Monday, October 27, 2014

Some naive thoughts on intuition

Much of the last two weeks has been spent trying to get my head around what a head is. A representative definition might be.
A head is the syntactic function of the dominant constituent within a phrase to which other constituents are subordinate or a two-word dependency relationship. “In the simplest account, other units depend on it” (Matthews 2007:191).
My starting point was Zwicky (1985), for no better reason than that’s what I stumbled on first. And this was where the issue of intuition came up. Zwicky writes,
The intuition to be captured with the notion HEAD is that in certain syntactic constructs one constituent in some sense ‘characterizes’ or ‘dominates’ the whole. (2)
On this first meeting with the word intuition, I can’t say I consciously noticed it, but throughout the paper, intuition is mentioned again and again. The first mention seems innocuous, but later, it is framed less as inspiration and more as evidence, as a compass that helps us find our bearings.
The intuitively correct assignment of certain items as arguments and others as functors can be guaranteed only if substantive assumptions about the relationship of semantics and syntax are made, as in Klein and Sag (to appear). (4n)
Both government and concord constitute arenas in which one constituent in a construct can intuitively be said to 'dominate' another. (7)
Intuitively, the difference is that subcategorization concerns the very possibility of one constituent's combining with some other co-constituent(s), while government concerns the form that a co-constituent has in such a combination. (7)
Again speaking intuitively, in both phenomena morphosyntactic features of one constituent can determine the morphosyntactic features of a sister constituent. (7)
The assignment of V and P as the governors in V + NP and P + NP is intuitively correct and has not been disputed to my knowledge. However, many have had reservations about saying that VP governs NP; it is somewhat counterintuitive to say that the VP 'causes' nominative marking on its NP subject. (8)
In truth, I didn’t think about all this talk of intuition much until I read Hudson’s response to Zwicky (1987), in which he begins by disagreeing, and justifies his disagreement with an appeal to intuition.
However crude it may be, I find this description very insightful, and I think it corresponds closely to the intuition that underlies many of the dependency analyses with which I am familiar. (113)
Writing almost a decade later, Croft takes up the discussion, arguing that Zwicky had the right criteria but misinterpreted them. He also picks up on the theme of intuition, and goes further than the first two writers, claiming that “the criterion [for establishing headship] should be founded on some sort of pretheoretical intuition that the criterion can be argued to reasonably capture” (1996:49). No justification for this claim is forthcoming. It’s simply stated as if it were obvious. But it seems to be an even less defensible position than that of Croft or Hudson: not just saying that intuitive appeal is evidence of correctness, but that it should be the foundation on which any description of heads is build.
None of the three writers elaborates on their ideas of what intuitions are or how they should be used in argumentation. It turns out that philosophers see it in quite a variety of ways (Pust 2012), and it’s not entirely clear to me which of these is intended, or if the authors are even thinking of the same concept. Overall, I found it somewhat puzzling, so I thought I’d dig into it a bit more.
My intention is not to argue that intuitions should be off the table. Clearly, they are often the motivation to examine a phenomenon[1] and, indeed, it may be impossible to avoid bringing them into the mix[2]. But if they are inevitable, then it is trivial to say, as Croft does, that they should be behind a theory. And if they are avoidable, it’s not at all clear why one should strive to build a theory on them. In fact, it seems that most philosophers of science today feel that a theory’s origins are of no consequence. “All that matters is how the theory stands with respect to the things it explains and predicts” (Curd & Cover 1998:410). So the idea of head would be useful if it explains something in language. Perhaps it could explain our intuition, but I don’t see why it would have to.
On a unificationist view, explanation involves bringing together a diverse range of phenomena under a single principle, or at least as few as possible (Godfrey-Smith 2009:196). This could simply mean finding a number of syntactic relationships that tend to act in predictable ways in relation to a particular set of words and calling this phenomenon head[3]. If the result is counterintuitive, so be it. This seems to be more what Cann has in mind (1993:47)[4]. It is also the basis for a variety of approaches in science, factor analysis being one of them (Rummel 1988)[5].
This doesn’t mean that the “shut-up-and-calculate[6]” faction has carried the day. If the result is intuitive, of course, it may face fewer objection of the sort that Einstein raised against quantum theory with his well-known claim that God doesn’t play dice (Hermanns & Einstein 1983:58), but it is the match with the evidence, not the match with intuition that matters. Any appeal, whether it be to intuition or to the outcomes of mathematical formulae or computational models, is pointless unless the process can be shared and evaluated. The assumptions behind all computations and their results need to be checked against the evidence, whether they’re on paper, in computers, or in vivo.
Intuition, even linguistic intuition, is merely one kind of evidence which must match with other evidence before it is given any credence (Devitt 2006). It is known to be susceptible to framing problems (Tversky & Kahneman 1986), confirmation bias (Morewedge & Kahneman 2010; but see Klayman & Ha 1987 for a more sanguine view), cognitive limitations (Clark 1996), and other weaknesses. This is because, as Heath puts it,
Intuitions are typically the result of heuristic problem-solving processes, rigidly adapted to deal with particular challenges that arose in the environment of evolutionary adaptation. Not only do they contain bugs, but they also tend to misfire when deployed in non-standard environments. (2012:97)
Linguists have begun to realize that intuition about grammaticality is suspect (Wasow & Arnold 2005), and there is no reason to think that our intuitions about syntactic relations should be any better. In fact, since this is something that we have not evolved to do, we might expect them to be much, much worse. And, as experts, we may be even more susceptible to overconfidence (Kahneman 2011:219).
Neither Zwicky nor Croft is explicit about whose intuitions are at issue. Presumably a layperson has no particular position on the question. My own intuition (admittedly) is that non-linguists would be very inconsistent in answering to the question: what is the head of this clause, or even the less technical what is the most important word in this sentence. Certainly, there is good evidence that statistical intuition is highly unreliable among non-statisticians and even among experts (Kahneman & Klein 2009). So this leaves the intuition of the syntacticians, but even here we find much disagreement, both synchronically and diachronically (Matthews 2007).
An important distinction needs to be drawn, then, between the use of intuition in the context of discovery, and its use in the context of justification (Bowers et al. 1990). Clearly, starting with intuition and finding that the data supports it is wonderful and rewarding; intuition can be helpful. It’s quite different from starting with intuition and finding the data to support it. But I see nothing wrong with starting with no clear theory or expectation, finding a cluster—apparently mere correlation—and deciding to label it for convenient reference head. If this cluster applies to a wide range of data, then it can be said to be explanatory in a unificationist kind of way, and intuition really has no bearing on it one way or the other.

Bowers, Kenneth S, Glenn Regehr, Claude Balthazard & Kevin Parker. 1990. Intuition in the context of discovery. Cognitive Psychology 22(1). 72–110. doi:10.1016/0010-0285(90)90004-N.
Cann, Ronnie. 1993. Patterns of headedness. In Greville G. Corbett, Norman M. Fraser & Scott McGlashan (eds.), Heads in grammatical theory, 44–72. Cambridge: Cambridge University Press. doi:10.1017/CBO9780511659454.
Clark, Andy. 1996. Economic reason: The interprlay of individual learning and external structure. In J Drobak & J Nye (eds.), The frontiers o fthe new institutional economics, 269–290. San Diego, CA: Academic Press.
Croft, William A. 1996. What’s a head? In Johan Rooryck & Laurie Zaring (eds.), Phrase structure and the lexicon, 35–75. Springer Netherlands.
Curd, Martin & Jan A Cover (eds.). 1998. Philosophy of science: The central issues. New York: W. W. Norton.
Devitt, Michael. 2006. Intuitions. Ignorance of Language. Oxford University Press. doi:10.1093/0199250960.001.0001.
Godfrey-Smith, Peter. 2009. Theory and Reality: An Introduction to the Philosophy of Science. Chicago: University of Chicago Press.
Harris, Zellig S. 1946. From morpheme to utterance. Language 22(3). 161–183. doi:10.2307/410205.
Heath, Joseph. 2012. Letting the World In: Empirical Approaches to Ethics. Les ateliers de l’éthique 7(3). Centre de recherche en éthique de l’Université de Montréal. 93. doi:10.7202/1014387ar.
Hermanns, William & Albert Einstein. 1983. Einstein and the Poet: In Search of the Cosmic Man. Branden Press.
Holmes, Richard. 2009. The age of Wwonder: How the romantic generation discovered the beauty and terror of science. HarperCollins Publishers.
Hudson, Richard A. 1987. Zwicky on heads. Journal of Linguistics 23(1). 109–132.
Kahneman, Daniel. 2011. Thinking, fast and slow. London: Penguin.
Kahneman, Daniel & Gary Klein. 2009. Conditions for intuitive expertise: a failure to disagree. The American psychologist 64(6). 515–26. doi:10.1037/a0016755.
Klayman, Joshua & Young-won Ha. 1987. Confirmation, disconfirmation, and information in hypothesis testing. Psychological Review 94(2). 211–228.
Matthews, Peter Hugoe. 2007. Syntactic relations: A critical survey. Cambridge: Cambridge University Press.
Mermin, David N. 1989. What’s wrong with this pillow? Physics Today 42(4). American Institute of Physics. 9. doi:10.1063/1.2810963.
Morewedge, Carey K & Daniel Kahneman. 2010. Associative processes in intuitive judgment. Trends in cognitive sciences 14(10). 435–40. doi:10.1016/j.tics.2010.07.004.
Pust, Joel. 2012. Intuition. Stanford encyclopedia of philosophy. (4 October, 2014).
Rummel, Rudolf J. 1988. Applied factor analysis. Northwestern University Press.
Tversky, Amos & Daniel Kahneman. 1986. Rational choice and the framing of decisions. Journal of Business 54(4). 251–278.
Wasow, Thomas & Jennifer Arnold. 2005. Intuitions in linguistic argumentation. Lingua 115(11). 1481–1496. doi:10.1016/j.lingua.2004.07.001.
Zwicky, Arnold M. 1985. Heads. Journal of Linguistics 21(1). 1–29.
Zwicky, Arnold M. 2007. Labels are not definitions. Language log. (14 October, 2014).

[1] See Holmes (2009) for a wide range of enlightenment examples.
[2] But with the complicated nature of much science, the difficult mathematics, and the prevalence of unobservable phenomena such as quarks, and mental representations, the types of intuitions may have changed over the years (Curd & Cover 1998).
[3] Of course, you might wish to call it giraffe or B21. As Zwicky points out (2007) labels aren’t definitions.
[4] Once Cann gets into the later parts of the article with the Chomskian arguments, I wasn’t able to follow it. Perhaps this data-based method is also similar to what Harris (1946) suggested? I haven’t yet read Harris, so I don’t know.
[5] My understanding of factor analysis is extremely superficial. Interestingly, Rummel notes that in the past, an intuitive process of “rotating the factors” (18) after factor analysis was not uncommon, but claims that “mathematical solutions of the rotation problem and the availability of high-speed computers have largely done away with this possible source of arbitrariness” (18).
[6] This is how David Mermin (1989) characterizes the Copenhagen school’s approach to quantum physics.

No comments:

Post a Comment