Ontological status of syntactic transformations

Question

Syntactic theories in the generative tradition involve transformations, i.e. movement of constituents, between deep structure and surface structure. What is actually meant by this? Is it intended as a model of the linguistic mind, a metaphor, or something else?

For example, take the statement "English passive clauses are derived from underlying active clauses by a set of transformations". I can think of two things this might mean:

1. It might be intended as an actual model of the linguistic mind. That is, the claim would be that when you form a passive clause, your brain first puts together the corresponding active clause, then applies the transformations to produce a passive clause out of it.

2. It might not be intended as a model of the linguistic mind, but simply as a formal system, i.e. a set of algorithms that can produce linguistic forms that correspond to those that actual speakers produce, but without necessarily going through similar processes.

If 1 is intended, what evidence is there that it's true?

If 2 is intended, what is the point of such a set of algorithms, when what we really want is to understand the actual workings of the linguistic mind?

Answer 1

OK, this is an Answer, not a book, and conversations are inhibited here.
So I'm going to observe some restrictions to keep this relatively short.

1. I'm not dealing with Chomsky here; I find Chomsky's proposals and theories about language and its supposed relation to human brains, minds -- and lately biology and genetics -- to be at best irrelevant, and at worst embarrassingly ridiculous. Either way, what he might say, or has said, is of no interest to me. For details, see The Chomskybot.
2. I'm restricting myself to a discussion of syntactic rules as they are developed in McCawley 1998, which is pretty basic and non-formal (formality in syntactic theory is only justifiable to the extent it makes theories machine-washable, and that is not the issue here).

That said, here's the presenting question:

• What is the point of 'a set of algorithms that can produce linguistic forms that correspond to those that actual speakers produce', when what we really want is to understand the actual workings of the linguistic mind? [all presuppositions by the original poster]

Well, there are an awful lot of different sentences in any language; if not an infinite number, at least an extremely large number, especially if you allow recursion without limit.
How does one describe them all? No matter when we start, we'll never finish.
That's not helpful.

It's the same problem with anything alive; there are a lot of worms, for instance.
How do we describe them? By noting similarities and differences, and positing clades, like species.
Same thing in syntax. We look for similarities and differences, and posit clades.

For instance, take the Passive construction in English.
I call it a Construction, but it could be called a Rule just as well, or an Alternation, though it's not cyclic. You have asked "Why say the active is basic and the passive derives from it?"

Look at the data. Assuming one can recognize Passive sentences and distinguish them from Active sentences, one will soon discover that for every Passive sentence there is a corresponding Active sentence -- perhaps with an indefinite subject when the Passive sentence lacks a by-phrase -- but that this is not true in reverse.
The following Active sentences, for instance, have no corresponding Passive sentences:

• Bill is sick ~ Mary ran yesterday ~ It's all over now. ~ He's being honest.

That is, there are Active sentences with no corresponding Passives, but there are no Passives with no corresponding Actives. This is a fact, and if grammar is about facts, it ought to be able to state this fact. One way to do this is to consider Passive a special effect that can be applied to most transitive sentences, but not to intransitive sentences.

Another point to make is that corresponding Active/Passive sentences mean the same thing.

• Bill ate the cake = The cake was eaten by Bill.
• Mary kicked Bill's ass. = Bill's ass was kicked by Mary.

These are synonymous (if one is true, so is the other, and if one is false, so is the other),
even though they look and sound different. So the meaning of the Passive is predictable from its corresponding Active. Furthermore, the form of the Passive is also predictable from its corresponding Active, by the following "algorithm":

a) Promote Object of main verb to Subject
b) Add auxiliary be
c) Follow be immediately with the past participle of main verb
d) If desired, express former Subject as object of by-phrase.

This allows the form and the meaning of passive sentences to be described, simply, by means of reference to the corresponding active sentence. And, by unwinding, to produce a corresponding active from a passive. This cuts the number of sentences that have to be described way down, because now Passives can be completely specified by reference to the prototype Active. And every time you discover another rule, you simplify the prototype that much more. That's the rationale, basically.

Surely any scientific theory of grammar should be able to describe such a simple correlation between simple sentences in a simple way. This is one such way. If you've got a neater way, tell us about it. And Passive is just one rule of several hundred in English, and this way works for all of them.

As for the tag in the question ...

when what we really want is to understand the actual workings of the linguistic mind

Who's this "we" you refer to?

There is no evidence that there is such a thing as "the linguistic mind" as a separate or separable entity (even ignoring problems with the ontological status of mind), except in the imaginations of theorists.

There is not even any evidence that different people speaking the same native language use the same internal syntactic processes. Much less all humans speaking all languages. Forget about "the linguistic mind" and "universal grammar"; it would be nice to be have telepathic X-ray syntax vision, but not even Superman has that.

Real linguists just have data to look at. And the data shows patterns. And the patterns can be viewed as algorithms, or as databases, or as formatting for ideas, or as social activity, or as cultural conventions, or as a lot of things. But you have to be able to state them somehow, and get the job done before the heat death of the universe.

Answer 2

The question is excellent. I would like to answer the question as well, adding an overlooked aspect of the issue.

Some derivationalists will likely defend the first interpretation of syntactic movement listed in the question, stating that syntactic movement (or copying) is a real condition of the human faculty for language. Humans have an innate ability to move syntactic units around in their "minds". Thus syntactic structures are generated in a bottom-up fashion, whereby certain constituents first appear in one position and are then moved to higher positions, and this all occurs in the mind before the actual structure is uttered by the speaker.

Other syntacticians and grammarians (including myself) reject this conception of how humans produce the syntactic structures of utterances. For me, a more plausible understanding of how humans produce utterances is that they access constructions in a lexicon (or constructicon) in an online fashion as they actually speak utterances. On this sort of account, the syntactic structures of utterances are produced left to right, not bottom up. They are produced left to right moving in the same direction that our speech organs can produce sounds and in the same direction that our comprehension absorbs meaning. This sort of account rejects the first interpretation of syntactic movement in the question, although it does not necessarily agree with the second.

A difficulty facing those syntacticians (like myself) who reject the innateness hypothesis of Chomskyan syntax is the fact that much of the terminology associated with the study of syntax was established in the 1960s and 1970s, at a time when Transformational Syntax (think deep and surface structure) were the main game in town. Thus since the terminology that we use today was established at that time, the presuppositions that that terminology bears remain intact in the field of syntax in general. Terms such as extraposition, extraction, scrambling, fronting, topicalization, shifting, etc. all carry the presupposition of movement.

If a non-derivationalist wants to examine, for instance, the phenomenon denoted by the term extraposition, he/she is faced with a dilemma. He/she can use the term extraposition (which presupposes that something has been "extraposed") so that his/her audience knows immediately the phenomenon that is meant, but in doing so, he/she has to accept that the term itself expresses the unwanted presupposition of movement. Or he/she can avoid this unwanted presupposition by choosing another term, but this option has the drawback that introducing unknown terminology is difficult. Doing so requires more patience of the reader/listener, a fact that will likely reduce the size of the audience.

What I am describing here is a situation that perpetuates the transformationalists' interpretation of syntax. The transformationalist has a major advantage, since if the non-transformationalist wants to be understood, he/she is almost forced to employ a terminology that actually supports the transformationalists' stance.

The long and the short of all this is that many more syntacticians, grammarians, and linguists would likely choose the second option listed in the question than the first, and there are others like myself who completely reject the first and distance themselves from the second, choosing instead a third interpretation which sees "syntactic movement" and "transformations" as constructions that are stored in the lexicon/constructicon.

Answer 3

As a minor technical correction, syntactic theories in the generative tradition do not involve transformations, they used to involve transformations, 50 years ago. Current generative syntax is very different, so the question is whether the query is about old-style syntax, or about generative syntax in general, abstracting over all instantiations. The one thing that remains constant is that “it is a model of the mind”, and it is not just a model of the data.

The same question can be asked about all theories of the mind or the universe – what evidence is there that the mind actually works the way that mind-scientists in that domain claim? One approach is to look for physical evidence, for example fMRI or EEG data. But that beg an important question about the relationship of brain to mind, which are not the same thing. The mind is not a physical object that can be inspected, so you can just as well ask “what is the evidence that this particular brain pattern is homologous to the actions of the mind?”. In the history of physics, one could (and did) ask question like “what is the evidence for these invisible ‘cell’ objects that you claim my dog is composed of?”, or we can ask about the evidence for molecules, atoms, particles etc. Relatively little of the universe can be directly seen with our eyes, but we have a good model of light that enables us to claim that we have “seen” cells in a microscope, and so on. The observables are a large part of the evidence.

The mentalist claims of GG are, in fact, not hypotheses, they are axiomatic stances that define the domain of discourse within which testable hypotheses can be made and evaluated. Whatever mechanisms constitute the faculty of language, they are by definition aspects of the mind. In contemporary GG, the closest that one can come to advancing an empirical claim directly addressing the mind is the existential claim that there exists a language faculty. There is a competing view that there is no language faculty, that apparent linguistic behavior is merely the result of generalized mental faculties such as “analogy”, and that language and painting are basically the same thing.

The mentalism claim imposes certain limits on what can be a model of language. For example, we can a priori reject a model of language that claims that all utterances of a language are stored from direct experience and directly retrieved when needed to produce or parse utterances (hopefully it is obvious why this is not a possible model of language). However, it is a possible Platonic model of the infinite set that constitutes “a language”. In fact, reified infinitudes were conceptual problems for the theory of metarules (in SPE and HPSG) and still-extant Optimality Theory.

The “what’s the point” question surrounding understanding the actual workings of the linguistic mind, incorrectly presupposes what we really want, although I do really want that. The “model of the data” view has always been very popular in linguistics, despite the mentalist stance taken by GG. The reason for diverging views of the mental-reality question is that the mind is not directly observable (a dog is directly observable), and we individually have to find a point on the evidentiary scale where we will say that we “know” something. One extreme position is epistemological nihilism, where one cannot claim to “know” anything – this is an actual position (held by Pyrrho and Berkeley, for example). Another imaginable extreme position (not actually held by anybody, ever) is that if you are aware of any evidence for a claim, you know that the claim is true. A related position comes perilously close to the position held by some generative grammarians, that any claim supported by uncontradicted observational evidence constitutes knowledge. I don’t intend to go into a long digression on epistemology here, but I find that the main split in working grammatical-theoreticians is between those holding an agnostic stance versus a gnostic stance, because of differences in what constitutes sufficient evidence, plus, whether you take “model of the mind” to be a defining desideratum of theory-formation. If you don't care about model-of-the-mind, you can have infinite storage in grammar.

I probably disagree with hmltn over the fundamental philosophical question about “real” and “theory” (we'll see if there's a "real" difference). What is “real” is that which exists. There is nothing more that you can say about “real”. Next we move to the epistemological domain, which is the relationship between that which exists and a mind. Asking whether a theory is “real” misses the point, that knowledge presupposes existence, and the right question to ask is “how should we evaluate this proposition”, i.e. does it constitute knowledge, suspicion, conjecture or imagination? A theory is never an approximation (or, should never be), it always says something precise. However a theory can always be evaluated on an evidentiary scale, so that we can compare theory A and theory B to see which is better-supported.

The valid sense of “is an approximation” that exists in linguistics is about the entity that exists. 70 or so years ago, we had only a vague idea what thing we were talking about when we talked about “sounds” in language, the same goes for “meaning”, “syntax” and “language”. Nowadays, we have to resort to auxiliary expressions to clarify whether we are speaking of “entailment” versus “pragmatic inference” when we talk of meaning, or for syntax are we just talking about constituency versus linear order? Progress has been made in linguistics when we narrow the claim by clarifying what “syntax” or “sound” means, leading to domain-splits (phonetics and phonology are completely different things, contrary to the original generative view).

Answer 4

The ontological status of theories of syntax is information, which is virtual. By virtual, I mean that information is a pattern that has to be carried out in some real-world thing, in order to exist, but it doesn’t matter what real-world thing is used for that purpose.

For example, sound travels as waves - vibrations in particles - through the air, to your ears, where they vibrate the eardrums, triggering some electrical current into your brain, constructing a conscious experience of a phenomenon.

However, a new technology called bone conduction, has led to the creation of headphones that produce sound (in your head), without vibrating air particles - instead, the wave patterns we would see, virtually, on the screen of a sonograph, are passed through cochlear bone, from where the eardrums pick them up, then routing them to the brain in the same way.

A cochlear implant goes yet further, and is a “neuroprosthesis” producing currents directly into the auditory nerve, and from there the nervous system.

---

I’m not qualified to answer this question robustly as the others did but I did want to lend my voice to a view on the question I think may be different from the other two answers. Or if it’s not, it would be at least shorter, for someone who may want a more elementary-level commentary.

In my opinion, there is not a true distinction between the two options posed:

• an actual model of the human mind
• a formal system that can produce the expressions of language, but not to be taken as the actual procedure by which the mind does so

When you ask “what is the ontological status of this thing”, meaning, what are the characteristics of its existence, what type of “thing” is it truly, we have to import the topic of ontology and ask what kinds of things one thinks there are, in general - for example, do you think things in the world divide into:

• mental or physical?
• virtual or actual?
• hypothetical or real?
• contingent or necessary?
• possible or impossible?
• exact or approximate?
• literal or metaphorical?

and so on.

Your question - a good one - is asking about if the “Platonic” forms of logical theories (of language), have real existence. But it’s not a question confined to linguistics, but a broad, philosophical one: what is real? And what is “real”?

---

A theory is always an approximation, until it gets more and more refined. If current syntax can come close to reproducing the forms of human sentences, but is not the procedure by which humans genuinely produce language, then it isn’t a good theory, and the theory’s not finished yet. It means it’s a work in progress; but in final form, it would indeed chart out those exact processes.

When that occurs, there would be no difference between an abstract, logical theory of syntax, and a physical, “actual” one. The point is to deduce the nature and functioning of a system. The rules of the system can be written down, identified in symbols and concepts humans use, to spell out patterns and relationships between things. But the typographical, symbolic expression only maps to a real-world, physical process - perhaps patterns in the brain, in electrons firing, in cells, in voltages, and amplitudes in sound waves, and light particles and so on. It wouldn’t matter, because as the approximate model gets more and more fine-grained, the distinction between real and non-real theory dissipates: there’s only information which describes something. (See Wittgenstein’s “picture theory of meaning”).

---

This viewpoint can apparently be called “structural realism”. According to the SEP:

Structural realism is considered by many realists and antirealists alike as the most defensible form of scientific realism. There are different forms of structural realism, and an extensive literature about their pros and cons, and how they relate to case studies from science and its history.