I’m a developer working for a little Start Up called Kevo Technologies S.L. located in the Basque Country, Spain. One of our main focuses is the language technology and we develop chat bots to attend at the phone or by text to automate some processes that keep some companies very busy.
Currently, we are using other companies’ solutions to develop our chat bots, so we are vendor locked. We liked a lot Botpress because it is promoted as a way of avoiding vendor lock-in and it is open source.
We are located in Spain so we want to use it in Spanish. We have made some tests but the results are so bad that we wanted to ask if the Spanish NLU is really this bad at the moment, or we are doing something wrong… I can only put one image so i will try to explain it well enough.
We have got a test chat bot using the Spanish NLU server locally. We have got only one intent called “que_tal_estas”. The intent is to respond to the question “¿Que tal estás?” which means “How are you?”. The first utterance is “Que tal estas” which works well when sending it to the bot using the emulator, and the confidence is 100%.
But, if we put an “Hola” in front of the exact utterance, the confidence becomes ridiculous:
We get only a 14% of confidence that the intent is “que_tal_estas” only by putting an “Hola”
in front of exactly the text of the first utterance, which would make this bot completely useless.
So, to summarize, we would like to know if the Spanish NLU is really this bad, or we are just doing something wrong, and, if we are doing something wrong, what would be the mistake.
Thanks in advance, and hope you can help us,