MP amateur looking for references to get up to speed

[poemworld]

Howdy y'all,
I'm the new guy on the block and I'm looking for references to get me up to speed and current with MP. Yeah, I know how to use teh google, but I often don't really know what I'm looking at or whom I should be reading, apart from NC. So far, I've made it through "Approaching UG from the Bottom Up", I'm plowing through "Phases", and I just started "Faculty of Language: What Is It, Who Has It, and How Did It Evolve". My degree is in math, so I'm checking out TAGs. I'm also interested in PIC. Anyone with insights or suggests, and especially elegant interpretations (or even not-so-elegant) would be richly appreciated. Thanks and have a nice weekend.

[Daniel]

Sounds like a good start. Let's see--

1. The Minimalist Program is a research agenda, not a theory. So there cannot be an "advances" made in it. The idea is 'perfect' as originally proposed: try to strip down the theory to only the necessary components. There can be advancements made in linguistic theory due to using this approach, but one odd result of the MP is that now in some ways the state of the art in Generative Grammar seems more splintered than it was before. With the GB theory, there was one mostly coherent whole to which new components could be tested and added. With the MP, everyone can propose a different optimized/minimal theory, and by their nature these theories aren't compatible. GB accepted additional components, while the MP makes a point of rejecting them unless absolutely necessary. So there isn't a single state of the art theory for the MP, and that means you'll need to pick one (or more) to read about for now. Some relatively accessible textbooks (by Hornstein for example, just to name one linguist) propose specific instantiations of MP-inspired theories that can actually be analyzed as theories. The MP itself isn't a theory, and isn't even falsifiable. There's no way to inspect it to see if it is coherent (yes, everyone agrees: the simplest still-functional theory is probably the best one). But you can inspect a fully developed theory like those proposed in recent years by Hornstein and others. The core of the MP then ends up being a debate between these different theories. Some argue that Agree is required, others that it is not. Some argue that Move and Merge must be maintained as separate components, while others say Move can be considered a subcase of Merge (so-called Internal Merge, for example, where a sub-constituent is moved to a modifying position and the first copy deleted). Because components of a theory must ideally be analyzed only in context of the whole theory, it's very difficult to compare different approaches by different linguists because the theories are different and thus the component aren't really the same thing. So you must make a few assumptions and then play with a toy grammar looking at just a small number of possibilities. And then repeat with other components/variation, and so on. Lots of trial and error. And to make things more complicated, the MP assumes the last 50+ years of syntactic research as its foundation, so to fully follow the MP you also need to know everything that came before it, both to avoid reinventing the wheel and also for specific empirical arguments that carry over or still need to be solved. In this sense, I think we can say that everyone is a student of the MP and no one is a master of it, even Chomsky, because it's a goal rather than any narrowly defined theoretical ideas. There are of course some trends such as Merge being central in almost everything written from a Minimalist perspective. You could propose an equally "minimal" theory not involving Merge, but it wouldn't really be "Minimal" I guess. It's complicated, so don't worry if you feel lost sometimes. I think everyone is.

2. Chomsky 2005 "Three Factors in Language Design" is a nice broad perspective on some aspects of the MP, and relatively accessible. Chomsky 2013 "Problems of projection" and 2015 "Problems of projection: Extensions" are very technical papers but if I had to pick something as sort of a central theme in the MP at the moment that would be it.

3. I also enjoyed reading the Chomsky's prefaces to his older works in the newly published 20th Anniversary Edition of The Minimalist Program and the 50th Anniversary Edition of Aspects of The Theory of Syntax, from MIT Press. These serve as a very nice perspective-explaining introduction to current research. What is still relevant, what has been rejected, and where to go from here. Not the most obvious thing to read necessarily, but I'd recommend them especially if you're not sure where to start. There are references too, which might lead you in other directions.

4. As a mathematician, you might enjoy some of Chomsky's earliest work if you haven't read it already. The Chomsky Hierarchy and other mathematically inclined perspectives on language. That was one of his early claims to fame before he was so famous as a linguist per se.

Obviously there are many other linguists to read beyond Chomsky. Don't forget that if you're looking for other perspectives, they'll come in two broad flavors:
1. Those that strongly disagree with Chomsky within Generative Grammar and even the MP. Some of his closest colleagues strongly disagree with him often. But within the same research agenda.
2. Those that fundamentally disagree with Chomsky's approach. The approach of Cognitive Linguistics, forever, rejects almost all of his axioms for lingusitic research.
To be able to evaluate whether you believe the MP is the best approach to understanding language, it would be, ideally, best to be familiar with both variation within GG and also entirely alternative approaches. At the same time you can look at Chomsky's work to see the sorts of questions he asks-- as a mathematician you'll probably be interested in competence (e.g., the language faculty) more than performance (e.g., language processing, based on experimental work), just as one example, but it's good to be aware of the other questions out there too.

As an outlook for future research, if your mathematical interests also reach into computational questions, then trying to understand now Generative theories could be implemented algorithmically (as opposed to just as abstract relationships in a grammar) is a major open area to be solved maybe in the next 100 years (though of course that also borders again on experimental work on processing). If the approach coming out of the MP is valid, it should be something that can somehow relate to (e.g., interface with) not just having the 'language faculty' but somehow using it. Questions of "is it enough?" and "is it too much?" asked often in the MP must at some point be evaluated from that broader perspective.

---

Now, as an aside, my dissertation research is about syntax, and the theoretical impact is about the syntax-semantics interface and linearization. Unlike many Generative linguists I take a less direct approach to the mapping between structure and form. I do at least tentatively accept some ideas from GG such as Merge, but I reject the idea that GG/MP theories are sufficient for effectively and efficiently producing the surface forms in language. So I think Constructions (along the lines of Construction Grammar) actually map the abstract structures onto pronouncable, linear strings of sounds (at least sometimes, when the form-meaning correspondence is loose). I'm looking at pseudocoordination ("try and do", "go and get") as a form-meaning mismatch, and some related phenomena. My approach is certainly controversial and also disconnected from the core of the MP, but I mention it just in case you were wondering about where I was coming from in writing this.

Just to follow up a bit, if you read Chomsky's latest work carefully, you'll find that Minimalism is drifting toward a very abstract focus, not so much on the quirkier aspects of what we actually pronounce. As an important example, Chomsky's latest work (see the Projection papers for example) rejects word order as a component of narrow syntax, and instead essentially sets that aside as something to be handled by phonology! So for my own reasons, and actually not too different from the direction Chomsky himself seems to be going, I have used the name "Minimal Minimalism" to refer to an idea of a Minimal theory (in the sense of the MP) that is also designed to explain minimal (or at least less) data. In other words, if we can set aside things like word order (or the quirks of pseudocoordination) then we can retain a very efficient and pure core to the grammar in the MP sense, succeeding in that goal. However, I strongly argue at the same time that some other component (Constructions, for example) is required to fully explain why we say the words we do in the way that we do. Therefore I think, and hope, that the MP might be going in that direction: being Minimal in two ways, and also thereby requiring an explanation for the other data set aside. Personally it's the quirky surface-level data that catches my attention the most, while for Chomsky it seems to be the most abstract core generalizations that can be made about the human language faculty, so that's one thing that sets us apart. I look for exceptions, and he looks for rules.

By the way, that leads to one other important area to consider reading: the idea of "Principles and Parameters"  has become less popular lately, especially with how incompatible it would be for a truly Minimal syntactic theory, yet some linguists such as Ian Roberts and Theresa Biberauer at Cambridge are reviving the parameters approach and pursuing it very seriously. That is worth looking at because it is perhaps the most divergent from Chomsky's work today within the most narrow sense of 'Generative Grammar' (not getting into LFG, HPSG, etc., and certainly not Cognitive Linguistics).

[poemworld]

WOOT! Thanks Daniel. I just saw this and will dig in.

[poemworld]

By the gawdz this is helpful! Thanks again!

I've noted "Projections" but hadn't pursued it yet. What's your take on Phases? I've read some stuff on the relationship between TAGs and PIC. Any comments there?

The part about everyone being confused is heartening, in a sort of diabolical way. Appreciated XD

[Daniel]

"Phases" are a little odd from the perspective of Minimalism. They may be necessary. But they seem eerily like processing to me. Merge simply takes A and B and makes AB, and repeat. It doesn't in itself predict phases. In fact, the system is much more complex than basic Merge (like that) if we allow the internal structure to continue to be manipulated as in Move (or Internal Merge as Move). Phases impose barriers to that sort of operation at certain levels, and it makes sense. But what it makes me think of is processing. Humans have memory limitations (that's a whole different area of research) that limit how much embedding is actually possible in real production (but not Competence according to Chomsky and others). Phases/barriers like that would in a way avoid that problem by "flattening" the structure below that barrier as essentially a finished, linearized sequence ready to pronounce. It makes sense. But it's a little less minimal, and again eerily like Performance, compared to the abstractness of Merge. In short, Phases bring back in the Linguistic Syntax that is lacking from the most abstract version of Minimalism, e.g., just Merge. If we do have Phases, then do we also have cross-linguistic parameters to deal with how they might vary? Or are they instead principles? Thus, they're part of UG either way, and somehow in our DNA. Optimally minimal syntax would avoid assuming that. Or, instead, are they actually not part of Chomsky's narrow syntax, maybe learned from input or due to third factors (processing, memory?). But in short, yes, something like Phases seems to be necessary. But how?

Tree-adjoining grammar (TAG) vs. standard GG: it isn't so clear to me what the important differences are, since evaluating that requires having something specific to evaluate. In other words, since I'm not sure about various basic assumptions/predictions of syntax it's hard to say which narrower approach is best. There is, interestingly, once you get into something like Lexicalized tree-adjoining grammars, not that big a leap to get into Construction Grammar. Note that Construction Grammar also has hierarchical structure, where one construction embeds another. The difference is simply that Construction Grammar is more powerful (and less mathematically precise/constrained) because Constructions can (at least pre-theoretically) do just about anything. Given that languages do seem to sometimes do just about anything, trying to make as an assumption of the theory that they don't do certain things is a risky approach in my opinion. Of course if it could be definitively shown that only certain types of operations occur in language (as Chomsky repeatedly argues) then it would be desirable to have the more constrained theory. So I guess I'm agnostic at this point about those things.

One approach that I find intuitively interesting, and you might as well, is Categorical Grammar, which uses valency-based types rather than the typical labels so that "NP/N" is the type for the word "the" because when you combine a noun with it, it becomes a noun phrase. There are some examples here:
https://en.wikipedia.org/wiki/Categorial_grammar
What I like most about that approach (again, it's hard to say whether it's better than the others, because there are various unknowns in the big picture) is that it has a very conveniently computational aspect to it. Various theories like this might actually be isomorphic to other theories, but it's interesting to look at some of the formal variants in notation and how they might be implemented computationally or how they might be processed, etc.

Back to Phrases: the Phase Impenetrability Condition (PIC) is an inherent assumption of the idea of Phases, of course. Phases would be pointless (or not exist) without it. As for implementing it with something like TAGs, that relates to my comments above. It makes sense in a way, but seems to bring in something like processing. And also like to think of this sort of thing as "chunking" in the way that people seem to memorize commonly pronounced elements ("Hello, how are you?" is memorized as a unit, of course, even though we also know its structure), which can lead to idiomatization, and then also to Constructions. But Phases are a narrower version of that in a sense because they are specifically syntactically defined (at certain nodes) as opposed to just based on usage.

I don't have more direct comments about TAGs/PIC except to say that it's good you're thinking of this in terms of implementation.


So those questions aren't quite narrowly related to what I study, but I hope my comments might be somewhat helpful still.

[poemworld]

Daniel, I want to have your baby. You have no idea what a service you are rendering to me. My gratitude runneth over. I wish I could be as useful and valuable to you as you are to me. Maybe someday... But for right now, you're connecting so many dots at once that I'm feeling a bit overwhelmed. Thank you again SO much. I'm in your debt. I have no idea why I never thought of looking for a linguistics forum before. I'm an idiot sometimes.

[poemworld]

When you say "narrow syntax", do you mean FLN or something else?

[poemworld]

Also, you mention "processing" with what sounds like wariness. Why?

[Daniel]

Glad it's helpful! Everything is a bit nebulous because things can be defined and implemented in different ways.

"Narrow syntax" can be FLN (Narrow Faculty of Language). Roughly what Chomsky seems to be trying to solve with Minimalism. I don't intend the term too precisely, because we're not sure what the boundaries are. (Syntax in general can include things like word order, but those seem to have been separated for now.)

Also, you mention "processing" with what sounds like wariness. Why?

Actually I really like research in psycholinguistics and attempts to connect syntactic theory to processing. And you can tell I like aspects of Cognitive Linguistics too. So I study Generative Grammar, but I'd skeptical about it and interested in other things.
But the wariness you correctly detected is me playing devil's advocate basically: Generative Grammar is so much about Competence that getting into Performance (processing, etc.) is sometimes like trying to mix oil and water. For example, you can actually exclude any data at all as "just Performance" and therefore not relevant to the more abstract and not directly observable "Competence". Chomsky and others use native speaker intuition and reflection to get at grammaticality judgments for studying Competence, but of course Performance (some kind of faulty intuition I suppose) could also cloud those judgments. It makes falsifying Generative Grammar very challenging, plus the fact that "Generative Grammar" (or especially the MP) cannot be falsified overall because it is an approach rather than one theory. Adjust the theory slightly, and continue.

So, personally, I have mixed feelings about this. But most Generative syntacticians seem to like separating out any sort of processing-related arguments for theoretical discussions of narrow syntax. Personally I might encourage you to look in another direction than Generative Grammar at least for addressing some questions. (I have seen absolutely no convincing evidence that GG is psychologically/biologically real, for example!) But I don't want to discourage your interest in it, and it is a very popular theory.

There are a few linguists who are actively trying to bridge the gap between performance and competence from a Generative perspective, even some who are serious Minimalists who are doing experimental work related to that. But they're in the minority, and mostly Minimalists are serious theoreticians looking at abstract rather than experimental data.

Any version of Generative Grammar that starts to rely on processing-like arguments drifts away from its foundation.

As an example of this, we often think of derivations as directional, starting with the lexicon and an abstract structure, and then applying rules sequentially until the form we pronounce is generated. But actually Generative Grammar (despite the name!) is not directional (in most accounts). It simply describes structural relationships between words. To compare it to math, I suppose it's just like how "1+1=2" doesn't mean that you must first add 1 to 1 then get 2, but also that 2-1=1, and various other relationships. It is, overall, completely unknown how Generative Grammar is actually implemented in the brain, and the theory cannot rely on such arguments. Chomsky himself does often discuss arguments about acquisition, but those are slightly different: since children learn X, they must be capable of learning X from the available input, and therefore the theory of syntax needs to explain what their UG is like so they can do that from that input. It borders on being about Performance, but it's actually about Competence. It doesn't explain HOW they do it. It just points out that they must be able to, so their Competence logically permits it.

Personally I'd be more than happy to rely on something like frequency or memory as a factor in explaining how syntax works. I don't think we need to explain everything in an abstract theoretical way if we can instead rely on getting it for free as part of something else (like memory). Chomsky wouldn't entirely disagree with me: he calls those things "third factors". But we may disagree (or may both not yet know) about where exactly to draw that line. As a basic example, I wonder if recursion can be accomplished as an iterative performance process rather than as something that must be defined in terms of competence. That could also potentially explain phases: there simply are boundaries that we do not cross for whatever reason in performance (maybe just communicative need not applying, or maybe they're difficult to process, or something else), and then those become conventionalized (Constructionalized?) in the grammar. Chomsky instead plays with the idea of those being encoded in UG, which I am skeptical about. But something like his Phases is needed, one way or another.

In short, I agree with Chomsky about something like Merge existing. And I also think we need something beyond standard Generative approaches (like Constructions). Everything in the middle I'm currently agnostic about.

So, above, I was trying to address your questions from the perspective of GG/MP (where "processing" is irrelevant!). I personally like the idea of connecting processing to syntax, but at the same time I'm not sure of how to do it (or exactly what would be connecting to exactly what!).

---
My knowledge of mathematics is limited, but I think you have a similar division, between "Analytical" solutions (like Competence, GG) and "Numerical" solutions (like Performance). The parallel isn't exact, but throwing in a Monte Carlo simulation in the middle of an Analytical problem would be about the same effect as referencing processing to explain narrow syntax in GG.

Similarly, more and more I'm starting to see different theories as not being "right" vs. "wrong" but as addressing different questions (like Analytical vs. Numerical mathematics). For example, the MP is great for explaining hierarchical structure, while Construction Grammar is great for explaining surface-level irregularities. So maybe it's not about figuring out which theory is best, but instead about which theory is best for which purpose. Chomsky's purposes are almost entirely about UG, FLN, Competence, etc. So in addition to the other many types of background reading for you to think about, also try to figure out what sort of questions you want to solve and whether those questions are the same as what Chomsky focuses on-- if so, the MP may be right for you. If not, there are other approaches. (Chomsky's work, for example, is probably the least likely to lead to a usable machine translation system today, for example. But that's not what it's for, either.)

[poemworld]

Daniel, everything, and I do mean EVERYTHING, you're telling me is revelatory. I'm going to need to think about this. I'm in the midst of writing something right now that you might find interesting. As a "theory" it's largely complete. You've basically confirmed that my intuition has been essentially correct. At this point, I'm willing to begin sharing with you, if you wish. However, I want to finalize what I've got in print before going any more public at all. My ideas are heterodox, to say the least, but well-grounded, as you've now all but confirmed for me. You deserve to know what I'm up to. You've earned that right. The choice is yours. I don't know how else to thank you except by being candid with you.

[Daniel]

You're welcome to share your ideas. Feel free to start a new thread, or you could send me a private message.

I don't know that I'll have time to read/analyze a full theory (happy to discuss things I'm familiar with but it takes longer to evaluate something new, independently), but I would at least like to hear about the general concept you're going for, or the questions you're trying to answer.

If you do want to start a new thread to discuss your ideas, we have a forum for that: "Outside of the box" (within the Linguist's Lounge). There's a rules thread there, but in short any unusual/controversial theories belong there but are welcome.

[poemworld]

No, I don't want to put this on a thread yet. I'm willing to send you a tentative abstract I've written, so it's not much to read. How would you prefer I send it?