PEOPLE who learn English as a second language tend to write with a 鈥渇oreign accent鈥�. Now software can examine a piece of written English and make a pretty good guess at the writer鈥檚 mother tongue.
The system works by looking for grammatical quirks that are characteristic of different languages. For example, people who speak Czech tend to omit the word 鈥渢he鈥�, while infrequent use of infinitive verbs, such as 鈥渢o go鈥� or 鈥渢o do鈥� is indicative of Spanish speakers. French people often create non-standard adverbs by adding the suffix 鈥�-ly鈥� to various words.
Moshe Koppel, Jonathan Schler and Kfir Zigdon at Bar-Ilan University in Israel took examples of writing by more than 1500 proficient but non-native English speakers, each about 700 words long. A grammar checker looked for idiosyncratic use of language, such as particular pairings of adjectives and nouns. Software then classified each of the documents into five native languages, according to similar use of these idiosyncrasies. The researchers found that they could identify the language of the writer with about 73 per cent accuracy.
Advertisement
鈥淭hat鈥檚 better than I鈥檇 expect,鈥� says Geoffrey Sampson an expert in natural language computing at the University of Sussex, UK.
The software is part of a general program aimed at profiling writers according to their gender, age and background.
