Friday, October 28, 2005

semantics syntax interface

OK, only the one linguistics reader I have will follow this, but I want to document my thoughts. It's 3:15 in the morning, and I have been stewing on this problem for almost two hours now. So, let's see. I am in a class on nativism in language acquisition, which is the argument that children at birth have certain language-specific knowledge, classically about grammar, and stunningly detailed. My prof has authored an argument designed to prove that this is the case. In Swahili, the verbs are marked with Tense morphemes, as well as mood, agreement, and other stuff. Children not infrequently, when learning their native Swahili, drop the tense marker. Nothing too exciting here, as children make all sorts of errors. What is interesting is that when they drop the tense marker they almost always drop the subject as well. If they don't drop the tense marker, they may or may not drop the subject, but if the tense marker is absent then without exception the subject is absent as well. Why?

The nativist explanation is that in generative syntax, the syntactic feature Tense licenses the subject. The exact reasons do not seem important. What is important is that these children are very young, 2-3, I think, and if you look at all the other, seemingly related patterns in Swahili, none of them seem to be viable models from which the child could learn to make this pattern of error. So, if the child can't learn it, then it is innate knowledge they are born with. Somehow, in their head, is the rule that if a sentence is -Tense, then you cannot have a Subject.

Now, the goal of the students in the class is to prove that nativism is a load of crap, so I've been puzzling on this for a few weeks now. I have to prove this argument is incorrect in some way. There are only a few possibilities. First, I cannot use syntax to solve the problem really, as the whole purpose of the argument is to show that kids are born innately knowing syntax. I lose no matter what the syntactic explanation is. The only way syntax is useful to me is if I can show that some piece of syntax, which is learned, can handle it. It's possible, in theory, that I could do that, but there would be problems. One is that if you alter the syntax too greatly from generative stuff, then the nativists have already stopped listening and you persuade no one. The second is the practical reason that I'd have to re-engineer a whole new syntax in my spare time.

So instead I have been trying to tweak either syntax or how syntax relates to other parts of language. I am going to assume, for now, that Tense does in fact license subjects. The goal then will be to try and figure out why this is the case, not relying on the syntactic theory (Case basically for anyone still reading). So, the most promising way would seem to have to deal with semantics (or pragmatics or discourse). There is something about the meaning of Tense that ties it to subjects when it gets sent into the syntactic module. In other words, meaning has structure too and when you run what you want to say (semantics) through how you want to say it (syntax), you end up with a link between Tense and Subjects. And it is this that the children have learned which causes the subject drop behavior in Swahili.

Nice theory, but does it work? Not yet. The bad news is that I know virtually nothing about semantics, so I have been trying to bone up on it in my great spare time, yet I have made a little headway. First, Tense is anaphoric in meaning. In a sentence like "If John comes to a party, Sue immediately leaves," when does Sue leave? She leaves right after John shows up. The time for her leaving is bound to the time that John arrives. This is what it means to be anaphoric. The time for Sue's departure can only be determined with reference to the time of John's arrival. This is similar to pronouns. Who "he" refers to can only be determined by a previous noun phrase that is indexed with the pronoun. Now, there is a huge literature on the concept of bound variables within syntax, starting with pronouns. They are called binding principles and they govern just how pronouns relate to their antecedents (loosely). But there is no such discussion that I have yet seen about Tense as a bound variable.

Why would I care if Tense is a bound variable? If a sentence has tense, then it is bound by something. It can't be bound in the syntax without really enhancing the Tense features, as they just don't handle the complexities of the meaning of Time. I am going to say it is bound in the semantics. That's my link from syntax to semantics that I have been looking for. Next, if something has no Tense, then is it bound? I am going to say, 'no', it is a free expression, like a referring noun phrase (any noun not a pronoun basically; you know what "tree" refers to without a previous antecedent to explain it.) This all has some potential because subjects, in the generative tradition, are supposed to be found in the Specifier position of the Inflectional Phrase. Interesting connection that subjects in particular are in the specifier of IP and nothing else. So what I am hoping for is that if I can give some content to the notion that -Tense is free, while +T is bound, that it will behave differently with regard to subjects.

And that's the theory. It's now 4:00 AM, and I'm really going to regret this in about 2 hours when B wakes up. I wish puzzles like this didn't get me so stoked, because I'm still rather wired and I need to sleep.


Charles Nelson said...

1. Why can't general cognitive processes work as well as language specific processes?

2. Assuming that all cognitive processes have the same beginning, then language specific processes emerge from general cognitive processes. How, if it does, this affect the nativist position?

pacatrue said...

1) I think all linguists would agree with your idea. There are certainly general cognitive mechanisms, such as associative learning, basic auditory analysis, motor skills in articulation, etc., that play a role in language. I think even the most hardened nativist would say this is true. What makes him or her a nativist, though, is that they say there are other language-specifc processes which are also at play. These would be any universal constraints on your phonology and, especially, the rules of generative grammar, such as phrase structure and rules on the binding of pronouns (among many others). The nativist claims that these extra bits are innate and specified from birth. A child's grammar is pretty much the same as an adult's. The child's job is "merely" to figure out how to set the small number of parameters that make languages different. One difference - apart from just this assertion of innate knowledge - is that a nativist thinks that this language-specific stuff is really the interesting thing. It's that which we want to explain and understand and the general things are sort of peripheral items which we can bother to look at in our spare time. The anti-nativist often thinks that it is the general processes which are of note, and we only fall back on innate knowledge, when the general processes fail.

2) I think the nativist would disagree immediately that all cognitive processes have the same beginning. Perhaps the biggest school of "anti-nativism" is the connectionist modeling group who specifically try to do what you suggest. They start with a brain that has a specific architecture, and maybe a slight bias on development, and then try to grow all of the rules of adult grammar. So even though the endpoint might be the same as the nativists want, they are considered anti-nativist specifically because the grammar grows based on architecture and learning, and not on genetically specified knowledge.

Now, is a nativist stuck with this? It's not clear they really need to be. All they really need to say is that the biology constructs the knowledge to develop in such and such a way apart from much input. That last part is critical, as the main argument for nativism is that children behave in a certain way - the whole dropping subjects when they drop tense thing - exactly in the absence of any input from which to learn this behavior. So in the end, the difference between the two positions is really one of "what role does learning play" in language acquisition? The radical nativist says almost none. The radical anti-nativist says almost everything is learning.