Speech Technology Magazine


Video: Rules-Based Chatbot Design, Part 1: Syntactical and Semantical Analysis

Aspect's Andreas Volmer discusses key components of a rules-based approach to chatbot design based on a robust language model in Part 1 of this two-part series from SpeechTEK 2018.
By The Editors of Speech Technology - Posted Jul 6, 2018
Page1 of 1
Bookmark and Share

Learn more about chatbot design at the next SpeechTEK conference.

Read the complete transcript of this clip:

Andreas Volmer: What is the basis for this chatbot? One basis we already mentioned is the ontology: a robust language model with synonyms for every single word. If you train the system for one word, it automatically will pick up all the known synonyms for this word in this particular language. And you can train the system with a certain food item by saying, “I don't just want this food item, but these rules are matched with other generalizations of this.” If I work with a simple phrase that says "turkey,” but is really supposed to apply to all poultry, then this is very easy to do. It drastically reduces the number of training phrases or phrases to consider because of the inherent language understanding the system already comes equipped with.

Another really important element is syntactical analysis and semantical analysis. This example, "I want to cash my check," is obviously different intent from "I want to check my cash.” If you just had a pure keyword-spawning approach, it would be really hard to distinguish these, so what you really need to see are not only the recognition of these words, but also an understanding of what function they have in the sentence and what we call the governance relationship between them is. In the first place, "cash" is a verb and it governs the noun "check" and in the second phrase, "check" is a verb and governs the noun "cash.”

So you need to be able to distinguish those. Other examples: "Where can I check in online?" Another one: "I want a large pizza and medium Coke." Again, you need to have the governance relationships "large" is governed by "pizza," "medium" is governed by "Coke.” So these are just concepts that need to be available when creating these rules, these semantic and syntactical rules to extract information and also to identify intents.

Another thing that's almost a no-brainer these days so you need to have abilities to recognize typical data types. The ,ost difficult one, typically, is dates and times. "Let's meet tomorrow at 2:00 PM.” “Do we have an open slot on 3/7 at noon?” “I'll be available two hours from now." All of this needs to be easily picked up, identified as a date or time, and matched to some form of representation that the application then can work with.

Page1 of 1