Date: Mon, 24 May 2004 12:02:18 -0400
Subject: some query language cases
To: Steven Bird <sb@csse.unimelb.edu.au>
From: Beatrice Santorini <beatrice@babel.ling.upenn.edu>

here's are some examples of query language requirements that have come 
up in my work on the emodeng corpus so far.  i'll let you know of more 
as soon as i can; it may be till the end of the week till our 
workstations are set up.

A. add/delete brackets; i.e. go from the (a) cases to the (b) cases, or 
vice versa.  this presumably is easy to do, at least adding the np 
part, perhaps not the -sbj), but would be quite useful because it is 
needed a lot.

(1)  a.  (cp-frl what i want) is not so easy to get.
        b.  (np-sbj (cp-frl what i want)) is not so easy to get.

B.  adding/deleting elements in easily identifiable contexts.  in the 
diachronic english corpora, we analyze indirect and direct questions as 
in (2) and (3).  (2b) contains a null complementizer by analogy to 
(2a), which is grammatical in earlier forms of english.

(2)  a.  i don't know [cp-que [wadvp why] [c that] [ip-sub he did that 
] ]   overt c(omplentizer)
        b.  i don't know [cp-que [wadvp why] [c 0] [ip-sub he did that ] 
]       null c(omplentizer)

(3)       [cp-que [wadvp why] [ip-sub did he do that ] ]    never with 
overt c(omplentizer)

mistakes occur as in (4).

(4)  a.  i don't know [cp-que [wadvp why] [ip-sub he did that ] ]
                                                                     
^^^^^  add null c(omplentizer) as in (2b)
        b.   [cp-que [wadvp why] [c 0] [ip-sub did he do that ] ]
                                                   ^^^^^  delete null 
c(omplentizer) as in (3)

C. small clauses - vanilla.  some frameworks/corpora analyze a sentence 
like (5) as in (6a); others as in (6b).

(5)     we saw smarty jones win.

(6) a. we saw [np-obj smarty jones] [vb win] .		monoclausal analysis
       b. we saw [s [np-sbj smarty jones] [vb win] ] .	small clause 
analysis

it would be useful to go from (6a) to (6b), as well as the other way 
around.  of course, from (6b) to (6a) is more difficult.

D. small clauses - a bit more complex.  in the vanilla case, the small 
clause predicate is a verb.  in other cases, the predicate is a whole 
phrase - say, a prepositional phrase.  one might want to rebracket (7a) 
as (7b) (or vice versa).

(7) a.  we consider [np-obj smarty jones] [pp in the running].          
  monoclausal analysis
       b.  we consider [s [np-sbj smarty jones] [pp in the running] ].   
   small clause analysis

however, one would not usually want to rebracket (8a) as (8b).

(8) a.  we put [np-obj the cats [pp in the carrier].           
monoclausal analysis
       b.  we put [s [np-sbj the cats] [pp in the carrier] ].    small 
clause analysis - no!

so it would be nice to be able to distinguish cases like (7) and (8) on 
the basis of the matrix predicate.

in complicated cases like this one, it might be best not to replace one 
structure by the other, but merely to flag the cases for inspection.

E.  conjunction.  in the current emodeng corpus, np conjunction is 
generally analyzed as in
(9a).  to make the emodeng corpus compatible with the mideng corpus, 
such structures will have to be rebracketed as (9b).  of course, one 
might want to go from (b) to (a) as well.

(9) a.  (np (np the first conjunct)
                   (conjp (np second conjunct))
                   (conjp (conj and)
                                (np third conjunct)))

       a.  (np (np the first conjunct)
                   (conjp (nx second conjunct))		determiner-less np 
replaced by nx
                   (conjp (conj and)                                    
(nx = n' = intermediate between np and n)
                                (nx third conjunct)))

note that (9) must be distinguished from (10), where the non-first 
conjuncts are unequivocally full np's with determiners or possessives 
of their own.

(10)   (np (np my first conjunct)
                   (conjp (np your second conjunct))
                                      ^^^^
                   (conjp (conj and)
                                (np the third conjunct)))
                                       ^^^

(11) shows a mixed case.  again, one might want to go from (a) to (b) 
or vice versa.

(10)  a.  (np (np my first conjunct)
                      (conjp (conj and)
                                  (np (np your second conjunct))
                                          ^^^^^^^
                                          (conjp (conj and)
                                                      (np third 
conjunct)))))
                                                       ^^^^

        b.  (np (np my first conjunct)
                      (conjp (conj and)
                                  (np (np your second conjunct))
                                          ^^^^^^  np stays because of 
possessive
                                          (conjp (conj and)
                                                      (nx third 
conjunct)))))
                                                       ^^^^  np replaced 
by nx because no det/poss