Machine translation

• US 20050015240A1
• Filed: 10/29/2002
• Published: 01/20/2005
• Est. Priority Date: 10/29/2001
• Status: Active Grant

• Current Assignee: British Telecommunications PLC (BT Group PLC)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Appleby, Stephen C
• Granted Patent: US 7,565,281 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/45 Example-based machine translation; Alignment

Abstract

A computer natural language translation system, comprising: means for inputting source language text; means for outputting target language text; transfer means for generating said target language text from said source language text using stored translation data generated from examples of source and corresponding target language texts, characterised in that said stored translation data comprises a plurality of translation components, each comprising: surface data representative of the order of occurrence of language units in said component; dependency data related to the semantic relationship between language units in said component; and link data linking dependency data of language components of said source language with corresponding dependency data of language components of said target language, and in that said transfer means is arranged to used said surface data of said source language in analysing the source language text, and said surface data of said target language in generating said target language text, and said dependency data and link data in transforming the analysis of said source text into an analysis for said target language.

Claims

1. A computer natural language translation system, comprising:

means for inputting source language text;

means for outputting target language text;

transfer means for generating said target language text from said source language text using stored translation data generated from examples of source and corresponding target language texts, characterised in that said stored translation data comprises a plurality of translation units, each comprising;

respective surface data representative of the order of occurrence of language units of said source and target languages;

respective dependency data related to the semantic relationship between said language units of said source and target languages; and

the dependency data of language units of said source language being aligned with corresponding dependency data of language units of said target language, and in that said transfer means is arranged to use said surface data of said source language in analysing said source language text, and said surface data of said target language in generating said target language text, and said dependency data in transforming the analysis of said source text into an analysis for said target language.

2. A system according to claim 1, in which, for some of said translation units, the number of said surface language units differs from the number of corresponding said dependency language units.

3. A system according to claim 1, in which for each said translation unit said surface data and said dependency data each comprise, for each language, a head language unit, and a data structure for storing daughter language units, each linked to said head language unit.

4. A system according to claim 3, in which said language units comprise constant data defining words.

5. A system according to claim 3, in which said language units comprise placeholder variables.

6. A system according to claim 1, in which said translation units include data from which a said source language analysis can be constructed which can be represented in the form of a graph.

7. A system according to claim 6, in which the transfer means is arranged to attempt to generate a source surface analysis graph using said surface data, and a source dependency graph using said dependency data.

8. A system according to claim 6, in which the transfer means is arranged to attempt to generate a target dependency graph using said dependency data and a target surface analysis graph using said surface data.

9. A system according to claim 6, in which said surface data is such that the or each said surface graph is constrained to have a tree structure.

10. A system according to claim 9, in which said tree is a planar tree.

11. A system according to claim 9, in which said dependency data is such that the or each said dependency graph is not constrained to have a planar tree structure.

12. A system according to claim 1, in which the alignment of the dependency data of language units of said source language with corresponding dependency data of language units of said target language comprises pairing of corresponding said source and target language units in a single paired record.

17. A computer program comprising code to execute on a computer to cause said computer to act as the system of claim 1.

13. A computer language translation development system, for developing data

for use in translation, comprising:

means for allowing corresponding source and target example texts to be linked into source and target language dependency graphs;

means for allowing corresponding translatable nodes of said source and target language dependency graphs representing translatable parts of the source and target texts to be aligned; and

means for automatically generating, from said source and target language dependency graphs, respective associated surface representative graph having a tree structure.

14. A system according to claim 13, in which the generating means comprises means for locating within a dependency graph an intermediate node which is located between two directly or indirectly translatable nodes;

and for causing the translatable node which depends on that intermediate node within said dependency graph to directly depend instead upon the translatable node on which said intermediate node depends, in said surface representative structure.

15. A system according to claim 13, in which the generating means comprises means for locating within a dependency graph a dependent node which depends on two translatable nodes, and for causing one of said translatable nodes, in said surface representative structure, to depend on the dependent node.

16. A system according to claim 13, in which the generating means comprises means for locating within a dependency graph a dependent node which depends from a higher node, so as to violate a planarity constraint, and for causing, in said surface representative structure, said dependent node to depend upon a node from which said higher node depends directly or indirectly, so as to satisfy said constraint in said surface representative structure.

18. A method for inferring a surface tree representation of the analysis of a text from a dependency graph comprising;

simplification of the dependency graph in a manner that is dependent on the alignment between source and target dependency graphs,

creation of an initial (not necessarily planar) surface tree from the above simplified dependency graph, and

creation of a final planar surface tree from the above surface tree.

19. A method of generating translation units for a computer natural language translation system, the method comprising the steps:

(a) providing examples of translations of source language text into target language text, and for each such example;

(b) generating for the source language text and target language text respective dependency graphs in which the respective language unit orders are retained;

(c) linking language units of the source language dependency graph with respective corresponding language units of the target language dependency graph where those language units are deemed directly translatable;

(d) processing the dependency graphs with a transformation algorithm to detect plural root nodes and, upon such detection, making one of the plural root nodes into a daughter, thereby producing corresponding trees having a single root node;

(e) processing said corresponding trees to detect any parent/daughter node dependency violating the planar tree projection constraint and, upon such detection, making the relevant daughter node dependent from a higher node of the tree; and

(if) for each linked pair of language units, generating a respective translation unit comprising those linked language units as respective head language units together with daughter language units, if any, and together with data relating to dependency change as the result of the operation of steps (d) and (e).