Understanding "[drop, put, swipe] X on the floor" and the final scopal argument for the verb

Based on probably too little data, I thought that I had cracked the code on verbs like “drop” and “put” that can take a prepositional phrase and interpret it in a few different ways. When I encountered “swipe” it didn’t work the same. So I’m wondering if I misunderstood.

For example, below are 3 readings I get from ACE in the 2020 grammar for “Drop the bag on the floor” and what I think are the proper interpretations of them (I get similiar readings for “put the bag on the floor”):

1. Drop the (bag on the floor) -> Drop the bag that is on the floor
    drop_v_cause__exx where ARG2 is set to the bag that is on the floor
    _on_p_loc__exx specifies the "bag on the floor" using ARG1 and ARG2

2. Drop the bag (on the floor) -> Drop the bag while you are on the floor
    drop_v_cause__exx  where ARG2 is set to the bag
    _on_p_state__eex where ARG2 is the floor and ARG1 is the drop event

3. Drop the bag on the floor -> Drop the bag and have it end up on the floor
    _drop_v_cause__exxh where ARG2 is set to the bag and ARG3 is a 
        handle to the "thing we want to be true at the end but is not 
        necessarily true now" namely: the bag being on the floor
    _on_p_loc__exx specifies the "bag on the floor" using ARG1 and ARG2 
         but unlike the first case above, this time it represents a future state 
         that results from the verb action and we can tell this because the 
         verb is referencing this predicate using a scopal argument

The problem is when I use ACE to parse “swipe the card on the reader”, I get the first two types of readings, but not the third.

I’m looking for the reading that means “swipe the card by pulling it through the reader”. The “card being on the reader” is a state that does not exist yet and so I imagine this is case 3 above. Thus, I expected there to be a swipe_v_1__exxh where ARG3 was a scopal argument pointing to _on_p_loc__exx. But this isn’t one of the readings.

Am I just confused about what this scopal argument on the verb is doing?

Any tips on how to think about this?

@Dan any tips on how to think about this?

You have the right model for the three readings with “drop”, where the third one with the PP as a second complement of “drop” is intended to describe the resulting state of affairs affecting the entity denoted by the NP direct object (the bag). Your example with “swipe” doesn’t work like “drop”, since the PP doesn’t describe a resulting state for “the card”. Your paraphrase of the desired reading has a by-PP modifying “swipe” that overtly gives you the notion of “by means of”, but when the PP is a simple locative phrase such as “on the reader” some level of inference is needed to supply that “by means of” part of the intended content of the sentence. It seems clear that it is the “swiping” that happens on the reader (necessarily involving “the card”), so it still seems reasonable to expect the second reading you get for the sentence to suffice for the intended reading, even though some additional inference is needed. In another paraphrase “swipe the card using the reader” the -ing VP more clearly modifies the “swipe” VP, again suggesting the necessary content to be inferred when providing just the locative phrase “on the reader”. Consider also “open the door with a key”, where “with a key”, a modifier of the VP “open the door”, provides a more overt sense of the instrumentality that has to be inferred for swiping “on the reader”.

Thanks @Dan for clarifying. So does this mean that the MRS for these two meanings of “swipe the card on the reader” is the same (shown below)?:

“swipe the card while you, the person doing the swiping, is on the reader”
“swipe the card, and do the swiping on the reader”

It seems intuitively like the first would conceptually have “_on_p_state” applied to the ARG1 of the verb (somehow) while the second would apply it to the verb itself. Is this just the way the _state pattern works: it could always mean either?

(I believe I saw the same pattern with “knock on the door” where the _on_p_state is applied to the verb and meant “the knocking should happen on the door” or “the knock should happen while the person is on the door”.)

[ TOP: h0
INDEX: e2
RELS: < 
[ _the_q LBL: h15 ARG0: x14 [ x PERS: 3 NUM: sg IND: + ] RSTR: h16 BODY: h17 ]
[ _reader_n_of LBL: h18 ARG0: x14 [ x PERS: 3 NUM: sg IND: + ] ARG1: i19 ]
[ pronoun_q LBL: h4 ARG0: x3 [ x PERS: 2 PT: zero ] RSTR: h5 BODY: h6 ]
[ pron LBL: h7 ARG0: x3 [ x PERS: 2 PT: zero ] ]
[ _the_q LBL: h9 ARG0: x8 [ x PERS: 3 NUM: sg ] RSTR: h10 BODY: h11 ]
[ _cardkey/nn_u_unknown LBL: h12 ARG0: x8 [ x PERS: 3 NUM: sg ] ]
[ _on_p_state LBL: h1 ARG0: e13 [ e SF: prop TENSE: untensed MOOD: indicative PROG: - PERF: - ] ARG1: e2 ARG2: x14 ]
[ _swipe_v_1 LBL: h1 ARG0: e2 [ e SF: comm TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: x3 ARG2: x8 ]
>
HCONS: < h0 qeq h1 h5 qeq h7 h10 qeq h12 h16 qeq h18 > ]

Yes, the grammar does not distinguish between a PP logically modifying the verb’s meaning or the denotation of the subject of the verb. This is due to the general conservative principle of only asking the syntax to produce distinct analyses if there is some chance that the grammar could help to disambiguate. For many, maybe most situations, we can’t even practically distinguish between verb or subject-of-verb modifier target: in “she sang on the stage” both “she” and the “singing” happened on the stage. As long as there is not some syntactic context where we could tease those two interpretations apart, there is no benefit (and some considerable cost) in doubling the number of parses each time we have a PP modifying a VP. We might still wish for an annotation/treebanking tool that would allow a human annotator to explicitly mark the distinction that you rightly observe for your “swiping” example, but akin to most word-sense ambiguity, this should be made to work from an underspecified syntactic analysis.

Got it. I understand the conservative approach, it seems like is a clear principle to follow and understand.

For my implementation I’ll just end up doing the combinatorics in the implementation of predications like _on_p_state since (like the ERG) I won’t know which is meant. So I have to try them and see which one works.