"The subject-matter of universal grammar"
« previous post | next post »
This semester, John Trueswell has been teaching a seminar focused on Lila Gleitman's recently-published collected works, Sentence First, Arguments Afterward: Essays in Language and Learning. Last week, the paper under discussion was Cynthia Fisher, Henry Gleitman, & Lila Gleitman, "On the semantic content of subcategorization frames", Cognitive Psychology, 1991. The start of its abstract:
This paper investigates relations between the meanings of verbs and the syntactic structures in which they appear. This investigation is motivated by the enigmas as to how children discover verb meanings. Well-known problems with unconstrained induction of word meanings from observation of world circumstances suggest that additional constraints or sources of information are required. If there exist strong and reliable parallels between the structural and semantic properties of verbs, then an additional source of information about verb meanings is reliably present in each verb’s linguistic context.
This reminded me of a passage from a much older work — John Stuart Mill, Inaugural Address Delivered to the University of St. Andrews, February 1, 1867:
Consider for a moment what grammar is. It is the most elementary part of logic. It is the beginning of the analysis of the thinking process. The principles and rules of grammar are the means by which the forms of language are made to correspond with the universal forms of thought. The distinctions between the various parts of speech, between the cases of nouns, the moods and tenses of verbs, the functions of particles, are distinctions in thought, not merely in words. Single nouns and verbs express objects and events, many of which can be cognized by the senses: but the modes of putting nouns and verbs together, express the relations of objects and events, which can be cognized only by the intellect; and each different mode corresponds to a different relation. The structure of every sentence is a lesson in logic. The various rules of syntax oblige us to distinguish between the subject and predicate of a proposition, between the agent, the action, and the thing acted upon; to mark when an idea is intended to modify or qualify, or merely to unite with, some other idea; what assertions are categorical, what only conditional; whether the intention is to express similarity or contrast, to make a plurality of assertions conjunctively or disjunctively; what portions of a sentence, though grammatically complete within themselves, are mere members or subordinate parts of the assertion made by the entire sentence. Such things form the subject-matter of universal grammar […]
I cut the passage off at that point because Mill then continues with his argument for the key educational role for Latin and Greek, because "no modern European language is so valuable a discipline to the intellect as those of Greece and Rome". Whether or not that's true, it's a distraction from the shared idea that "the principles and rules of grammar are the means by which the forms of language are made to correspond with the universal forms of thought."
Richard said,
October 8, 2020 @ 3:55 pm
This made me wonder if universal forms of thought exist and how you would establish that they do exist and then to wonder how a form of thought might be different from a thought.
Rick Rubenstein said,
October 8, 2020 @ 4:26 pm
I'll use this as a moderately decent segue to request a post on the astoundingly good grammar (in the linguistic sense) of the GPT-2 and -3 text generation AIs. For example, I've been following https://twitter.com/StarTrekAI (curated by Ryan North of Dinosaur Comics fame), where GPT-3 generates blurbs for made-up Star Trek episodes. Naturally they're hilarious; but I'm amazed at how utterly coherent they are, with fairly complex embeddings, consistent pronoun referents across multiple sentences, correct use of punctuation like dashes, etc., and hardly any errors to speak of. This is quite a leap from earlier language generators. Sometimes it's really hard to remember that GPT-3 doesn't actually "understand" what it's writing.
I'm also wondering if any similar leaps are forthcoming on the parsing side, which I haven't heard much about. (I don't follow the academic literature, though; my main source of what's up in AI research is Janelle Shane's AI Weirdness blog: https://aiweirdness.com)
Rick Rubenstein said,
October 8, 2020 @ 4:28 pm
Dammit, I knew I'd screw up the markup somewhere. I pine for editable posts. Kill the italics between "astoundingly" and "really".
Jerry Friedman said,
October 8, 2020 @ 8:01 pm
I suppose the (very strong?) (naive?) Whorfian claim is that what Mill thought were the universal forms of thought were actually dictated by the grammars of the languages he knew and the rules of logic derived from those grammars.
But I'd have thought Fisher, Gleitman, and Gleitman were only investigating whether those strong and reliable parallels existed in a given language and could help children acquire that language, not whether the parallels were universal. As may be obvious, I know nothing about their work (or most of linguistics).
Jerry Friedman said,
October 8, 2020 @ 8:03 pm
*derived from those grammars.
I pine with Rick—though I suppose careful proofreading would be another possibility for me.
JPL said,
October 13, 2020 @ 5:24 am
"The principles and rules of grammar are the means by which the forms of language are made to correspond with the universal forms of thought."
"How are the structures of sentences related to the propositions they encode?" (From the Fisher, Gleitman and Gleitman paper)
Two decent responses to the question "what is the domain of morphosyntax?" (BTW, I wonder, was Mill thinking of the "universal forms of thought" in the way Boole was conceiving them at that time?)
In any case, your juxtaposition of these texts put me in mind of something Hilary Putnam said, referring to formal language practice: he said, "mathematicians point to the inscription and think the meaning." This expression separates clearly the dual aspects of the act of language use, whose basic logical unit is the sentence: the part that is perceived and the part that is understood. Even in linguistics, talk of sentences is always infused with talk of meaning. The analysis of what was interpreted as the constituent structure of sentences by appealing only to the observable facts of distribution was a way of separating off the perceivable aspect of the utterance of a string of phones or morphs from the unobservable (and thus suspect) "meaning" that analysts who understood those forms couldn't help including when talking of sentences. But why do sentences have this type of pattern of distributional facts, rather than some other type? The existence of this type of form and not some other one is, it seems, dependent on something else other than the string of phones or morphs. So where does this form, the hierarchical inclusion structure with internal differentiation, come from? Given a "thought", if you will, there is a difference between the expression of that thought (in a sentence, using the available morphs and ordering, etc.), and the description of that thought as an object of inquiry. (BTW, I'm using "thought" in the sense of Frege, something objective and abstract, not in the psychological sense.) Focus on the sentence tends to involve description of the expression of the thought, but, as opposed to the role of the inscription, I want to know what is thought in the thinking, I want a description of the thought expressed. So, is there an internal form of the proposition that is distinct and different in its principles of construction (not just order; perhaps, e.g., a structure not describable by the principle of compositionality) from the form of the sentence, where description of the latter is restricted to, say, a morph by morph gloss of the categories indicated? And, given two sentences, s1, s2, expressing slightly different meanings, what is it that makes possible the differences in the meanings expressed, i.e., the creation of the thoughts, from the speaker's point of view? How is the meaning "done"?