Talk:Open domain question answering

[Untitled]

This would perhaps be better as a section under question answering. Anyone agree/disagree? pgr94 (talk) 17:37, 21 June 2010 (UTC)

Agree very much. Moreover, this article fails to make clear what exactly is the "open-domain" part of ODQA (invented acronym, for brevity) and how ODQA differs from regular QA. If the distinction is not made clear, I would say this should not even be a section, but rather be fully merged into the QA article. --Waldir ^talk 21:44, 21 June 2010 (UTC)

I disagree. Open domain question answering is such an interesting subfield that it deserves it's own article. You may think it is too short, and at this point it is probably too short to warrent its own article, but it will probably become substantially larger. At least, this is what I hope. It would be really nice to see detailed descriptions of different algorithms in pseudo code. I'd also like to see links to actual open source implementations.174.63.50.121 (talk) 10:51, 26 February 2013 (UTC)

Besides needing to be a section of question answering, is this too specific about how it works?

Agree that maybe this could be merged, but also seems like we're taking the outline of one ODQA system (maybe most existing ones) and saying that's what ODQA is. Seems like ODQA is really just defined by the goal, and anything about specifics has to clearly be saying "this is what some of the problems are" or "this is how some QA systems work," ideally with reference to specific systems.

So, broadly: question comprehension is a piece of it; searching is a piece of it; transforming search results into answers is a piece of it. For all of those stages, there are a bunch of techniques: question comprehension might try and make a formal diagram of the sentence or use statistical methods or just extract key words or some of each; search might look in structured/unstructured data, or use synonyms or such; phrasing answers involves some playing with the grammar of results and referring back to how the question was phrased. And most importantly we need citations for all of this in the plentiful QA literature so it's not just my or somebody's idea of how a QA system should work.

(Updated to add: ...and we may want to work a little harder to avoid jargon. Seems like the sort of toplevel introductory algorithm a total layperson would read, as opposed to something more specific about an algorithm, technique, etc.)

Obviously I got brought here by reading about IBM Watson, so maybe a reference to that would be nice.173.2.143.48 (talk) 16:53, 17 January 2011 (UTC)

I don't understand this sentence

"A score is then given to each of these candidates according to the number of question words it contains and how close these words are to the candidate, the more and the closer the better"

The part I don't understand is "and how close these words are to the candidate". This means "and how close the question words in the candidate are to the candidate" which I don't understand.174.63.50.121 (talk) 10:59, 26 February 2013 (UTC)