Semantic map generation from natural-language-text documents
A computer-implemented process includes obtaining a natural-language-text document comprising a first and second clause and determining first and second embedding sequences based on n-grams of the first and second clauses. The process includes generating data model objects based on the embedding sequences and determining an association between the first data model object and the second data model object based on a shared parameter of the first and second clauses. The process includes receiving a query including the first category and the first n-gram and causing a presentation of a visualization of data model objects that includes shapes based on the data model objects and a third shape based on the association between the first data model object and the second data model object.
1 . A computer-implemented method of extracting a knowledge graph from a legal document, the method comprising:
obtaining, with a computer system, an unstructured natural language legal document;
converting, with the computer system, tokens in the natural language legal document into a sequence of embedding vectors corresponding to respective tokens;
generating, with the computer system, with a trained language model, based on the sequence of embedding vectors, a set of semantic triples characterizing rights, obligations, or prohibitions of the unstructured natural language legal document; and
storing, with the computer system, the set of semantic triples in memory, wherein the language model is configured to process a current representation vector corresponding to part of the sequence based on a plurality of attention heads that apply attention weights to a plurality of previous representation vectors corresponding to other parts of the sequence to affect processing of the current representation vector corresponding to the part of the sequence.
2 . The method of claim 1 , wherein the set characterizes rights, obligations, and prohibitions of the unstructured natural language legal document.
3 . The method of claim 1 , comprising generating a visualization based on the set of semantic triples.
4 . The method of claim 1 , comprising reasoning, with the computer system, about the legal document based on a knowledge graph formed by the set of semantic triples.
5 . The method of claim 1 , wherein one of the semantic triples has as a first field, a party to an agreement, as a second field, an indication that the party is obligated to do something, and as a third field, a description of what the party is obligated to do.
6 . The method of claim 1 , comprising structuring information of the legal document into a normative, semantically annotated form.
7 . The method of claim 1 , wherein at least one of the plurality of attention heads is configured to form a query vector, a key vector, and a value vector.
8 . The method of claim 1 , comprising annotating the embedding vectors with part-of-speech indicators.
9 . The method of claim 8 , wherein the part-of-speech indicators are vectors in another embedding space.
10 . The method of claim 1 , comprising:
using the set of semantic triples to answer a what-if question about the legal document.
11 . A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising:
obtaining, with a computer system, an unstructured natural language legal document;
converting, with the computer system, tokens in the natural language legal document into a sequence of embedding vectors corresponding to respective tokens;
generating, with the computer system, with a trained language model, based on the sequence of embedding vectors, a set of semantic triples characterizing rights, obligations, or prohibitions of the unstructured natural language legal document; and
storing, with the computer system, the set of semantic triples in memory, wherein the language model is configured to process a current representation vector corresponding to part of the sequence based on a plurality of attention heads that apply attention weights to a plurality of previous representation vectors corresponding to other parts of the sequence to affect processing of the current representation vector corresponding to the part of the sequence.
12 . The medium of claim 11 , wherein the set characterizes rights, obligations, and prohibitions of the unstructured natural language legal document.
13 . The medium of claim 11 , the operations comprising generating a visualization based on the set of semantic triples.
14 . The medium of claim 11 , the operations comprising reasoning, with the computer system, about the legal document based on a knowledge graph formed by the set of semantic triples.
15 . The medium of claim 11 , wherein one of the semantic triples has as a first field, a party to an agreement, as a second field, an indication that the party is obligated to do something, and as a third field, a description of what the party is obligated to do.
16 . The medium of claim 11 , the operations comprising structuring information of the legal document into a normative, semantically annotated form.
17 . The medium of claim 11 , wherein at least one of the plurality of attention heads is configured to form a query vector, a key vector, and a value vector.
18 . The medium of claim 11 , the operations comprising annotating the embedding vectors with part-of-speech indicators.
19 . The medium of claim 18 , wherein the part-of-speech indicators are vectors in another embedding space.
20 . The medium of claim 11 , the operations comprising:
using the set of semantic triples to answer a what-if question about the legal document.
21 . A computer-implemented method of extracting a knowledge graph from a legal document, the method comprising:
obtaining, with a computer system, an unstructured natural language legal document;
converting, with the computer system, tokens in the natural language legal document into a sequence of embedding vectors corresponding to respective tokens;
generating, with the computer system, with a trained language model, based on the sequence of embedding vectors, a set of semantic triples characterizing rights, obligations, or prohibitions of the unstructured natural language legal document; and
storing, with the computer system, the set of semantic triples in memory, wherein the language model is configured to process a current embedding vector in the sequence based on an internal memory gate that causes a previous embedding vector in the sequence to affect the processing of the current embedding vector in the sequence.