This is a prototype of language detection for short message service (twitter). with 99.1% accuracy for 17 languages

1. Extract model directory tar xf models/[select model archive]

2. Detect ldig.py -m [model directory] [text data file]

As input data, Each tweet is one line in text file as the below format.

[label]\t[some metadata separated '\t']\t[text without '\t']

[label] is a language name alike en, de, fr and so on. It is also optional as metadata. (ldig doesn't use metadata and label for detection, of course :D)

The output data of lidg is as the below.

[correct label]\t[detected label]\t[original metadata and text]

ldig has a estimation tool.

./server.py -m [model directory]

Open http://localhost:48000 and input target text into textarea. Then ldig outputs language probabilities and feature parameters in the text.

• cs Czech
• da Dannish
• de German
• en English
• es Spanish
• fi Finnish
• fr French
• id Indonesian
• it Italian
• nl Dutch
• no Norwegian
• pl Polish
• pt Portuguese
• ro Romanian
• sv Swedish
• tr Turkish
• vi Vietnamese

• Presentation in English

• Presentation in Japanese

• Blog Articles about ldig

  • Language Detection for twitter with 99.1% Accuracy
  • Precision and Recall of ldig (twitter language detection)
  • Estimation of ldig (twitter Language Detection) for LIGA dataset
  • Why is Norwegian and Danish identification difficult?

• (c)2011-2012 Nakatani Shuyo / Cybozu Labs Inc. All rights reserved.
• All codes and resources are available under the MIT License.