Patent EP0887748A2 successfully refreshed from Google Patents.
EP0887748A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Multilingual terminology extraction system
Assignee: International Business Machines Corp
•
Filing: 1998-04-07
•
Granted: —
•
Expires: 2018-04-07
Abstract
Abstract The present invention is a terminology extraction system which allows for automatic creation of bilingual terminology. A source text which comprises at least one sequence of source terms, is aligned with a target text which also comprises at least one sequence of target terms. The term extractor builds a network from each source and target sequence wherein each node of the network comprises at least one term and such that each combination of source terms is included within one source node, whereas each combination of target terms is included within one target node. The term extractor links each source node with each target node, and through a flow optimization method selects relevant links in the resulting network. Once the term extractor has been run on the entire set of aligned sequences, a term statistics circuit computes an association score for each pair of linked source/target terms, and finally the scored pairs of linked source/target terms that are considered relevant bilingual terms are stored in a bilingual terminology database. The whole process can be iterated in order to improve the strength of the bilingual links.
Inventors
- 👤 Jean-Marc Lange
- 👤 Eric Gaussier
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-10 00:51
CPC Classifications
G; G06; G06F; G06F40/00; G06F40/40; G06F40/42; G06F40/45
Claims Analysis
Total Claims: 11
Claim 1
Independent Claim
211 words
A computer based terminology extraction system (205) for creating bilingual terminology from a source text aligned with a target text, the source text comprising at least one sequence of source terms, the target text comprising at least one sequence of target terms, a term being composed of at least one word, said system comprising: a term extractor means (304) which operates on at least one pair extracted from the aligned texts and consisting of a source sequence aligned with a target sequence. a term statistics means (306) coupled to said term extractor means for computing a statistical association score for each pair of source and target terms detected in the whole set of aligned source/target sequences; a memory means (312) coupled to said term statistics means for storing the scored pairs of source/target terms that are considered relevant bilingual terms; characterized in that said term extractor means comprises means for building a network (402,404) wherein each node (Si,Tj) comprises at least one term from said at least one pair of aligned source/target sequences, such that each combination of source terms is included within one source node and each combination of target terms is included within one target node; and means for linking each source node (Si) with each target node (Tj).
Claim 2
Dependent Claim
37 words
The system of claim 1 further comprising means for operating a flow optimization algorithm such that each link (406,408) between a source node (Si) and a target node (Tj) is characterized by a capacity and a flow.
Claim 3
Dependent Claim
16 words
The system of claim 2 wherein said flow optimization algorithm is a maximum flow/minimum cost algorithm.
Claim 4
Independent Claim
44 words
The system of claims 1 to 3 wherein said linking means further comprises means for selecting a group of source nodes not to be linked to target nodes according to predetermined criteria, such as ignoring source nodes already stored in said memory means (312).
Claim 5
Independent Claim
35 words
The system of claims 1 to 4 further comprising a word statistics means (302) for computing association scores for each pair of source and target terms wherein said source terms are composed of one word.
Claim 6
Dependent Claim
15 words
A machine-aided translation system (103) comprising the system of anyone of claims 1 to 5.
Claim 7
Independent Claim
169 words
A method for creating bilingual terminology from a source text comprising at least one sequence of source terms aligned with a target text comprising at least one sequence of target terms, a term being composed of at least one word, said method comprising the steps of: a) reading a first pair consisting of a source sequence aligned with a target sequence; b) building a network (402,404) wherein each node (Si,Tj) comprises at least one term from said first pair of aligned source/target sequences, such that each combination of source terms is included within one source node and each combination of target terms is included within one target node; c) linking each source node (Si) with each target node (Tj); d) computing an association score for each pair of linked source and target nodes; e) selecting at least one link such that the corresponding source/target linked terms are considered relevant bilingual terms; f) storing the relevant bilingual terms; g) repeating steps a) to f) for all pairs of aligned sequences.
Claim 8
Dependent Claim
35 words
The method of claim 7 further comprising after step g) the step of: h) computing alignment statistics for the stored bilingual terms according to a scoring function; i) selecting scored bilingual terms as relevant candidates.
Claim 9
Dependent Claim
74 words
The method of claim 8 further comprising after step i) : j) using the statistical results of step h) for each pair of source and target terms to improve the single word statistics for those words which compose said terms; k) updating the association scores used in step d) according to the results of step j); l) repeating steps a) to k) with the updated single word statistics until an end criterion is met.
Claim 10
Dependent Claim
51 words
The method of claim 9 further comprising after step i): m) repeating steps a) to k) until an end criterion is met, while not taking into account, in the network building step b), those nodes and links associated with terms that have been selected in step i) of the previous iteration.
Claim 11
Independent Claim
21 words
The method of claims 7 to 10 wherein the selection step e) is performed according to a network flow optimization algorithm.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.