The Routledge Handbook of Translation and Technology by Minako O'Hagan;

Author:Minako O'Hagan;
Language: eng
Format: epub
Publisher: Taylor & Francis (CAM)
Published: 2019-02-26T16:00:00+00:00

Machine interpreting (technology-generated interpreting)

Machine interpreting requires a combination of automatic speech recognition and machine translation systems with an optional speech synthesis or text-to-speech system for spoken target-language output (see Ciobanu and Secară in this volume). Technological solutions designed to automate interpreting can be traced back to the 1990s but a viable breakthrough has not yet been achieved.

Most currently available systems operate in a consecutive fashion whereby a speaker produces an utterance, the system processes and translates it, and delivers the translation either in written form (speech-to-text translation) or in spoken form (speech-to-speech translation). However, speech translation systems that simultaneously translate unsegmented, continuous speech have also been explored (Cho et al. 2013).

Early systems were restricted to experimental domains (e.g., ‘conference registration’), underpinned by initially hand-built lexicons. They focused on simple dialogues between two interlocutors in a limited number of language pairs (Waibel et al. 2017). More recent systems are corpus-based, i.e. have a larger vocabulary, and can cope with a greater variety of speech genres and languages. In the early 2000s, a focus emerged on domain-specific applications with real-life commercial, military, law-enforcement and humanitarian uses (Waibel et al. 2017: 15). These systems relied on a range of techniques to mitigate reliability problems (e.g., back-translation).

Further stages of development leading up to the present day are mobile solutions and general-purpose systems (notably Skype Translator, following Microsoft’s acquisition of Skype). Many of the available solutions now offer alternative output options, i.e. synthetic voice or text (in the form of speech bubbles, subtitles, heads-up displays in wearable technologies). However, whilst the research in this area has doubtlessly moved machine interpreting solutions forward, their application to situations in which highly accurate professional language mediation is required remains a non-trivial challenge. The rapid evolution of the neural paradigm, which is relevant for all major components of machine interpreting, may lead to sizeable progress. Especially neural networks which can learn from previous tasks and shift attention according to the relevance of an element in the source speech may have the potential to make machine interpreting more human-like (Waibel et al. 2017: 48).

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.