"Probably, the cat sleeps" vs. "The cat probably sleeps"?

grammars ERG
1

I was messing with trying to get scopal modifiers to work, and I found that I can either have an MRS where the INDEX comes from probable.ARG0, or an MRS where the INDEX comes from sleep.ARG0.

I get different generation results each option, and I’m confused about why both of these constructions are permitted and what the intention behind each version is.

INDEX = sleep.ARG0

[ TOP: h15
  INDEX: e11 [ e TENSE: pres ]
  RELS: < [ _probable_a_1 LBL: h10 ARG0: i8 ARG1: h14 ]
          [ _sleep_v_1 LBL: h13 ARG0: e11 ARG1: x1 ]
          [ _the_q LBL: h4 ARG0: x1 RSTR: h2 BODY: h3 ]
          [ _cat_n_1 LBL: h7 ARG0: x1 ] >
  HCONS: < h14 qeq h13 h2 qeq h7 h15 qeq h10 > ]

Probably the cat sleeps.
Probably the cats sleep.
Probably, the cat sleeps.
Probably, the cats sleep.

INDEX = probable.ARG0

[ TOP: h18
  INDEX: i8
  RELS: < [ _probable_a_1 LBL: h10 ARG0: i8 ARG1: h17 ]
          [ _sleep_v_1 LBL: h13 ARG0: e11 [ e TENSE: pres ] ARG1: x1 ]
          [ _the_q LBL: h4 ARG0: x1 RSTR: h2 BODY: h3 ]
          [ _cat_n_1 LBL: h7 ARG0: x1 ] >
  HCONS: < h2 qeq h7 h17 qeq h13 h18 qeq h10 > ]

The cat probably sleeps?
The cat probably sleeps.
The cats probably sleep?
The cats probably sleep.
2

I have also found the same thing to be true with negation:

INDEX = sleeps.ARG0

[ TOP: h17
  INDEX: e8 [ e TENSE: pres ]
  RELS: < [ neg LBL: h13 ARG0: e11 [ e TENSE: tensed ] ARG1: h12 ]
          [ _sleep_v_1 LBL: h10 ARG0: e8 ARG1: x1 ]
          [ _the_q LBL: h4 ARG0: x1 RSTR: h2 BODY: h3 ]
          [ _cat_n_1 LBL: h7 ARG0: x1 ] >
  HCONS: < h12 qeq h10 h2 qeq h7 h17 qeq h13 > ]

The cats do not sleep.
The cat does not sleep.

INDEX = neg.ARG0

[ TOP: h16
  INDEX: e11 [ e TENSE: tensed ]
  RELS: < [ neg LBL: h13 ARG0: e11 ARG1: h12 ]
          [ _sleep_v_1 LBL: h10 ARG0: e8 [ e TENSE: pres ] ARG1: x1 ]
          [ _the_q LBL: h4 ARG0: x1 RSTR: h2 BODY: h3 ]
          [ _cat_n_1 LBL: h7 ARG0: x1 ] >
  HCONS: < h12 qeq h10 h2 qeq h7 h16 qeq h13 > ]

The cat does not sleep.
The cats do not sleep.
The cats don't sleep.
The cat doesn't sleep.
3

Thanks for what turns out to be a welcome bug report. Back in 2020 at the DELPH-IN summit, it was noted that in order to get the semantics right for conjoined VPs such as “Kim arrived and didn’t leave” (where the conjunction relation takes as its two arguments the ARG0s of the two conjuncts), the INDEX of “didn’t leave” needs to be the ARG0 of the neg_rel, not of the “leave” relation. This improvement was, sadly, not carried through consistently, neglecting negated modals such as “shouldn’t”, the lexical rule that adds “not” as a complement to get “Kim [should not] leave”, and sentence-initial scopal adverbs. These shortcomings have now been corrected and will be checked in for the next grammar release later this year, so now for both negation and scopal adverbs, the only MRSs that will successfully generate are the ones where the INDEX of the overall MRS is identified with the ARG0 of the negation or adverb relation.