Ideas: multilingual models, training, personality, GPT-2

rasa-nlu
(Vic) #1

Hi, i am new to Rasa, but not new to Ai dev. I have experience docs and want to ask several questions that are not clear from docs and blog posts i read recently

  1. Why dont you use Polyglot python library for multilingual support of NER, sentiment, etc ? polyglot · PyPI For me its a crucial part when i try to choose what framework to use. Spacy is good, but little lang supported. Is there a simple path to implement Polyglot?

    Tokenization (165 Languages) Language detection (196 Languages) Named Entity Recognition (40 Languages) Part of Speech Tagging (16 Languages) Sentiment Analysis (136 Languages) Word Embeddings (137 Languages) Morphological analysis (135 Languages) Transliteration (69 Languages)

  2. Can i use non-official language pack for Rasa, for example this for RU: GitHub - buriy/spacy-ru: Russian language models for spaCy ?

  3. Rule-based bot is bad approach, i agree with level 5 paradigm from blog, but cannot find extensive tutorial how do i implement pretrained models for my language as a helpers and then supply these with domain-trained model on LSTM from Rasa core Take a look GitHub - Desklop/Voice_ChatBot: Chatbot in russian with speech recognition using PocketSphinx and speech synthesis using RHVoice. The AttentionSeq2Seq model is used. Imlemented using Python3+TensorFlow+Keras. here several bin pretrained models, how do i integrate them to get Ai bot up and running, no teaching from zero. i need a bot that GENERATE answers, not stupid repeat from list of predefined

  4. Personality: how do i add personality to bot, to be able reproduce character and general knowledge and skills? I mean emotional bot, not business, example: GitHub - jddunn/emoter: Sentiment analysis library integratable into a chatbot that can generate text corpora from Facebook messages