Who is working with NLP in the Spanish language?

Who is working with NLP in the Spanish language?(neuraan.com)

3 points by IsraCV 3 years ago | 3 comments

IsraCV 3 years ago |

We’re Israel & Mario, cofounders of Neuraan (https://neuraan.com). It is difficult for natural language processing developers in LatAm to get their applications to have the same accuracy in Spanish as in English. We create an API that helps them correct those inaccuracies and increase the adoption of their solution.

We have +6 years of experience developing chatbots and voice assistants for large enterprises across LatAm and was very common to end up doing some research and developing algorithms to enhance the accuracy that the commercial solutions like Dialogflow, Watson, Amazon Lex, and Azure Cognitive deliver because they provide a general solution for non-English languages.

I'll like you to try or invite someone to try our API and share your feedback with us.

gus_massa 3 years ago |

You should have linked the English landing page https://neuraan.com/en

Do you have a video in English about your project? It would be nice at least to have subtitles. [Anyway, hi from Argentina!]

The example says:

> "input": {"text": "Como has estado"}

It is missing a tilde. it should be

> "input": {"text": "Cómo has estado"}

Is that on purpose? Can you handle accented characters? laitn1 vs uf8? What about this kind of problems when someone forget the tilde?

Maxsob86 3 years ago | |

Hi, thanks for the feedback. Yes, the API is able to handle accents and punctuation in utf-8. I have not made proofs using latin1 encoding but I will check how it's handling it (thanks for the hint). The example was taken from a knowledge base (utterances and examples to tell our AI what to recognize things inside the chatbot) of one of the users. They feed the AI algorithm with all sorts of examples, some with good punctuation and orthography but the majority have spelling errors, which are also very common in the users' interactions with the bot, and we get good recognition results with and without the tilde. We are also handling spelling errors and enhancing this since it seems (surprisingly) to be a big issue for a lot of standard NLP engines. Regards