[Paper Review] Self-Organizing Machine Translation: Example-Driven Induction of Transfer Functions

Patrick Juola|ArXiv.org|Jun 3, 1994

Natural Language Processing Techniques3 citations

TL;DR

This paper proposes a self-organizing machine translation system that induces transfer functions from bilingual sentence pairs using a novel *marker-normal form* grammar formalism, embedded in a multivariate optimization framework. The approach successfully infers English-to-French and English-to-Urdu transfer functions from small corpora, achieving linguistically plausible and competitively accurate results compared to Hidden Markov Model baselines.

ABSTRACT

With the advent of faster computers, the notion of doing machine translation from a huge stored database of translation examples is no longer unreasonable. This paper describes an attempt to merge the Example-Based Machine Translation (EBMT) approach with psycholinguistic principles. A new formalism for context- free grammars, called *marker-normal form*, is demonstrated and used to describe language data in a way compatible with psycholinguistic theories. By embedding this formalism in a standard multivariate optimization framework, a system can be built that infers correct transfer functions for a set of bilingual sentence pairs and then uses those functions to translate novel sentences. The validity of this line of reasoning has been tested in the development of a system called METLA-1. This system has been used to infer English->French and English->Urdu transfer functions from small corpora. The results of those experiments are examined, both in engineering terms as well as in more linguistic terms. In general, the results of these experiments were psycho- logically and linguistically well-grounded while still achieving a respectable level of success when compared against a similar prototype using Hidden Markov Models.

Motivation & Objective

• To develop a machine translation system that learns transfer functions directly from bilingual example pairs, avoiding reliance on hand-coded rules.
• To integrate psycholinguistic principles into machine translation by modeling language data using a formalism compatible with cognitive theories.
• To enable the system to generalize from small parallel corpora to translate novel sentences using inferred transfer functions.
• To evaluate the linguistic and engineering validity of the induced transfer functions in cross-lingual translation tasks.
• To compare the performance of the example-driven system against a Hidden Markov Model-based prototype in terms of accuracy and linguistic plausibility.

Proposed method

• The system uses a new formalism called *marker-normal form*, a variant of context-free grammars designed to align with psycholinguistic models of language processing.
• Bilingual sentence pairs are represented in marker-normal form to encode syntactic and semantic correspondences between languages.
• A multivariate optimization framework is applied to infer the optimal transfer functions that map source-language structures to target-language structures.
• The optimization process minimizes a loss function based on the alignment quality between example sentence pairs and the induced transfer rules.
• The system generalizes by applying the learned transfer functions to novel, unseen sentence pairs for translation.
• The method is evaluated using small parallel corpora for English-French and English-Urdu translation tasks.

Experimental results

Research questions

• RQ1 Can transfer functions be effectively induced from small bilingual corpora using a formalism compatible with psycholinguistic theories?
• RQ2 Does the example-driven induction of transfer functions lead to linguistically plausible and accurate translations?
• RQ3 How does the performance of the self-organizing system compare to a Hidden Markov Model-based prototype in terms of translation accuracy?
• RQ4 To what extent can the system generalize to novel sentence structures not present in the training data?
• RQ5 Can the marker-normal form formalism adequately represent cross-lingual syntactic and semantic correspondences?

Key findings

• The system successfully induced English-to-French and English-to-Urdu transfer functions from small parallel corpora, demonstrating feasibility of the approach.
• The induced transfer functions were found to be linguistically well-grounded, reflecting plausible syntactic and semantic mappings.
• The system achieved a respectable level of translation success, comparable to a Hidden Markov Model-based prototype in quantitative evaluation.
• The use of marker-normal form enabled a more cognitively plausible representation of bilingual data than traditional formalisms.
• The optimization framework effectively balanced accuracy and generalization, allowing the system to translate novel sentences with reasonable fidelity.
• The results suggest that example-driven induction of transfer functions is a viable alternative to rule-based or statistical methods in low-resource settings.