MST HPSG Parsing Exercise Results



Score (max 5)

349481, 3382009, 331917

You wrote that "every object needs to have a case"[sic!] as a constraint. Actually, this holds true for every object of type "noun", but not for every object. Some structures were are not complete (i.e. missing).


358083, 383702, K-9881

The feature structures are fine, except that there are several SYNSEM errors (though no point cuts were made for this, as it was optional, and I do appreciate the effort): the content of "cats" is "loved" not "loving"; the shared synsem feature of lamp and green can be something like "color" (I know we did this using the lexicalized approach, but still "greenable" is a bit of a contrivance, to me at least); the content of "sing" is "singer" and "sang" (You seem to have parsed this as "Sing!" and not "Sing."), as in "Mr. Belvedere [1singer] sang [1singer, 2sang] a song [2sang]" (he might sing it to someone, but that is non a minimum requirement for the AS of the verb to be realized, so we would relegate it to CONTEXT as a pragmatic thing). Finally, the sheets [that] you have submitted this on are tiny and made me feel like a giant attempting to use a late 90's Nokia mobile.


389393, 378350, 383601

I could not find task one on any of the sheets, so we need to cut one point for that... Lamentable, for the synsems were a valiant attempt, evidential of a desirable predilection with the wonders of syntax. And yet, this predilection notwithstanding, they feature a small measure, a smidgen, or air, if you will, of inconsistency with these signature. Notably, the "LOCAL" component is omitted. So, basically, for syntax the path should be: SYNSEM|LOCAL|CAT|SUBCAT<(E/NE)list>. For semantics it should be: SYNSEM|LOCAL|CONTENT. Furthermore, the lexicalized semantic content of "walk" is not "walk" but "walker" (the location is not compulsory and may therefore be relegated to CONTEXT if a sentence would specify such, but it does not). But, again, this was optional, so no points were cut here. Moving on, there an ever so slight problem with  task 2, wherein you (Ye? Y'all?) stated that "[pronunciation/case/semantic role] should be given"[sic!]. This is a problem because in any CBG, everything indicated by constraints MUST be given. This answer was also overly specific. The preferred answer would be simply to state that "this slot requires a non-empty list", etc.


370226, 377941, 326075

Your work has not been anonymized. This is a no-no. Opted for no SYNSEM. That is your prerogative. 2. is also a bit too specific, but correct. Again, I would prefer something mentioning a "non-empty list". The only nitpick that reflects on assessment pointwise is that you used HEAD NOUN as an object type. As our signature specifies, the HEAD should be a feature of which things like noun, pronoun, verb, etc. are types (list objects). I.e. "CAT|HEAD|noun" and not "CAT|head noun". So, I will nick half a point for that.