Skip to content

Natural Language Processing


Natural language processing covers a set of techniques that allows the computer to understand a voice, a text, and to act on it.

Some examples to get an idea:

  • Identify the overall sentiment of user feedback, based on time and products.
  • Annotate, or classify, documents automatically by extracting metadata from the text.
  • Identify a user’s request and choose the answer to it.
  • Write down an oral request and act on it.
  • Answer questions by extracting related information from a large body of material.
  • Write summaries of documents… has experience in developing models based on several technologies. From light models, in which each word is simply indexed by its position in a dictionary, to the most advanced models, such as Transformers, which allows to transform whole sentences into numerical vectors synthesizing the information they contain and integrating the contextual information of each word.

To speed up the training of these models, we start with networks pre-trained on huge text corpora. Like the BERT network, trained by Google, which has “read” all of Wikipedia and several book libraries. To this training, we add a second domain-specific run. For example, reading thousands of technical documents related to the domain concerned.

Leave a Reply