A statistical model for lost language decipherment
ACL '10 Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics
Unsupervised multilingual learning
Unsupervised multilingual learning
| Hi-index | 0.00 |
Nearly all of the core data that computational linguists deal with is in the form of text, which is to say that it consists of language data written (usually) in the standard writing system for the language in question. Yet surprisingly little is generally understood about how writing systems work. This tutorial will be divided into three parts. In the first part we discuss the history of writing and introduce a wide variety of writing systems, explaining their structure and how they encode language. We end this section with a brief review of how some of the properties of writing systems are handled in modern encoding systems, such as Unicode, and some of the continued pitfalls that can occur despite the best intentions of standardization. The second section of the tutorial will focus on the problem of transcription between scripts (often termed "transliteration"), and how this problem--which is important both for machine translation and named entity recognition--has been addressed. The third section is more theoretical and, at the same time we hope, more fun. We will discuss the problem of decipherment and how computational methods might be brought to bear on the problem of unlocking the mysteries of as yet undeciphered ancient scripts. We start with a brief review of three famous cases of decipherment. We then discuss how techniques that have been used in speech recognition and machine translation might be applied to the problem of decipherment. We end with a survey of the as-yet undeciphered ancient scripts and give some sense of the prospects of deciphering them given currently available data.