Monday, February 13, 2012

FootySpeak: Natural Language Generation for AFL Reports in Aboriginal Languages and English

Greetings from the School of Language Studies at the Australian National University in Canberra, where Mark Dras, Department of Computing, Macquarie University is speaking on "Arrernte Footy: A Computational Grammar and Natural Language Generation System for Arrernte". The idea is to generate natural language new articles in English and Arrernte automatically from the data about an AFL football match. This is of interest to researchers, as well as having commercial application in financial and sports reporting, where there is a regular structure to what is reported and lots of statistics. This is also of interest for teaching languages and making languages more understandable. It occurs to me that the English language description of an AFL match is not really in English, it is in "FootySpeak", a subset of English (like Seaspeak). The same approach might be used to generate weather reports and emergency warnings, in a readable, but very precise language. The problem is that messages have to be created quickly in high stress situations.

A content selector, planner, deep realizer, dictionary complier, XLE and Weighted Finite-State Transducers (WFSTs) are used in turn to create the natural language sentences. A lexical functional grammar is used to describe the structure of sentences in the language. For Arrernte some featured not required in English are used, such as for reduplication. Also Glue semantics could be used, if the software would support it.

Some of the research was carried out at the Ngurratjuta Lightning Carnival (Lightning football is a shortened version). One use for the automatically generated text would be for teaching reading in schools, as football will be of many interest to many students than other topics.

Mark needs some recording or text of descriptions of football matches in Arrernte, so if you have any contact him.

ps: It would be interesting to generate Roy and HG commentary.

No comments: