Loki is the acronym of Linguistic Oriented Keyword Interface. Unlike other Machine Learning based NLU engine, Loki is specially designed for Mandarin language. In linguistics, cooperative principles describe the features of humans’ conversations to make it efficient. Basing on these principles, Loki classifies NLU tasks into three levels namely “scenario”, “intent” and “syntax pattern.”
The transparency feature of Loki’s language model helps the developers see through the model with regular expression and the programming logic across every step in the algorithm. This obviously brings more convenience and less requirement of training data.
The history of Chatbot can be dated back long. However, many approaches of building chatbots that have been proven valid do not work when chatbots are facing Chinese language. Years after, “chatbot” has downgraded from “chatting” to “clicking” and is better described as “clickbot.” Some not-clicking bots are more like “keyword bot” and still not a “chatbot.”
As these technologies would not bring us a chatbot that can really “comprehend” the intent in a sentence and calculate a proper response, Droidtown Linguistic Tech. takes advantages of our Articut NLP AI system, and use the POS/NER taggings in the output to build a syntactic-pattern based NLU classification engine. It implements the three levels of conversation (i.e., “scenario”, “intent” and “syntax pattern.”)
We would continue releasing syntax patterns, intents and scenarios that can go with the intents to our Github repository under MIT license (https://github.com/Droidtown/ArticutAPI). Instruction videos are also leased to Youtube.
We try to introduce the four steps of “open source”, “sharing”, “feedback” and “re-develop” to bring out a real chatbot application that can show human-level intelligence.
About PeterWolf (Wen-jet Wang)
Founder of Droidtown Linguistic Tech. Co., Ltd.