Knowledge-Based Machine Translation (MT)

1.2.2. Definition of Machine Translation

Machine Translation, sometimes referred to by the abbreviation MT (not to be confused with Computer-Assisted Translation (CAT), Machine-Aided Human Translation (MAHT) and interactive translation) is a sub-field of computational linguistics that investigates the use of computer software in translating from source language (SL) into the target language (TL) (Reddy & Hanumanthappa , 2013, p. 64).

“Machine translation (MT) is the process whereby a computer program translates a text from one natural language into another […].The main focus of MT is to produce high-quality natural language output, translating from one language into another, quickly and cost-effectively.” (Flanagan, 2009, p. 27). Hutchins

Computational Architecture

1.2.3.2.1. Rule-Based Machine Translation (RBMT)

It is also known as ‘Knowledge-Based Machine Translation’ or the ‘Classical Approach’ of MT. It concerns machine translation systems based on the encoding of linguistic information about source and target languages (morphology, syntax, semantics, etc.) in the form of rules (Soudi, Farghali, Neumann, & Zbib, 2012, p. 2). Traditional rule-based approaches to MT require large scale grammars and rules, which means that developers require extensive linguistic expertise and a substantial amount of manual labour (Gough, 2005, p. 1).

There are three different types of rule-based machine translation systems: Direct Systems, also called (Dictionary Based Machine Translation) which use basic rules i.e. a dictionary will map each word in the source language to the appropriate word in the target language, Transfer RBMT Systems which use morphological and syntactical analysis and Interlingua RBMT Systems which use an abstract meaning (Koehn, 2010, p. 15). The problem with Rule-Based Machine Translation (RBMT) systems is that they are unable to learn from their mistakes according to Winiwarter, 2007 (as cited in Flanagan, 2009, p. 30).

1.2.3.2.2. Corpus-Based Approach

1.2.3.2.2.1. Statistical Machine Translation (SMT)

“It is a purely statistical, language independent approach implementing a highly developed mathematical theory of probability distribution and probability estimation” Carl and Way, 2003 (as quoted in Flanagan, 2009, p. 31). However, Lopez (2008) defines SMT as “an approach to MT that is characterized by the use of machine learning methods” (p. 2). He adds that SMT treats translation as a machine learning problem via applying algorithm to text previously translated which is known as parallel corpus. (p.2)

It integrates a word-level alignments and phrasal alignment as well, taking into account source text words, phrase co-occurrence, etc. (Koehn, 2010, p. 56). In other words, it does not use the sentence example directly, but rather uses derived units from the data.

Flanagan (2009) states that SMT is a more dominant approach that is usually used in commercial translation engines such as Google Translate (p. 31). Soudi et al. (2012) add that it is the approach that makes use of supervised machine learning algorithms in order to learn translation model parameters from a corpus of translated sentences (p.