Google announces the results of the study of the common language model, showing excellent semantic understanding.

CTOnews.com, March 7, according to foreign media reports, Google recently announced the results of the study of the common language model that was put into use in November last year. The results show that the Google language model can now recognize more than 100 languages. In some common language recognition, the semantic understanding of Google language model is even more eye-catching. Compared with the large-scale language model Whisper published by OpenAI, the error rate of semantic recognition is lower.

Pixabay according to Google official introduction, Google language model can carry out continuous self-learning, and can constantly modify the overall structure of the language model with the blessing of BEST-RQ algorithm, so as to complete continuous analysis, learning language structure and other operations. In addition, Google language model will also transfer text injection and supervised loss functions in semantic understanding, so as to make the semantic understanding of language model more accurate. At present, the error rate of Google language model can be kept below 30% when dealing with more than 70 language translations.

It is worth mentioning that the information obtained by CTOnews.com shows that the Google language model is more accurate than Whisper in the FLEURS tests of spoken English CORAAL, mixed accent SpeechStew and many other languages used by African-Americans.