as claimed in claim 1 , wherein the similarity function is a maximum cosine similarity function between the embedded representation of the user query and the one or more testimony sentences.
4. The processor implemented method as claimed in claim 1 , wherein the identified one or more testimony sentences and the one or more non-testimony sentences are stored in a database.
5. The processor implemented method as claimed in claim 1 , wherein the subset of the set of predefined linguistic rules satisfied by the set of non-testimony sentences are (i) presence of at least one statement-indicating verb, (ii) dependency subtree rooted at the at least one statement indicating verb should contain at least one of the following: a clausal complement or open clausal complement, (iii) the statement verb should have at least one legal role mention within its subject or agent dependency subtree.
6. A system ( 100 ), comprising:
a memory ( 104 ) storing instructions;
one or more communication interfaces ( 106 ); and
one or more hardware processors ( 102 ) coupled to the memory ( 104 ) via the one or more communication interfaces ( 106 ), wherein the one or more hardware processors ( 102 ) are configured by the instructions to:
receive a user query for retrieving one or more prior court cases, wherein the one or more prior court cases comprise a set of testimony sentences and a set of non-testimony sentences, wherein the user query and each testimony sentence of the set of testimony sentences comprise an event structure with a predicate and one or more arguments;
obtain an embedded representation for the event structure of the user query, using a trained denoising auto-encoder, wherein the denoising auto-encoder is trained using one or more testimony sentences from the set of testimony sentences wherein the one or more testimony sentences are identified using a trained classifier and a set of linguistic rules, wherein the trained classifier is trained using a set of training data, wherein the set of training data comprises of
(i) the set of testimony sentences satisfying a set of predefined linguistic rules,
(ii) the set of non-testimony sentences satisfying a subset of the set of predefined linguistic rules and
(iii) a set of sentences which are neither identified as testimony nor non-testimony by the linguistic rules and
wherein the set of predefined linguistic rules satisfied by the set of testimony sentences are:
(i) presence of explicit or implicit witness mentions wherein the implicit mentions is any one of (a) pronouns (b) person-indicating common nouns (c) actual person names,
(ii) presence of at least one statement-indicating verb,
(iii) dependency subtree rooted at the at least one statement indicating verb should contain at least one of the following: a clausal complement or open clausal complement,
(iv) the statement verb should not have a child which negates itself, and
(v) the statement verb should have at least one witness mention within its subject or agent dependency subtree but should not have any legal role mention within its subject or agent dependency subtree;
estimate a similarity between the obtained embedded representation of the user query and an embedded representation of each of the one or more testimony sentences using a similarity function;
assign a relevance score to the one or more prior court cases associated with the one or more testimony sentences in accordance with the estimated similarity; and
retrieve a predefined number of prior court cases out of the one or more prior court cases, wherein the predefined number is decided based on the assigned relevance score.
7. The system of claim 6 , further comprising training the denoising auto-encoder, the training comprising:
(i) encoding the event structure by masking either the predicate or one or more arguments of the event structure of each of the one or more testimony sentences;
(ii) reconstructing the embedded representation of the encoded event structure of each of the one or more testimony sentences; and
(iii) storing the embedded representation of each of the one or more testimony sentences in the database.
8. The system of claim 6 , wherein the similarity function is a maximum cosine similarity function between the embedded representation of the user query and the one or more testimony sentences.
9. The system of claim 6 , wherein the identified one or more testimony sentences and the one or more non-testimony sentences are stored in a database.
10. The system of claim 6 , wherein the subset of the set of predefined linguistic rules satisfied by the set of non-testimony sentences are (i) presence of at least one statement-indicating verb, (ii) dependency subtree rooted at the at least one statement indicating verb should contain at least one of the following: a clausal complement or open clausal complement, (iii) the statement verb should have at least one legal role mention within its subject or agent dependency subtree.
11. One or more non-transitory machine readable information storage mediums comprising one or more instructions, which when executed by one or more hardware processors causes a method for:
receiving a user query for retrieving one or more prior court cases, wherein the one or more prior court cases comprise a set of testimony sentences and a set of non-testimony sentences, wherein the user query and each testimony sentence of the set of testimony sentences comprise an event structure with a predicate and one or more arguments;
obtaining an embedded representation for the event structure of the user query, using a trained denoising auto-encoder wherein the denoising auto-encoder is trained using one or more testimony sentences from the set of testimony sentences wherein the one or more testimony sentences are identified using a trained classifier and a set of linguistic rules, wherein the trained classifier is trained using a set of training data, wherein the set of training data comprises of
(i) the set of testimony sentences satisfying a set of predefined linguistic rules,
(ii) the set of non-testimony sentences satisfying a subset of the set of predefined linguistic rules and
(iii) a set of sentences which are neither identified as testimony nor non-testimony by the linguistic rules and
wherein the set of predefined linguistic rules satisfied by the set of testimony sentences are:
(i) presence of explicit or implicit witness mentions wherein the implicit mentions is any one of (a) pronouns (b) person-indicating common nouns (c) actual person names,
(ii) presence of at least one statement-indicating verb,
(iii) dependency subtree rooted at the at least one statement indicating verb should contain at least one of the following: a clausal complement or open clausal complement,
(iv) the statement verb should not have a child which negates itself, and
(v) the statement verb should have at least one witness mention within its subject or agent dependency subtree but should not have any legal role mention within its subject or agent dependency subtree;
estimating a similarity between the obtained embedded representation of the user query and an embedded representation of each of the one or more testimony sentences using a similarity function;
assigning a relevance score to the one or more prior court cases associated with the one or more testimony sentences in accordance with the estimated similarity; and
retrieving a predefined number of prior court cases out of the one or more prior court cases, wherein the predefined number is decided based on the assigned relevance score.