Hi all, While reading through and trying to understand Ubiquity's English parser, I came across a semi-bug and wrote a patch. The patch does also disallow a certain type of legal input though, so I'd love to get some other's opinion. (I'd also appreciate if someone could take a look at the code diff for sanity and style.)
## The bug Right now when you have a command with a modifier (like translate) it'll let you put the direct object at the end of the command: "translate to Hebrew hello". However, if you have a word before to and also after the language argument, it'll try to put them together into a direct object. I.e., "translate hello to Hebrew goodbye" will give you the translation of "hello goodbye." Things get even weirder with multi-argument verbs... I just wrote a test verb "move" which takes a from argument (noun_arb_text) and at (noun_type_number). When you enter "move truck from tokyo at 1 pm," it'll try putting the "pm" at the end of Tokyo and at the end of truck, so it returns two possible options: "move truck pm from tokyo at 1" "move truck from tokyo pm at 1" neither of which really makes sense... the pm was obviously just a misunderstanding/misuse of that at argument. ## The patch This patch makes it so that any extra words after modifier-arguments which cannot be parsed as part of the arguments are dropped completely, e.g. "move truck from tokyo at 1 pm" will just give you the same result as "move truck from tokyo at 1". This personally makes sense to me and disallows for those weird multiple-candidate cases described above. ## Consequences of the patch The main consequence is that Ubiquity with this patch will quit parsing direct objects after modifier-arguments. In other words, "translate to Hebrew hello" will no longer be valid... you will have to use "translate hello to Hebrew". ## How the patch works (technical details) The patch basically works by changing the _recursiveParse function in ubiquity/modules/parser/locale_en.js so that it doesn't carry over the `unusedWords` from the end of the argument string in the `lastWord` for-loop. It will `return` the `completions` whenever a valid partial parse is produced, rather than waiting for all iterations of `lastWord`. This means it will do a "greedy matching" for arguments... basically the longest string between each of the modifiers which satisfies the noun type will be taken and shorter substrings will not even be considered. It also now returns only the `completions` that do not include any `_invalidArgs`. As a bonus, in some cases with verbs with multi-word modifier- arguments, this may in theory reduce the parsing time, though it didn't feel particularly faster on my machine compared to the older parser. Here's the diff: http://pastebin.mozilla.org/62711 Thanks! mitcho -- mitcho (Michael 芳貴 Erlewine) [email protected] http://mitcho.com/ linguist, coder, teacher --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "ubiquity-firefox" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/ubiquity-firefox?hl=en -~----------~----~----~----~------~----~------~--~---
