What AI techniques does IBM use for its Watson platform (specifically, its natural language processing part)?
$\begingroup$
$\endgroup$
0
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
0
As a person who works with people who work on Watson, perhaps I can give some insight.
The name Watson is casually thrown around a lot whilst many people aren't aware of its evolution into a larger suite of systems and services. We now have Chef Watson, Watson Health, and many other developing projects along the "cognitive" route. Watson is really an amalgamation and varied application of the different cognitive computing routes IBM is pursuing.
So, there are many forms of NLP that Watson conducts and has conducted, developed by different teams to fit different processes, interconnected in different ways. Additionally, much (probably all) of it is proprietary/classified since, as one would imagine, ongoing research is constantly being conducted and added to Watson. This is likely your largest obstacle. The precise workings of the NLP of the Jeopardy flavor of Watson are probably themselves still classified (I can't find anything in the time I've just spent looking myself)
There are, thus, many answers to this question; many outdated, and others not always applicable. The full answer is very complicated and, by the time you find out what the answer is today, it's probably already been advanced. The researches I know are always working on new, cutting-edge algorithms and processes for text classification and the related NLP topics. See, for example, this article http://www.aaai.org/Magazine/Watson/watson.php, where it is stated
For the Jeopardy Challenge, we use more than 100 different techniques for analyzing natural language, identifying sources, finding and generating hypotheses, finding and scoring evidence, and merging and ranking hypotheses.
and
Watson's metalearner uses multiple trained models to handle different question classes as, for instance, certain scores that may be crucial to identifying the correct answer for a factoid question may not be as useful on puzzle questions
Most of these techniques have likely already been modified, adapted, or dropped altogether for other methods.
To point you to more information, take a look at the links below.
General
The DeepQA Research Team, specifically, have a look at their publications
Articles
- Building Watson: An Overview of the DeepQA Project (2010) written by the IBM Watson Research Team, led by David Ferucci, and published in AI Magazine Fall, 2010.
Videos
- How does IBM Watson work? (2018), by IBM Watson, which states that Watson uses deep learning and transfer learning
Other
- Natural language processing: an introduction (2011), a good but technical paper on natural language understanding (that acknowledges Watson)
