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

Not wanting to be a wet blanket on this party but this system has a looong way to go. Too many of the queries I made either gave me dumb error messages or decided to answer a different question.

Try and search for the shortest river. No dice. Given that it knows about second, third and the like I tried to rephrase this as "last longest river". All I got back was the longest river, it dropped 'last'. It knows average but not mean. It reports 'average length of a river' as 1117.2391304347826087. That's a great deal of precision for something without a unit of measure. What is this in: meters, kilometres, centimetres, feet, inches, miles? Given that the data is from America I would expect this to be in miles but I suspect that it is probably in kilometres.

Has this really taken 18k lines of code?

Stay put and knock this thing into shape on the universities dime then start your own company.



Hi Peter,

> this system has a looong way to go

yes it does... Natural language interfaces are a very hard area. Lots of failed attempts (e.g. MS EnglishQuery). This in fact is what attracted me to the area. If as a researcher I can help make NLIs more robust/usable, then that would be a 'result'.

> Has this really taken 18k lines of code?

Actually that is an approximate figure and includes test routines, comments, etc. Plus what is a line of lisp code? A large amount of code is to support non-linguist authoring of the interface. So in fact I was able to patch the system for "shortest river" and "mean" in a jiffy.

> Stay put and knock this thing into shape

Yes that's what I will probably do. I toned down Anaphoric's web-site and will work with this system from more of a research perspective. I will try my best to avoid open sourcing it, but in the end circumstances may compel that move...




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

Search: