Machine Translation and its Philosophical Accounts

Proceedings of the EACL 2009 Workshop on the Interaction between Linguistics and Computational Linguistics, page 1, Athens, Greece, 30 March, 2009. c⃝2009 Association for Computational Linguistics Machine Translation and its philosophical accounts Stelios Piperidis Institute for Language and Speech Processing, “Athena” Research Centre spip@ilsp.gr Abstract This paper attempts to explore the interrelation between philosophical accounts of language and respective technological developments in the field of human language technologies. In doing so, it focuses on the interaction between analytical philosophy and machine translation development, trying to draw the emerging methodological analogies. 1 Introduction Philosophical accounts of science and respective technological development bear a tight interrela- tion and continuous interplay. Likewise, philo- sophical investigations of language bear their own implications on how technology processing language, in monolingual or multilingual set- tings, evolves. In the multilingual setting, machine translation feasibility, its presuppositions and implications, brings forth a range of questions, applicable to human translation as well, including, but not lim- ited to, linguistic and ontological relativity, inde- terminacy of translation, inscrutability of refer- ence, representational function of language, the problem of meaning. Bar Hillel’s claims on the infeasibility of ma- chine translation and the Sapir-Whorf hypothesis with the linguistic determinism and relativity principles, partly backed up by Quine’s ontologi- cal relativity and later Wittgenstein’s private lan- guage and variability of language games have haunted the way of thinking in machine transla- tion development. Indeterminacy, relativity, and the consequent abolition of the one gold transla- tion idea, however, as well as the necessity for frameworks integrating pragmatic and be- havioural data in translation have played their role in advancing machine translation design paving the way for observed paradigm shifts at all stages of development. In broadly dividing machine translation history in the rule-based and corpus-based eras, in the 50’s and 80’s respectively, one can draw the analogies that would rather point to a tight inter- action between philosophical accounts and ma- chine translation paradigms. Early contemporary analytic philosophy, through conceptual, reference-bound analysis and com- positionality principles, provided the foundation for representational, rule and knowledge-based approaches of early machine translation, from 50’s through the 80’s. The turn, in analytic phi- losophy, to an understanding of meaning though use, to pragmatics and behaviourism, may be paired and seen as laying the foundation for the machine translation paradigm shift observed in the 80’s. In this pairing, it is the use of the much required parallel (or comparable) translation data that could be seen as constituting the behavioural data base, with each alignment function being conceived of as the result of a radical translation process, where a source language sentence pro- vides the sensory data and a target language sen- tence provides the linguistic observation. In such a framework, this aligned data source does pro- vide the “translation manual”, which after a se- ries of inductive operations does converge to a potentially usable set of translation relations. Along this line, we will discuss, in this talk, the continuous relations between analytic philoso- phy, linguistic science and human language …

Content truncated to 500 words