Linguistic Data in Machine Translation Systems

Study Notes

Linguistic Data in MT Systems

Linguistic data in MT systems can be divided into two categories: lexical data and grammatical data
Grammatical data refers to the information used by analysis and generation routines, stated in terms of acceptable combinations of categories and features
Lexical data refers to the specific information about each individual lexical item (word or phrase) in the vocabulary of the languages concerned

Lexical Data

Lexical data in MT systems differs from that found in conventional dictionaries
No need for information about pronunciation, etymology, synonyms, definitions, or examples of usage
Required information includes:
- Grammatical category
- Morphological type
- Subcategorization features
- Valency information
- Case frames
- Semantic features
- Selection restrictions
The term lexicon is preferred to 'dictionary' due to these differences

Design Decisions and Organization of Lexical Data

Design decisions have implications for the organization of lexical data
In MT systems of the direct translation design, there is typically one bilingual lexicon containing data about lexical items of the source language and their equivalents in the target language
Each entry in the lexicon combines grammatical data for the source item, its target language equivalents, and the information necessary to select between target language alternatives and change syntactic structures

Direct Translation Approach

The direct approach is an MT strategy that lacks intermediate stages in translation processes
Processing of the source language input text leads directly to the desired target language output text
The approach is still valid today, but first-generation direct MT systems have a more primitive software design
First-generation direct MT systems:
- Began with a morphological analysis phase
- Identified word endings and reduced inflected forms to their uninflected basic forms
- Input results into a large bilingual dictionary look-up program
- No analysis of syntactic structure or semantic relationships
- Lexical identification depended on morphological analysis and led directly to bilingual dictionary look-up providing target language word equivalences
- Followed by local reordering rules to give more acceptable target language output
- Target language text was produced
The direct approach can be characterized as 'word-for-word' translation with some local word-order adjustment
It has severe limitations, resulting in poor translation quality

Description

This quiz is about the types of linguistic data required in Machine Translation (MT) systems, including lexical and grammatical data. It covers the information embodied in grammars and specific information about individual lexical items.