System and method for estimating accuracy of an automatic natural language translation

A computer system for a natural language translation, comprising:

• a translation process that translates a source natural language segment of one or more source words into a target natural language segment of one or more target words;
• a parser that parses the source natural language segment into a source parse structure with one or more source elements including one or more syntactic relationships among the source elements;
• and
• an evaluation process that determines a translation complexity of the computer system translating from the source natural language segment to the target natural language segment by measuring a complexity of choices during the translating and measuring a parsing complexity of the parsing, and further determines a confidence measure of the natural language translation by a complexity combination of the complexity of choices and the parsing complexity, the confidence measure indicating less confidence as the complexity combination increases.

A computer system for natural language translation, comprising:

• a parser that parses a source natural language segment of one or more source words into a source parse structure with one or more source elements including syntactic relationships among the source elements;
• a source evaluation process that generates a source indication of a complexity of choices in producing the source parse structure;
• a transfer process that converts the source parse structure into an initial transfer structure in which source elements are replaced by target elements and target relationships are created;
• a transfer evaluation process that generates a transfer indication of a complexity of choices in producing the initial transfer structure;
• a structural process that converts the initial transfer structure into a target language structure;
• a structural transfer evaluation process that generates a structural indication of a complexity of converting the initial transfer structure into the target language structure;
• a target generation process that converts the target language structure into a target natural language segment of one or more target words;
• a target generation evaluation process that generates a generation indication of a complexity of the converting the target language structure into the target natural language segment;
• and
• a combiner that aggregates any one or more of the source indication, the transfer indication, the structural indication, and the generation indication to produce a single measure of the complexity of the translation by the computer system and determines a confidence measure of a natural language translation that indicates less confidence as the complexity increases.

A method executed on a computer system for natural language translation, comprising the steps of:

• parsing a source natural language segment into a source parse structure with one or more source elements;
• translating the source parse structure into a target natural language segment through a target language structure;
• determining a translation complexity of translating the source natural language segment to the target natural language segment;
• and
• determining a confidence measure related to a parsing complexity that is a complexity of parsing the source natural language segment, where the confidence measure indicates less confidence as the parsing complexity increases.

A computer system for natural language translation, comprising:

• means for parsing a source natural language segment into a source parse structure with one or more source elements;
• means for translating the source parse structure into a target natural language segment through a target language structure;
• means for determining a translation complexity of translating the source natural language segment to the target natural language segment;
• and
• means for determining a confidence measure related to a parsing complexity that is a complexity of parsing the source natural language segment, where the confidence measure indicates less confidence as the parsing complexity increases.