Microsoft is globally rolling out a large scale multilingual spelling correction model, called Speller 100 with high accuracy and high recall in more than 100 languages. According to Bing around 15% of queries submitted by users have wrong spellings, which can lead to incorrect and unsatisfactory answers. To solve this problem, Bing has created its most comprehensive spelling correction system ever.
Bing observed the following results in A/B testing queries with and without Speller100:
- The count of pages with no results decreased by up to 30%
- The number of times users had to manually modify their query decreased by 5%
- The number of time users clicked on spelling suggestion increased to 67%
- The number of times users clicked on any result on the page went to 70%.
So, how they made this possible, continue reading to know more about Speller100.
Spelling correction has been a priority for Bing for long, and the search engine is taking it to another level by including more languages from across the globe.
According to Bing, “In order to make Bing more inclusive, we set out to expand our current spelling correction service to 100-plus languages, setting the same high bar for quality that we set for the original two dozen languages.”
Speller100’s launch represents an important step forward for Bing and it became possible due to latest advancements in AI.
Speller100 technology of Microsoft Bing
According to Bing zero-shot learning is a significant advancement in AI which assisted in making Speller100 a reality.
Zero-shot learning allows an AI model to learn accurately and correct spelling without any additional language-specific labeled training data. Unlike the traditional spelling correction solutions which rely completely on training data to learn spellings of a language. This becomes really challenging when there is insufficient amount of data. Zero-shot addresses this problem very efficiently.
Bing said, “Imagine someone had taught you how to spell in English and you automatically learned to also spell in German, Dutch, Afrikaans, Scots, and Luxembourgish. That is what zero-shot learning enables, and it is a key component in Speller100 that allows us to expand to languages with very little to no data.”