"Is the light on/off?" Which parse is right?

Parsing the sentence “Is the light on?” (same thing happens with off) with ERG erg-2018-osx-0.9.31, I get the below options. Assuming I want the meaning to be “is the light illuminated?” Which is the right parse?

It seems like #1 can be discarded since “‘on’, as in illuminated” is not a location so _on_p_loc is not right…

It seems like #3 can be eliminated because nothing has been elided from the phrase so ellipsis_ref is not right.

So that leaves #2, my intuitive answer is that parse #2 has a stative preposition on which is modifying the verb be and this means “to be in the on state”, but I’m completely speculating here.

Any help is appreciated!

#1
[ _the_q LBL: h4 ARG0: x3 [ x PERS: 3 NUM: sg ] RSTR: h5 BODY: h6 ]
[ _light_n_1 LBL: h7 ARG0: x3 [ x PERS: 3 NUM: sg ] ]
[ _on_p_loc LBL: h1 ARG0: e2 [ e SF: ques TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: x3 ]
#2
[ _the_q LBL: h5 ARG0: x4 [ x PERS: 3 NUM: sg ] RSTR: h6 BODY: h7 ]
[ _light_n_1 LBL: h8 ARG0: x4 [ x PERS: 3 NUM: sg ] ]
[ _on_p_state LBL: h1 ARG0: e9 [ e SF: prop TENSE: untensed MOOD: indicative PROG: - PERF: - ] ARG1: e2 ]
[ _be_v_id LBL: h1 ARG0: e2 [ e SF: ques TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: i3 ARG2: x4 ]
#3
[ _the_q LBL: h4 ARG0: x3 [ x PERS: 3 NUM: sg ] RSTR: h5 BODY: h6 ]
[ _light_n_1 LBL: h7 ARG0: x3 [ x PERS: 3 NUM: sg ] ]
[ _on_p_state LBL: h1 ARG0: e8 [ e SF: prop TENSE: untensed MOOD: indicative PROG: - PERF: - ] ARG1: e2 ]
[ ellipsis_ref LBL: h1 ARG0: e2 [ e SF: ques TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: x3 ]

Yes reading #2 seems the appropriated one. But paraphrasing the semantics is not easy. @guyemerson is good on that!

_be_v_id is for a copula connecting noun phrases, e.g. the sentence “Is the light an electric lamp or a gas lamp?” would need _be_v_id. In this case, _be_v_id introduces an event for the clause. On the other hand, adjectives and prepositions already introduce events, so it’s not necessary to introduce another predicate. So for the sentence “Is the light on?”, I would expect an analysis like in #1.

As for the different predicates associated with “on”, be aware that they aren’t mutually exclusive, but rather are arranged in a hierarchy (see: http://svn.emmtee.net/erg/tags/2018/fundamentals.tdl). In particular, _on_p_state is a subtype of _on_p_loc. The _on_p_loc predicate constrasts with _on_p_temp, as in “on Monday”. These various subtypes aren’t specific to “on”, but also exist for other prepositions.

Do you mean that _loc doesn’t stand for location? It is fine that loc is a subtype of state, but we are talking about a situation - more general - not about a location - specific space position.

I got the point about _be_v_id being unnecessary here.

My understanding of _state prepositions is that they are “stative” prepositions that are used in sentences like “What do you see on the table?” meaning “what do you see while you are in the state of being on the table”. So I guess both the _loc and _state prepositions are talking about a location and they both make the same amount of sense in this context.

I guess my real confusion here is that on does not seem to be acting like a preposition at all here (from my naive perspective). I guess I expected the MRS to use on_a like you see with off_a when you parse “get the off light”. Was this just a ERG design choice to just reuse an existing predication instead of making another, or is it really a preposition in this context?

[edit] Or maybe it is just an idiom in this context and nothing would have been exactly right so it didn’t matter?

The ERG only makes coarse-grained sense distinctions which follow from a syntactic distinction. So the process of designing a hierarchy of predicate types for “on” means considering different usages of “on”, and seeing whether the usages can be grouped according to some syntactic contrast. The contrast between _on_p_loc and _on_p_temp has to be understood syntactically (and in this case, is part of a larger predicate hierarchy for prepositions). The choice of labels loc and temp is somewhat arbitrary – in this case, it also acts as a mnemonic, but it’s not a definition.

For ease of reference, the relevant part of the hierarchy (in the above file):

_on_p_rel := prep_mod_rel.
_on_p_loc_rel := _on_p_rel & nontemp_loc_sp_rel.
_on_p_dir_rel := _on_p_loc_rel & dir_rel.
_on_p_state_rel := _on_p_loc_rel & state_loc_rel.
_on_p_temp_rel := _on_p_rel & temp_loc_sp_rel.

As for whether “on” is a preposition here, it’s not a preposition in the sense used in old-fashioned grammar books (which would insist that a preposition has to prepose a noun), but we don’t need to stick to old-fashioned grammar books. Huddleston and Pullum (2019) discuss this point at length, and say:

It is also clear that words like in, up, down, over, through, etc., are sometimes not followed by a noun (or NP) or anything at all.

But in any case, from a semantic point of view it doesn’t matter whether it’s a preposition or an adjective or even a verb. In this usage, it’s a two-place predicate whose ARG0 is an event.

The _off_a_1 predicate is specifically for an attributive use (see: https://lr.soh.ntu.edu.sg/~bond/cgi-bin/ERG_mo/showtype.cgi?lexid=off_a1), e.g. “off-season”, “off-trail”, “off-budget”. It seems to me that it would make sense to have a corresponding _on_a_1, but this use isn’t what you’re looking for anyway.

My intuition is also that this is an adjective, and you would need this for sentences like The switch is in the on position.

1 Like

I am already familiar with the coarse-grained senses of ERG. But I guess I will only understand the @guyemerson discussion on the syntactic contrasts if I can find the rules that uses the _loc and _state senses (and the others mentioned above). I mean, it is not obvious to me when the loc sense is used and when the state sense is used.

The examples in this thread could be handled with an extra pair of lexical entries for “on” and “off”, but these aren’t the only prepositions that have this adjective-like behaviour, e.g. as well as “the on position”, we could also have “the up position”, “the down position” etc. Predicatively, there are other examples like “the time is up”, “the game is over”, “the server is down”, etc. I don’t think there’s something special about “on” and “off” (but of course I may have overlooked something). If prepositions regularly behave like this, it makes sense to deal with this using syntactic rules (rather than extra lexical entries), and I think that’s what the ERG currently does.

As for _state, it contrasts with _dir, as two different subtypes of _loc. My vague understanding is that some verbs select for one or the other. In “the light is off”, it’s not being selected for, so it uses the underspecified _loc.

Perhaps @Dan could comment on the above points with more precision than I can.

1 Like

In my experience, they have both been generated as alternative parses when you have a sentence like “get the book on the table” or “is there a mouse on the table?” You can’t tell from the syntax, in those examples at least, whether on_loc or on_state is correct, so the parser gives you each alternative. The parse that gives you the correct semantics has to be decided from other context (like whether or not the person is actually on the table).

Not sure if this is what you were asking @arademaker

[BTW: This is purely for my understanding of how the system is designed, I now understand enough to code up what I need] I’m confused by this, @guyemerson, and I’m sure its because of my lack of linguistics knowledge (I did read that paper you sent, BTW, makes me sad about all the time my kids spent learning grammar). But, if prepositions regularly behave like this, wouldn’t you want to introduce a tag like _p_adj to aid in understanding the semantics? I do notice, running through a few examples like “the game is over” and “the time is up” that the preposition form actually chosen is different. For example, _over_p_dir__ex and _up_p__ex. I’m not sure if this is meaningful or just a choice that had to be made since there isn’t a form like _p_adj that is introduced. For my uses, as long as it stays consistent for the particular preposition, I can work around it, but it is a bit confusing.

Yes, this is where I landed. As long as the predication chosen is the same when a preposition is used adjectively, regardless of what the tags on the predication name are, things are pretty straightforward. I was mostly just trying to understand the ERG design choices and why there wasn’t a predication variant like on_p_adj that got created.

Thanks, Guy and Alexandre, for good explanations of the grammar’s intentions with these prepositions. There are a group of prepositions including on' and off’ that can be used with either a directional sense or with a stative one: “the hamster ran under the table” can either mean that it moved from somewhere else to a spot under the table, or that it stayed under the table (say in its cage) and ran there. Since some verbs syntactically insist on a PP with the directional sense (“the hamster sneaked under the table” but not “*the hamster sneaked”), the grammar has to represent this directional/stative distinction, but to avoid a proliferation of preposition lexical entries, we use underspecification of their predicates as you’ve seen, with loc specializing to dir or state when grammar constraints insist on it. I can see why you want to view intransitive uses of “on/off” as more adjective-like in “is the light on/off”, but I fear that this predicational use of prepositions is not sharply defined enough to be encodable in the grammar. Consider “she stayed on for a a few moments longer before hurrying off” referring to an actor on a stage. A light being on or off is a little more vexed, since it’s a different kind of state than for an actor on a stage, but these specialized senses conditioned by the semantic sort of the noun or event being modified are beyond the scope of the ERG, when we don’t have grammatical constraints to determine the choice at least some of the time.
Seeing that intransitive “on” etc. don’t always carry a clear sense of location, though, I’m thinking I should change their lexical entries so the predicate name is just say “_on_p_rel” instead of “_on_p_loc_rel”. Let me try that improvement, to see if the generalization of that predicate name causes any trouble.

3 Likes

I didn’t mean to disillusion you about traditional grammars! There are plenty of things you learn at school which you have to “unlearn” if you study the subject in more detail. (Not too long ago, I was reading about biological classification, and was amazed to see that “protoctista” isn’t regarded as a valid grouping in modern biology, despite what I learnt at school.) If your kids can classify different uses of “before”, that still shows linguistic understanding, even if modern linguistics can give us a better analysis.

I think Dan’s already answered your other questions better than I can, but I’d be happy to discuss anything further.

1 Like