Translation, information theory and cognition

Elke Teich,José Martínez Martínez,Alina Karakanta
DOI: https://doi.org/10.4324/9781315178127-24
2020-05-31
Abstract:The chapter sketches a formal basis for the probabilistic modelling of human translation on the basis of information theory. We provide a definition of Shannon information applied to linguistic communication and discuss its relevance for modelling translation. We further explain the concept of the noisy channel and provide the link to modelling human translational choice. We suggest that a number of translation-relevant variables, notably (dis)similarity between languages, level of expertise and translation mode (i.e., interpreting vs. translation), may be appropriately indexed by entropy, which in turn has been shown to indicate production effort. This chapter sketches a formal basis for the probabilistic modelling of human translation that may provide a common basis for product-oriented and process-oriented research on translation and makes communicative explanations of translational choice and its linguistic traces possible: shining through, normalization, simplification, etc. It provides a definition of Shannon information applied to linguistic communication—called surprisal in research on comprehension and information density in production research—and discusses its relevance for modelling translation. Information theory allows to define such probabilities in context on the basis of the amount of information that is conveyed by a given unit measured in bits. The perspective of predictability in context is extremely fruitful for the study of language use, variation and change. Casting translation in terms of rational communication with information theory as a formal basis allows to formalize some core notions of translation theory as well as provide the theoretical underpinnings for some less explained areas in Translation Studies.
What problem does this paper attempt to address?