Hi Oliver,

I took the path to consider that some deep NLP is a must have when building a chatbot. GATE provides many out of the box features and plugins that I use in order to clean, optimize, tag the sentence given by a user. As example, here is a rough set of a pipeline of plugins processing the line of a user:

- Language identification module
- Tokenizer
- Stemmer
- Stanford Parser
- Part of Speech tagger
- Wordnet custom plugin (for each token, it searches in Wordnet its synonyms. Because maybe the original token is unknown, but one the synonyms and therefore can be used to know what the user is talking about)
- Gazetteer (Identifying Name Entities using the JAPE transducer. Meaning patterns recognition and tagging of tokens, like “Influenza->Virus”)
- Some chunker
- Pronominal Coreferencer
- Custom Plugin adding annotations about the relations between tokens (like extraction of triplets subject-predicate-object, or tokens with possessives relationship)
- Also a custom plugin using OpenEphyra when the user entered a question (my goal here is to use the questions-answers patterns capabilities of OpenEphyra)

Once this pipeline is done, I have many annotations full of information that I can use.

My current step (learning process) is to recognize those triplets and to store them into an ontology if they were not already there.

Once I will be happy with the learning capabilities, I will start looking at identifying what is the most relevant information to be returned, eventually using AIML as back bone for the static part of the speech…