More circumambience in syntax

🕑 2 min • 👤 Thomas Graf • 📆 November 14, 2019 in Discussions • 🏷 subregular, syntax, movement, complementizer agreement, relative clauses, English, French

This is a very short follow-up to my previous post on circumambient patterns in syntax. I just realized that there’s another, very robust example that makes all the cases look even more mundane: complementizer alternations in English relative clauses.

Circumambient patterns in syntax

🕑 12 min • 👤 Thomas Graf • 📆 November 11, 2019 in Discussions • 🏷 subregular, syntax, phonology, tone plateauing, movement, complementizer agreement

Last week I gave an invited talk at UMass on the subregular program and the computational parallels it reveals between syntax and phonology. If you’re curious, the slides are on my website. The talk went over a lot better than I expected, and there were lots of great questions. UMass has a tradition of letting students ask questions first before the faculty get to chime in, and the students were relentless in a good way. I think there was only 5 minutes left for faculty questions at the end. It was a great experience, and probably the best question period I’ve ever been on the receiving end of.

Anyways, after the colloquium Brian Dillon asked a few questions about more complex movement cases, and those are very interesting because they’re yet another instance of computational parallelism between phonology and syntax: tone plateauing = movement-driven complementizer agreement.

Shellshock

🕑 8 min • 👤 Thomas Graf • 📆 November 06, 2019 in Discussions • 🏷 syntax, Larsonian shells, constituency, CCG, Minimalist grammars

This semester I am teaching a seminar on computational syntax. It’s mostly on subregular syntax, but I started out with a discussion of CCG. CCG is noteworthy because it is a theory-rich approach that has managed to make major inroads into NLP. It would be cool if we could replicate this with MGs, but in order to do that you need a killer app. Subregular complexity might just be that because CCG doesn’t have a regular backbone, so it can’t have a subregular one either (more on that in a future post). CCG’s killer app was flexible constituency and a one-to-one mapping from syntax to semantics. You combine that with a corpus (CCGbank) and an efficient parsing algorithm (e.g. supertagging with A* parsing), and you have something that is both linguistically sophisticated and sufficiently fast and robust for practical applications. Anyways, this post collects some of my thoughts on flexible constituency and how it could be emulated in MGs. Spoiler: shells, lots and lots of shells.

We need a framework?

🕑 7 min • 👤 Thomas Graf • 📆 October 30, 2019 in Discussions • 🏷 methodology, subregular

As you might know, Stony Brook hosted AMP, the American Meeting on Phonology, a week ago quite a while ago (yikes, almost November again). Jane Chandlee started off the conference with an invited talk on the subregular view of phonological mappings from underlying representations to surface forms. It was well received, but during the question period Bruce Hayes (no, not that Bruce Hayes) made a point that I found puzzling: “You need a framework!” Unfortunately I didn’t have time to ask Bruce afterwards what exactly he meant by that. But every conceivable interpretation I’ve come up with I vehemently disagree with, and I think Bruce’s demand for a framework stems from incorrectly applying the linguistic modus operandi to computational work.

Some musings on corpora

🕑 9 min • 👤 Thomas Graf • 📆 September 05, 2019 in Discussions • 🏷 syntax, corpus linguistics, Minimalist grammars, Combinatory categorial grammar

Pro tip: Don’t start a multi-part series of posts on locality right before the beginning of the semester and when you have a pile of papers to review. On the upside, this will give you guys some extra time to digest all the concepts in the three previous posts. In the meantime, here’s a quick and dirty post on corpus linguists and why it should be part of our syntax curriculum. I didn’t even proofread it, so beware.

KISSing semantics: Subregular complexity of quantifiers

🕑 9 min • 👤 Thomas Graf • 📆 July 26, 2019 in Discussions • 🏷 subregular, strictly local, tier-based strictly local, monotonicity, quantifiers, semantics, typology

I promised, and you shall receive: a KISS account of a particular aspect of semantics. Remember, KISS means that the account covers a very narrowly circumscribed phenomenon, makes no attempt to integrate with other theories, and instead aims for being maximal simple and self-contained. And now for the actual problem:

It has been noted before that not every logically conceivable quantifier can be realized by a single “word”. Those are very deliberate scare quotes around word as that isn’t quite the right notion — if it can even be defined. But let’s ignore that for now and focus just on the basic facts. We have every for the universal quantifier $$\forall$$, some for the existential quantifier $$\exists$$, and no, which corresponds to $$\neg \exists$$. English is not an outlier, these three quantifiers are very common across languages. But there seems to be no language with a single word for not all, i.e. $$\neg \forall$$. Now why the heck is that? If language is fine with stuffing $$\neg \exists$$ into a single word, why not $$\neg \forall$$? Would you be shocked if I told you the answer is monotonicity? Actually, the full answer is monotonicity + subregularity, but one thing at a time.

I'm not done KISSing yet

🕑 3 min • 👤 Thomas Graf • 📆 July 25, 2019 in Discussions • 🏷 methodology, semantics

I just got back from MOL (Mathematics of Language) in Toronto, and much to my own surprise I actually got to talk some more about KISS theories there. As you might recall, my last post tried to made a case for more simple accounts that try to handle only one phenomenon but do so exceedingly well without the burden of machinery that is needed for other phenomena. My post only listed two examples for syntax as I was under the impression that this is a rare approach in linguistics, so I didn’t dig much deeper for examples. But at MOL I saw Yoad Winter give a beautiful KISS account of presupposition projection (here’s the paper version). That’s when it hit me: in semantics, KISS is pretty much the norm!

KISSing syntax

🕑 7 min • 👤 Thomas Graf • 📆 July 12, 2019 in Discussions • 🏷 methodology, syntax

Here’s a question I first heard from Hans-Martin Gärtner many years ago. I don’t remember the exact date, but I believe it was in 2009 or 2010. We both happened to be in Berlin, chowing down on some uniquely awful sandwiches. Culinary cruelties notwithstanding the conversation was very enjoyable, and we quickly got to talk about linguistics as a science, at which point Hans-Martin offered the following observation (not verbatim):

It’s strange how linguistic theories completely lack modularity. In other sciences, each phenomenon gets its own theory, and the challenge lies in unifying them.

Back then I didn’t share his sentiment. After all, phonology, morphology, and syntax each have their own theory, and eventually we might try to unify them (an issue that’s very dear to me). But the remark stuck with me, and the more I’ve thought about it in the last few years the more I have to side with Hans-Martin.

Vision on P vs. NP

🕑 3 min • 👤 Thomas Graf • 📆 July 03, 2019 in Discussions • 🏷 fun allowed, complexity theory

Come and listen to the Vision of the Avengers, who has saved this planet thirty-seven times. Listen to his story of P vs. NP. No, seriously, the following is an excerpt on complexity theory from Tom King’s Vision.

News from the MG frontier

🕑 3 min • 👤 Aniello De Santo • 📆 June 24, 2019 in Discussions • 🏷 MGs, parsing, NLP

True to my academic lineage, I’m a big fan of Minimalist grammars (MGs): they are a pretty malleable formalism, their core mechanisms are very easy to grasp on an intuitive level, and they are close enough to current minimalist syntax to allow for interesting computational insights into mainstream syntax. However, I often find that MGs’ charms don’t work that well on my more NLP-oriented colleagues — especially when compared to some very close cousins like TAGs or CCGs. There are very practical reasons for this, of course, but two in particular come to mind right away: the lack of any large MG corpus (and/or automatic ways to generate such corpora) and, relatedly, the lack of efficient, state-of-the-art, probabilistic parsers.

This is why I’m very excited about this upcoming paper by John Torr and co-authors (henceforth TSSC), on a (the first ever?) wide-coverage MG parser. The parser is implemented by smartly adapting the $$A^*$$ search strategy developed by Lewis and Steedman (2014) for CCGs to MGs (basically, a CKY chart + a priority queue), and coupling it with a complex neural network supertagger trained on an MG treebank.