Appraches to machine translation

Appraches to machine translation

1. Subfield of computational linguistics that investigates the use of computer software to translate text or speech from one natural language to another (Arnold et al., 1994).

   • Goal- To develop a system that accurately produces a good translation between human languages (Adeoye, 2012).
   • Approaches
     • Single
       • Charaacteristics
         • Fails to achieve satisfactory performance. Sometimes inconsistent, inflexible for largescale applications Uses shallower representation of knowledge. Resulting in lower quality and fluency of the output.
       • Methods
         • Rule-Based Machine Translation (RBMT)
           • Involves the application of morphological, syntactic and semantic rules in the analysis of the source language text and the synthesis of the target-language text.
           • Helps to handle word-order problems and trace parse errors using linguistic knowledge.
           • Can be divided into: Transfer based Interlingua based
         • Direct-based Machine Translation (DBMT)
           • 1ST Replaces words in the source language with words in the target language in the same sequence without much linguistic analysis and processing.
             • 2ND Syntactic rearrangement and re-ordering
         • Corpus-Based Approach
           • Characterized by the use of machine learning methods. Learning algorithm applied to a large body of previously translated text (parallel corpus). Learner is able to translate previously unseen sentences.
         • Knowledge-Based Approach
           • Uses extensive semantic and pragmatic knowledge in machine translation.
           • Entails the ability to reason about concepts
     • Hybrid
       • Characteristics
         • Combines the strength of two or more approaches to improve the overall quality of machine translation.
       • Methods
         • Word-based Models
           • Model the lexical dependencies between single words.
         • Phrase-based Models
           • Introduction of phrases as the basic translation unit. Phrases Can be any substring Allow Local re-orderings Translation of short idioms Insertions and deletions sensitive to local context.
         • Syntax based Model
           • String-Based Systems
             • Input is a string to be simultaneously parsed and translated by a synchronous grammar.
           • Tree-Based Systems
             • Perform translation in two separate steps: Parsing Decoding
         • Tree-Based Systems
           • Tree-to-string
             • Input is a string Output is a parse tree
           • String-to-tree
             • Parses a tree at the input Outputs a string
           • Tree-to-tree
             • Extracts rules using parse trees from both side(s) of the bi-text.