R&D Focus Areas

Parsing

This is the first step in a Natural Language (i.e., English) Processing application. The input text is parsed sentence by sentence. Every word is identified as to its grammatical category (such as noun, verb, adjective, adverb, conjunction, preposition, etc.). Morphological information, such as the tense of a verb, or whether a noun is plural or singular, is extracted. The function of constituents is identified (subject, verb, object, complement, adverbial). A constituent of a sentence is a clause, the constituent of a clause is a phrase, and the constituent of a phrase is one or more words or collocations.

Our parsing technology goes beyond traditional parsing, by assigning meaning to sentence constituents.

Logic-based Representation of Language

Being able to represent a piece of text as a logic-form (i.e., using Ontological entities and relations) is the key to being able to understand language. Even simple things that humans take for granted, such as "if someone just arrived to a place, he or she is in that place", or "if A killed B, then B is dead." require 'understanding'.

Within certain limits, our technology is able to correctly answer questions on a factual English text.

Ontological Engineering

Ontologies are critical in Natural Language Processing since they encode concepts unambiguously. Words are inherently ambiguous, and highly context dependent. Consider the three senses of the verb 'order' in the following examples: 1) A commanding officer orders individuals under his command to carry out a specific action. 2) Somebody orders pizza. 3) A mailman orders letters by zip code.

An ontology first classifies things into categories, and then specifies how one thing relates to another through the use of logic predicates. Cyc, SUMO or Dolce are examples of ontologies.

Reasoning

To be able to answer questions on a piece of text (queries), the text and the question have first to be translated into logic form. A program called a reasoner may then 'reason' on both logical forms, and see if and how the logical form of the query can be satisfied by the logic form of the text.

Dialogue Management and Natural Language Generation

Natural Language Generation deals with converting a logic form back into human language.

Dialogue Management is concerned with keeping track of the user's goals, clarifying user intent, providing feedback to the user and structuring the output effectively.