In common use today, we have two different temperature measurements. Degrees Celsius, and Fahrenheit.

In cooking books is usual to have all temperatures specified like: "180C (356F)" or "365F (180C)", depending on where they are published.

Very few of us can readily think in both. So questions/answers tend to use one or the other.

I feel like it wouldn't be too hard to perform a little (extremely basic) natural language processing, and automatically recognise when someone is talking about a temperature in one unit, and then automatically convert it to the other. Placing the results in brackets.

The Community User is the stack exchange bot that does several useful things, including editing posts in very limited ways (it does for example put in a "possible duplicate of" link). A bot could be set up do this task -- which would mean it doesn't break the normal editing process and would show in the edit history. It could be smart enough to not re-edit after its edit has been removed. (It could also be done without using a web crawler bot).

I feel like this could be done, with a very low rate of false positives. I think this would provide much utility. Very few other sites on the SE network care about temperature as much as we do.

  • I think this is a very old feature request already :-) Feb 7, 2015 at 15:31
  • or of this here Inline Unit Conversion? Feb 7, 2015 at 16:17
  • 1
    NLP is an extremely difficult problem that nobody has really solved yet, so... probably not going to happen.
    – Aaronut
    Feb 7, 2015 at 19:02
  • 1
    @Aaronut This particular problem is probably within the grasp of current technology if the ambition is lowered to producing a suggested edit. Recognizing temperatures is easy enough with a few regular expressions if the unit is supplied. Certainly there will be false positives("2C" may refer to "2 cups" but could equally refer to refrigerator temperatures) but a human reviewer of the suggested edit ought not let those pass. I suggest that only temperatures given without the unit of measure require lots of context and non-trivial NLP. Feb 7, 2015 at 19:47
  • @ChrisSteinbach: I interpreted the request as detecting when people (usually Americans) don't specify the unit at all, which happens quite frequently. Sure, if they specify C or F then it's an easier problem to solve, but you don't really need NLP for that, just a few regexes and sanity checks. This problem is easy for the community to solve - my script's been around forever, and can definitely be improved upon. I honestly wouldn't want to waste the dev team's time on such a trivial issue, unless they plan to solve the problem for all UOMs on all sites.
    – Aaronut
    Feb 7, 2015 at 19:54
  • Aaronut, I mostly mean catching things like 320F and 200 degrees Celsius. Get the easy ones, so people don't have to and can focus your hard ones. Regex+sanity checks is (very simple) NLP. Can you link your script? Feb 7, 2015 at 23:21
  • found Aaronut's script meta.cooking.stackexchange.com/a/1395/21622 and am now using it. Feb 7, 2015 at 23:29


Browse other questions tagged .