Language detection accuracy & requirements

[Atul Vaidya]

I tried the Dandelion language detection API with a few short snippets of text in English.  While it managed to report back with a high degree of confidence for English for most snippets I found that it failed with simple text.  For instance

• https://api.dandelion.eu/datatxt/li/v1?token=3....1&text=I+am+here came back with a confidence of 0.055 for English and 0.56 for Danish
  

This is an extreme example and my app has other, pre Dandelion, safeguards that will stop this from happening.  However, I am wondering...  are there any general guidelines on the amount of text to submit for reliable language detection.  Also, is it advisable to have one or more words with more than three letters to improve the chances of correct detction?

[ber...@dandelion.eu]

Hi Atul,

no, there are no general guidelines except for the size of the text.

Longer is the text more accurate is the language detection.

regards

Giacomo Berardi

Dandelion team