Saturday, 30 May 2015

Athens Day 2

Athens Day 2

Quote of the day: Rizzi: ``I guess I agree with Norbert to the extent that he agrees entirely with me’’

The first session of Day 2 was a panel on the interaction of syntax with morphology and phonology (Anagnostopoulou, Cheng, van Riemsdijk).

When it comes to phonology, van Riemsdijk made a plea that we should do more of it because it might actually have some interesting internal properties that look a bit like syntax. Cheng argued specifically that prosody and intonation do  in fact have the `syntactic’  property of  hierarchy/recursive units, and that the mapping to syntax is one of the big exciting open issues.  They both argued for a reassessment of the arguments strictly separating syntax and phonology, and for building bridges.

Once it came to morphology,  it was immediately clear from the discussion that this was no bridge building exercise.   Every syntactician in the room was already convinced that morphology was in bed with syntax on one way or another, and considered it part of their domain of interest.  Specifically, the questions were of the form is the type  of syntax found inside the word level the same as that you find outside the word level. Are the features and operations identical? In a subset relationship? An overlapping relationship?  At some point on Day 2 or perhaps even Day 3, the discussion was so rich (and specific) about the issues at stake,  Bobalijk had to reiterate the point that actually the vast majority of morphologists in the world don’t believe this at all, and think that morphology is its own thing and is distinguished from syntax in having no internal structure.  For about half a second, the entire room went completely silent. And then  we just carried on.  Well, there really was nowhere to go from there, since there was no one in the room to argue with.  I have not tried to reproduce the details of the issues and individual contributions in this post since that way lies madness; I am hoping that the contributors will make their slides available for everyone to see the specifics of what was discussed.  Here, for example, are the slides from Anagnostopoulou, who has kindly provided me with a link to them here:

I cannot summarize the nitty gritty of this discussion, although it was extremely rich, specific and informed. One big picture question stood out, though, perhaps as a result of the syntax-all-the-way-down assumption. This was raised by Pesetsky some time in the middle of some morphological discussion. It  can be summarized thus:  ``Words. What’s up with that?’’ 

What followed was a flurry of discussion back and forth about whether this was a deep question or not, or whether we had an explanation or not, and are Words even a  Thing  such that there can be a mystery about them. But the discussion slowly converged on a general acknowledgement that there was a real linguistic unit here, crucially of our own making (i.e. we cannot hand it off to the phonologists to define it), and that we really have no idea how to explain  its existence (this is as opposed to being able to code  its existence), or  explain its internal variability across languages. 

In the morphology session there were no bridges to build because you don’t need to build a bridge from a thing to itself.

In the semantics session, Merchant and I seemed to agree that there was no real bridge.  Because the  proper hard core formal semanticists don’t seem to want one, and the syntacticians think that someone else is building it.

Specifically, Merchant worried that semanticists seemed very often to construct representations that used their own tools and ontologies that were a very poor fit with the shapes and ontologies of syntactic representations and their lexical items.  This gives rise to some very funny LFs if you believe that one is allowed to transform syntactic representations to make them fit better with the structure of the semantic formalism.  I agreed with this assessment, and went further to claim that a huge and coherent field of semantics exists,  which actually does not care about the match to syntax, and which is concerned only with the proper description of meaning.  Conversely, the semantic toolbox is in principle so powerful, that it does not provide any substantive constraints on syntactic representations. What is needed is an actual narrowing down of the semantic tools--- an actual theory  of semantic primes and primitive combinatoric operations that underlie natural language(s).    And not that many people are actually doing this.  (Deal agreed with me, but repeatedly asserted that she thinks this is just a temporary trend, and will be over soon once the field as a whole rediscovers the joys of the syntax-semantics interface). 

I provocatively suggested that syntacticians (since that was my audience) cannot simply sit back and relax in the belief that the problem of mapping to semantics is being taken care of somewhere else.  I even suggested that syntax-semantics might not even be a mapping problem at all, but in fact part of the core computation.   I argued that Structural Semantics is so inextricably tied up with syntactic hierarchy that the syntactician ignores it their peril.  By structural semantics I meant core things such as property predication over things of type e, modification and compositional semantic embedding.  This in turn massively underdetermines truth conditional meaning.  Let’s call  that Big Meaning.  Big Meaning  (albeit underwritten by structural semantics) is not part of the syn-sem computation, It actually sits over in the nonlinguistic C-I module. 

(In case anyone is getting upset, we can all (as syntacticians) think of a bunch of wonderful syn-sem people out there who are doing a great job. Yes. These are the ones we know. And guess what. The semanticists have a name for them---- Syntacticians, also sometimes known as NotRealSemanticists .)

When it comes to the interface with experimental and psycholinguistics, I am going to be a bit eclectic and fold in  remarks  and minidiscussions  that came from outside the actual panel slot, since the interesting controversies only seemed to emerge gradually in this domain.  Again, there were a number of specific results and ideas presented and discussed, but here I only have time for the big picture and methodological issues that struck me as interesting and stimulating.

Some people (e.g. Müller) were skeptical towards experimental linguistics, arguing that it really hasn’t discovered anything new for us, and that psycholinguistics  and especially neurolinguistics are remote from our concerns to be relevant to the kinds of questions we are asking.  But other people were more positive (e.g Rizzi).     Why this difference of opinion and what are the roots of the problem here?  One of the big problems seems to be that even though psycholinguists were eager for good input from us for what kinds of things to drive research questions,  we do not seem to be consistently providing them. This is not because we are not generating results and analyses, but because those `results’  are so  framework internal and notationally specific so as to make them useless to the granularity of the questions that are answerable by experimental or psycholinguistic techniques.  We need to scale up our questions to a different kind of granularity if we are going to reap mutual benefits.  Scaling up our questions is also therapeutic for us, I suggest, since it allows us to see the bigger issues on which the smaller debates depend.

I illustrate with an example from Rizzi’s recent work on the processing of object vs subject relatives, which he presented.  One might ask oneself  concerning certain dependencies that are known to be sensitive to intervention effects from a `like’  element, whether this sensitivity is due to a deep fact about the grammar (i.e. what we know when we know a particular language) or about the parser and its limitations.  To answer this question, Rizzi hypothesized that (i) if the effect was found equally in comprehension and production, then it was more likely to be a grammar problem than a parser specific problem, and (ii) if it was sensitive differentially to grammatically active  `like’  features as supposed to inactive features (language dependent classification),  it was probably a grammar issue not a processing or performance issue. Obviously this kind of question, experiment, and answer,  tell us absolutely nothing about the coding or implementational properties of the dependency relation if indeed it is found to be a grammatical as opposed to  processing fact.

Another big and recurring issue in the room, was the hard and (if I understand it correctly) largely open question of how to model the relationship between the parser and the grammar so that both the performance systems (comprehension and production) can make crucial reference to it while embedded in the online processes of actually producing and understanding language.   The problem is hard because we have make all kinds of decisions about working memory and things like that which have a huge impact on the result, but which we currently have no independent understanding of to guide us.    If we use a derivational metaphor in constructing our grammar description (minimalism), does that have any impact at all on its psychological plausibility?  In fact, no.  Graf says that it’s mathematically easy to write top down parsers for minimalist grammars.  Still the details of embedding grammatical knowledge within a performance system are largely unresolved to date.   Here’s a place where we need to continue to keep talking to each other.

Finally, we come to the issue of the communication of syntax with the outside world.    Van Riemsdijk  opened the very first day of the conference saying that although we have done great things in the past, our glory days are over and we fighting for our very survival. We need to shore up our position with respect to communicating with others and make friends which those who are antagonistic to us.  But also,  he claimed, we are simply not very good at generating research questions any more--- we  cannot allow ourselves to stay narrow.   The mood in the room on the end of the second day however was much more upbeat, even though we know that there really is a problem,  I think we were upbeat because we had all just spent two days have a very enjoyable and stimulating set of discussions so it was very hard to be grumpy.   Everybody put aside their differences of opinion and had a good hearty discussion containing suggestions for improving outreach to the community outside of our small group. There were different emphases, with some arguing that we need to get linguistics into schools, and others saying that communication with our fellow non linguist academics in related fields would be the most efficacious.  It was notable that most felt sufficiently positive that there were important and cool results to communicate, and all agreed that we need to do better in these various ways.  Each to his or her own talents, and sharing the burden around. 

I am going to save practical suggestions and outcomes for my discussion of the final day and end rather with A Salutory Tale  conveyed to us from the Greek linguist community.    Apparently,  a few linguists wrote a textbook for the schools, which was rather traditional in most respects, but which contained the linguistically uncontroversial assertion that Modern Greek contains five phonological vowels.  It took a while for the traditionalists to notice this, but what ensued was a furore of Orwellian proportions.  Now, as we know,  Greece is the proud inheritor of an ancient language with great prestige and cultural significance, and ordinary people have very strong feelings and opinions about it.   The problem is this: Ancient Greek had seven vowels which correspond to seven distinct orthographic marks.  Now, the Linguists had come along and were trying to steal vowels from the Greek nation.  Everyone was outraged. The book had to be banned!   Over a hundred Greek linguists signed a petition to explain  to the highest reaches of government that this was not a Mistake, and that there was difference between Ancient Greek and Modern Greek, and between phonology and Orthography.  But it was no good. The issue became a national scandal overnight and the whole country was in uproar.  The linguists (many of whom were in the room as this tale was being told) were then subjected to shouting and internet abuse, and public scrutiny of their credentials by ordinary citizens who were convinced that they knew better.  (In the end I think the Linguists managed to make their point stick in the end,  although I do not think they ever managed to win over the Common Man).

I think this is a particularly extreme example of the kind of problem the field as a whole faces. It highlights the difficulty of getting our message through to regular people, who have in many cases strong feelings and opinions about language.  But I tell this story also because the Greek Linguists have been our hosts these three days and I want to stand in solidarity with them.  They have been honourable in sticking to their ideals under trying circumstances;  they have been gracious in hospitality; they have been stimulating in conversation; they continue to maintain their standards and their good nature under the most difficult of conditions.  We are grateful to them for hosting this event.

In my blog for Day 3, I will try to assess what we got out of all of this navel-gazing.   Practical outcomes.   Concrete suggestions.   Affirmatory Lists.   Day 3 will have it all.


  1. You mention features passingly above. Have they been quite marginal to the discussion?

    1. Daniel, features were very prominent in the morphology panel part of the discussion, I just did not focus on them in the blog post. In general, I have been choosing to stay away from specific agreements and disagreements and concentrate on points of philosophy and methodology.

  2. This comment has been removed by the author.

  3. Thanks again for the summary, Gillian.

    Psycholinguistics is unlikely to generate a bunch of new MLGs. And in that respect, Gereon Müller is right that it hasn't done a lot for syntacticians. It certainly can be relevant in situations where there are disputes over whether a MLG really is a grammatical generalization at all.

    As for the broader question of the relation between sub-fields, some of the perceived difficulty of the issue is partly due to comparing apples and orange juice. Differences between tasks (e.g., speaking vs. making an acceptability judgment) are not the same as differences between levels of analysis (e.g., modeling real-time processes vs. abstracting away from time and just modeling the output), and those are not the same as differences between separate cognitive systems (e.g., the assumption that there are multiple language systems, some dedicated to specific tasks). I think that the conflation of these distinctions has led to a situation that is understandably confusing. I of course have strong views in this area, but those are not so relevant here; what is relevant is to simply distinguish the questions.

    Meanwhile, I'm not so sure that we're held back by a lack of independent understanding about "memory and things like that". There's actually a fairly rich literature out there that grammatically informed psycholinguists have been working closely with in recent years. It has its origins in psychological memory research that deals with much simpler representations than sentence structures, but it has led to lively discussions, often helped by explicit computational models. People like Rick Lewis (Michigan), who are equally at home talking with minimalists and memory geeks, played an important role in getting this line of research going. The work has mostly appeared outside the regular linguistics journals and conferences, which is why it might be less apparent to the syntax community. To take just one example, an interesting paper by Ming Xiang and colleagues at U of Chicago is about to appear in the Journal of Memory and Language that uses evidence from memory models to make an argument for how the scope of wh-in-situ is resolved in Mandarin. There are many other examples.

    Looking forward to the Day 3 report.

    1. Thanks, Colin. It is really useful to get specific recommendations from you about what to read in this area.
      I hope to address more of these issues in the Day 3 summary which is currently in the works.

    2. Ok, then a couple of specific suggestions.

      Shevaun Lewis and I have written a couple of recent pieces on this topic. The best starting point would be:
      S. Lewis & C. Phillips. (2015). Aligning grammatical theories and language processing models. J. of Psycholinguistic Research, 44, 27-46.
      An earlier paper by us focuses more on the mentalistic commitments of syntactic theories (especially claims about derivations). And we're currently wrapping up a short piece on competence and performance. All available at

      For a broader sampling of the recent research on real-time models of grammatical computation you could take a look at the ongoing Research Topic on "Encoding and navigating linguistic representations in memory", hosted by Frontiers in Psychology. 20 articles published already, and many more to appear in the coming months.
      In that line of work, one doesn't find a whole lot of explicit discussion of the relation between grammatical theories and theories of real-time processes (various reasons for that), but it's more or less taken for granted at this point that the real-time computations make full and direct use of grammatical constraints. The finding that real-time processes are grammatically sensitive is now so commonplace as to be almost boring.

      Coming from a different tradition, work by Ed Stabler and his students (such as Thomas Graf) takes on the issue from a formal-computational angle.