Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

The real question I have is according to that episode Watson does not take into account the category only the 'answer' and the already shown questions. It seems _really_ odd that they would ignore this bit of information.


Since there are a lot of pun-based categories on Jeopardy, I can understand why they would have problems with it.


They could at least factor it in selectively, in obvious situations. For example, all categories in jeopardy that contain a quoted string include that string in all of the answers. Wouldn't be too hard to use that information.

edit: furthermore, sometimes it's impossible to narrow down the answer if you don't understand the category. For example, if the category is '"C" you later' or something, there might be like 8 possible synonyms that fit, but only one starts with a "C".


Exactly - the category label probably is very little help in narrowing down possible answers to a text-processing algorithm.


This is a significant disadvantage for some categories. For example one category from Friday was "WARE"-ING with each question containing "ware" in the response.

After typing out this comment it occurs to me that if you carefully consider each answer there is only only logical question for all the answers in this category. Although I don't have any specific examples, there are times when a question is ruled incorrect because of the category name (e.g The category called for a specific number of letters in the response).


I would have to disagree. Every bit of information helps. I was even surprised that they didn't implement a voice to text system so that he could learn what the other contestable guessed incorrectly and utilize that information if only to not guess the same thing on his turn.


They have added this. This was actually a problem in earlier versions where it would sometimes guess the exact same wrong answer as a previous contestant.


<spoiler>This happened on the first episode tonight... Ken Jennings guessed "1920s" and Watson guessed the same thing</spoiler>

Either they ended up not implementing this, or Watson didn't hear him correctly :)


The problem could have been that Jennings said "the twenties" which a computer may not have recognized as the same thing as "the nineteen twenties"


IIRC, they said they fixed that.


They said in the end of the episode that they started feeding him the correct questions once they are known (Thus in the example learning that all of the questions are names of months), but they didn't mention the categories that I saw.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: