"no matter"

[nschneid]

In EWT it is currently: advmod(matter/ADV/RB, no/ADV/RB), which is counterintuitive to me.

For the internal structure, GUM uses a compositional analysis: det(matter/NOUN/NN, no/DET/DT)

In PTB guidelines I don't see any discussion of the tags or the internal structure of the expression. There is an example where "no matter" forms an ADVP with an SBAR complement. CGEL p. 761 says "no matter" is a single lexeme that takes a subordinate clause complement, and the bottom of p. 764 claims that the complement can be reduced. Simple English Wikipedia (based on CGEL's POS inventory) classifies it as a preposition (analogous to the PTB IN tag). Rawlins (2008, p. 80) tentatively suggests "no matter" is a complementizer. Kim and Lee (2011) seem to treat it as a noun expression that takes a complement.

"No matter" typically takes an interrogative clause complement, but it can take a "concealed question NP" as well—e.g. "the time", "the reason" (Kim and Lee 2011, p. 962). The meaning and external syntax are similar to "regardless of".

This leaves open the possibility of treating "no matter" as a marker, which suggests it should be connected via fixed rather than det. For "no matter what it costs", or "no matter whether it costs too much", it would be mark; and case seems natural for "no matter the cost". But if we treat it as formed via det, the case for attaching it as mark or case becomes weaker.

Or, it could be treated as a nominal formed by det that takes an acl complement. This works for "no matter what it costs" (or "no matter what", which can be considered to contain a reduced clause). But is acl appropriate for "no matter the cost"? Should we say that is obj (analogous to worth/ADJ having an object)? Or nmod:npmod, perhaps?

In any case, I don't think this expression licenses a relative clause, which is annotated in some places in EWT. I will implement the following for now:

• For the internal structure, follow GUM and treat the expression as transparent: det(matter/NOUN/NN, no/DET/DT)
• Treat what follows as a dependent with acl (if a clause or expression suggestive of a reduced clause) or nmod:npmod (if it has a nominal complement which does not look like ellipsis).
• For the attachment of the full "no matter..." phrase (headed by "matter"): obl:npmod

[nschneid]

Kim and Lee argue that the internal structure should be regarded as fixed because it is not productive: *little matter, *not any matter, *no possible matter, *no matters. AFAICT all the above sources treat it as a fixed expression. I think fixed would be warranted on this basis, with the marker (mark/case) analysis paralleling several other expressions that we treat as fixed.

[amir-zeldes]

I don't necessarily think it's not fixed, but I also don't like inflating the fixed list which annotators already find hard to keep in their heads, and there is no real differences between this 'no matter' and other 'no NN' phrases, so internally I'm happy enough if it's det.

Externally I think acl is misleading, because it makes it look like there's an NP headed by "matter" which has a clause expansion, when in fact I think it's some kind of predication. I would prefer to assign a structure to it that is more similar to "X does not matter", or "regardless of X". The latter is hard because it would require 'no matter' to be mark, and it doesn't really express the predication. How about treating the X as a subject?

Do it no matter the cost
advcl(do, matter)
nsubj(matter, cost)

Do it no matter whether they like it
advcl(do, matter)
csubj(matter, like)

Alternatively you could view it as a secondary adverbial/obl:npmod, thinking the whether case is like 'if'. This has the advantage of fitting a 'when' clause better:

Do it no matter the cost
advcl(do, matter)
obl:npmod(matter, cost)

Do it no matter whether/if/when they arrive
advcl(do, matter)
advcl(matter, arrive)

[nschneid]

"no matter the cost" with nsubj like 'the cost is no matter': I see the logic semantically but it looks totally weird to have a nominal predicate preceding its subject with no copula. I would rather have a noun with an object than this.

It's not obvious to me that "no matter the cost" is a clause...seems not too far from "apart from the cost" or "regardless of the cost".

[amir-zeldes]

I mean, if there's a predication in there then I think it should be a clause. If it's not a subject, then how about the advcl analysis? We need to be able to explain what's happening with "no matter if/whether" etc.

[nschneid]

I'm not convinced "no matter the cost" has a predication:

1. Clauses can usually be negated, right? Even verbless clauses like "With his feet on the table, ..." can be negated to "With his feet not on the table". But you can't say "no matter not the cost" or "no matter no cost" or remove the "no".
2. How is it more predication-like than "regardless of the cost"?

The fixed+case/mark analysis seems simplest to me—that way we say it can mark a clause or an NP. But I'd like to hear opinions from others.

[amir-zeldes]

Clauses can usually be negated, right?

No, not necessarily. Some clauses are always and only negative ("lest" clauses come to mind), and some are only positive (comparative corelatives, you can't say "*the not faster the better"). Some languages systematically use different clause types based on the positive-negative axis (Coptic is notable in having lots of these).

How is it more predication-like than "regardless of the cost"?

Because you can say "no matter if they come or not". "Regardless" requires mediation with "of". That's not necessarily a deciding factor, I just find the pattern a bit odd for a standard 'preposition'. If it only took nominals I guess I could see it as equivalent to "regardless" (though really I think that's an adverb which takes a PP complement headed by "of"). But what would you do with the "no matter whether" cases? Would you want double mark here?

[nschneid]

"Lest" clauses can be negated: "Lest we not arrive on time, ..."

But you're right about comparative correlatives. Those are one of the minor clause types discussed in CGEL pp. 944-945.

My point about "regardless of" was not to make an argument about the POS of "regardless", just to say that if it's not considered a predication, then I'm not sure why "no matter the cost" should be considered a predication.

UD does allow P + "whether" clauses as advcl: "They disagree about whether..." etc. I think those are properly considered complements (the whole cobl discussion), but formally speaking, there would be nothing exceptional about double mark on an advcl.

As true adjuncts, there are also "whether or not..." and "whether X or Y" type clauses, in case that is relevant.

So I guess our options are

• decide that "no matter X" is a minor clause type—the only adjunct type that prohibits negation to my knowledge—that does not begin with a marker, or
• decide that "no matter" has grammaticalized as a marker, and can function as case or mark, including before "whether".

[amir-zeldes]

Huh, I only find "lest we not" followed by "forget", and discussions about whether that's correct (or whether that even has double negative meaning):

https://www.quora.com/What-does-lest-we-forget-mean-1
https://groups.google.com/g/alt.usage.english/c/3IDMCNlDc0I?pli=1

I think there are some theoretical papers about constructions like these which allow inserting a negation without a change in meaning.

I think there are a bunch of constructions that come with built in negation but are still clauses, like "not to brag, but..." where you can't do a positive "??to brag, ...". Of course, this shows some degree of fossilization, but I have taken the UD policy to be preferring a compositional analysis whenever possible, including within names and other fossilized expressions. And in languages other than English, asymmetry between positive and negative clause possibilities is sometimes much greater, so either way I wouldn't take negation as an indication of clause status.

As for the options, TBH I'm not categorically opposed to a case or mark analysis, but I'm just wondering if it's really necessary. Essentially I think the list of 'fixed' in English is getting bigger than it should be, so I'm a bit loathe to see it grow. The internal structure of "no matter" seems pretty normal to me (similar to "no reason to fret" and others), so I'd like to see the 'det' relation survive (also fits the established xpos).

[nschneid]

I mean "lest" is largely archaic to begin with. But I see no reason to believe "lest" clauses resist negation. Examples from the web:

• Lest the reader not understand, Bailey specified the sort of “small dogmatisms” that ought to be put away.
• We see this principle reflected also in restrictions against workers moonlighting in second jobs, lest they not have the energy to devote themselves fully to their first job
• I'm always a bit disinclined to change config files for things like, lest they not get changed back

"Not to brag" is just a normal negated infinitival clause that has become formulaic. I think the same syntactic construction as "To elaborate, ...".

[amir-zeldes]

I see no reason to believe "lest" clauses resist negation

OK, I'm convinced about lest - but there are definitely clauses that can only appear in the positive/negative, even though it's very rare in English (in Coptic it's systematic for certain tenses/aspects).

just a normal negated infinitival clause

I think syntactically that's right, but I also think "no matter" is just a normal negated noun phrase like "no reason". They both share the property that you can't just un-negate them ("*to brag, ..." just doesn't work/mean what the idiom means). That doesn't mean that either of them is totally opaque or unstructured, and there are variants like "not to brag too much", or for "matter" we can find:

• Little matter that ninety percent of the money he earns comes from framing pictures (Paul Auster)

Or similar expressions with another noun instead of 'matter', such as 'wonder':

• No wonder you're hungry
• Little wonder that he quotes with relish from David Lodge's hilarious academic novels
• Small wonder that you feel as bad as you do this morning

I just don't want all of these to be a huge list of fixed expressions - they are certainly collocations from a frequentist perspective, but I think UD's job is just to describe the syntactic structure, which is a noun head with a negation. I'm less passionate about it being an advcl, or having a subj, I'm just trying to avoid a long shopping list of fixed, and especially when I think the internal structure is not headless.

[nschneid]

I absolutely would not want to use fixed for the "little matter"/"no wonder" ones. That appears to use a lexically more productive construction, and I'd say the noun heads the clause. It is perhaps historically related to "no matter whether it costs too much", but that and "no matter the cost" can only function as adjuncts, not as main clauses. Also:

• *No wonder the cost!
• *Little matter the cost.

[amir-zeldes]

OK, but is there any analysis you'd be OK with that keeps "matter" as the head of "no", so we can keep the normal structure there? Otherwise we're pretending it's not a negation, which I think it still is.

[nschneid]

That depends on the structure IMO. If we treat "no matter" as a grammaticalized marker, then I think fixed makes the most sense. We're saying the historical determiner relationship is no longer productive in that construction. We can still keep DET and NOUN as tags.

But I want to hear others' opinions.

[amir-zeldes]

If we treat "no matter" as a grammaticalized marker, then I think fixed makes the most sense

Yeah, I'm the least comfortable with this part - "no" is still acting as a normal negation here, meaning that it doesn't matter. So I think "matter" should be the head.

I want to hear others' opinions.

Me too - @dan-zeman @manning , any thoughts?

[dan-zeman]

I would do det(matter/NOUN, no/DET). That is, the structure attributed to GUM in the first post in this thread.

And I think I would use obl(superordinate-clause, matter)... although advcl seems justifiable, too.

[nschneid]

@dan-zeman What would you do for the relation between "matter" and "cost"?

…

On Thu, Oct 27, 2022, 3:16 PM Dan Zeman ***@***.***> wrote: I would do det(matter/NOUN, no/DET). That is, the structure attributed to GUM in the first post in this thread. And I think I would use obl(superordinate-clause, matter)... although advcl seems justifiable, too. — Reply to this email directly, view it on GitHub <#376 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAHQRLYDUMWDVY5ZPFO4W6DWFLWMDANCNFSM6AAAAAARMEXFOY> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[dan-zeman]

@dan-zeman What would you do for the relation between "matter" and "cost"?

nmod(matter, cost)

[amir-zeldes]

The English guidelines reserve nmod (and obl) to cases where there is an explicit preposition as a case marker, so that would introduce an inconsistency. We could go with nmod:npmod, but if we view the 'no matter' as advcl, then I would argue it should be obl:npmod as a clause modifier (since it doesn't actually belong to an NP headed by 'matter'). This is also one of the options I suggested above.

In sum, I think either obl:npmod or nsubj (if we view this as an argument structure like "it doesn't matter that X") are the best options.

[nschneid]

Yes, as @amir-zeldes points out we use the :npmod subtype for non-prepositional nmod/obl. Given that I gather @dan-zeman is arguing for

det(matter, no)
obl:npmod(PRED, matter)
nmod:npmod(matter, cost) in "no matter the cost"
acl(matter, costs) in "no matter whether it costs too much"

I think this is the least objectionable "shallow" analysis, i.e. the analysis that treats "no" and "matter" as literally as possible in their compositionality. But it is not ideal: a noun licensing either acl or nmod:npmod seems exceptional for the language (most nouns that license complements would have PP complements if nominal, thus nmod rather than nmod:npmod—e.g. "the fact of" vs. "the fact that").

My native speaker intuition is that "no matter" really functions as a marker (case/mark) that happens to be spelled as two words for historical reasons, but is effectively a single lexeme with build-in negation semantics. I.e. it is a single word with negation in the same way as "nobody", "without", or "notwithstanding". It happens to be conventionally spelled as two words, but I wouldn't be surprised to find people writing "nomatter", and in fact this is attested on Twitter.

Note that we decided to allow some VERBs as case/mark because they have become similarly grammaticalized.

But I do see the practical appeal of shallowly recognizing the "det" in no+matter as it would be if "no matter" were used in other contexts ("It is no matter", etc.).

[nschneid]

P.S. No matter what we decide :), I predict annotators will have to look it up on the fixed page (which is essentially a mini-constructicon).