Lg Acq & Ling Thy |
A weblog for GRS LX 700
|
I intend to continue with my office hours on the same schedule as they were during the semester over the upcoming week (Monday 11-12, Tuesday 11-12, Thursday 3-4), although I also expect to be in my office much of the time next week and it’s no problem at all to arrange meetings outside of those time slots.
And beyond next week, I’ll still probably be around much of the time, too, at least until about 12/21.
I have a conflict with my normal office hours time on Monday 12/5, so they’ll be from 1-2pm instead on that day.
If you’re going to be doing an experiment with actual people (as opposed to working with a corpus like CHILDES), you’ll need to get your experiment and informed consent form approved by an IRB (Institutional Review Board).
In the past, people in this course have done this through the School of Education, and so let me post the information I have about doing it that way. Bruce Fraser has provided the forms and policy statements (scanned PDF file) as well as some examples of informed consent forms that you might model yours on (PDF, RTF, Word). I’ve also posted these in the “Internet resources” section of the course home page.
What you do is fill out the form and write up something short about the experiment (in broad terms), and submit your informed consent form. It’s mainly the informed consent form that is being evaluated. When it’s ready, you can take it to Bruce Fraser at SED.
The second question mentions that a lot of speakers picked “2″ for both sentences and asks if “2″ is a legitimate choice in either language. The hint I gave you is “no”.
I stand by that, but it’s kind of confusing and it’s probably not worth banging your head against a wall about it. Of course, the DP labeled 2 is an option in both languages in both sentences. The thing is, their choices were “1″, “2″, and “1 and 2″. So, if they picked “2″, that means “only 2″—that is, they disallow “1″.
I hope that clarifies what a response of “2″ is supposed to mean.
Just to clarify something on the homework, in a sentence like
(1) Mary knew that Susan heard Ann’s description of SELF
all three names count as “subjects.” This is a somewhat technical sense of “subject”, since even a possessor in the specifier of a DP will count as a subject. Mary and Susan are the usual kind of subject we think of, the subject of a clause. But Ann is also a subject, being in the specifier of the DP Ann’s description of SELF. Since all three are subjects, it would be possible for a monomorphemic reflexive to be bound from that position (unless there is some additional condition prohibiting binding, such as being separated by a tensed Infl [=T] node).
In (1), SELF can refer to Ann and Susan because they are both subjects and neither is separated from SELF by a tensed Infl. (Susan counts as close enough, because it is inside the TP headed by the tensed Infl.) Mary on the other hand, is too far away, because it is not within the smallest tensed clause containing SELF.
I’m slowly catching up on grading things, my profuse and profound apologies for getting so far behind. Also: if you look at the Courseinfo site and see a “−” instead of a score, that means that I haven’t entered a score yet for that item (not that you’ve gotten no points for it!). I anticipate having everything graded—and comments on the experimental designs as well—back to you on Tuesday, and probably all entered into the Courseinfo site by tomorrow.
Belatedly, let me say a couple of things about the ideas in homework 1 (although I realize it requires remembering a bit far back). This was the one looking at the Legendre et al. (2000) proposal to account for the acquisition of tense and agreement in French, as compared to the Wexler (1998) and Schütze & Wexler (1996) ATOM proposal.
The Wexler (1998) proposal functions pretty much in the same way Optimality Theory does. The part of his proposal that makes it like OT is the Minimize Violations constraint, which basically says to obey as many constraints as you can. Meanwhile, the rest of the constraints, most notably the UCC, set up a conflict such that it is impossible to satisfy all of the constraints (at least in the languages where AgrP requires that the subject check its feature, the non-null-subject languages). So, when I spoke of Wexler’s proposal having three constraints (versus Legendre et al.’s four constraints), I wasn’t even thinking of the Minimize Violations constraint. There are three others, but one of them is barely mentioned at all in Wexler’s paper. One says that Tense must be in the structure, another is the UCC. The third is one that says that Agr must be in the structure. It’s important to have that last one, or else there would not have been any conflict, and Agr would simply be left out all the time (satisfying both the constraint requiring Tense and the UCC).
The four constraints in the Legendre et al. proposal are the three that are basically shared with Wexler’s proposal (Parse T, Parse Agr, *F2), and then one other, *F.
The assumption I made is that Minimize Violations is not a constraint on a par with the others, that it is basically inviolable. One might question that assumption, and if you do allow Minimize Constraints to be violable, then the differences between Wexler’s predictions and our predictions pretty much disappear.
The main difference that I wanted to try to point out is that, if you assume Wexler’s three constraints (Have T, Have Agr, UCC [don’t have both]) in the context of Minimize Violations, then there are only three possibilities (which makes sense, there are only three constraints the kid could opt to violate): Agr is missing, T is missing, or neither is missing. What that means is that it makes the prediction that T and Agr are never both missing at the same time.
On the other hand, this possibility (T and Agr both missing) is predicted by the Legendre et al. system, and in fact relied on, since that is the situation under which an NRF is claimed to appear in French. In the Schütze & Wexler (1996) paper, they do suggest that this situation arises in English, specifically in utterances with a genitive subject, such as My playing. But the ATOM model of Wexler (1998) does not actually predict this would ever occur.
So, that’s really the predictive difference between the two, the existence of structures with neither T nor Agr. Now, if we were to allow Minimize Violations to itself be violated, then perhaps it would be possible. The MV constraint seems as if it has a somewhat different status, and it’s a little bit weird to suppose that it might be violated, but it is nevertheless a possibility. If you violate MV, then you can violate as many of the others as you like, and in particular you could violate both the constraint requiring T and the constraint requiring Agr. And, then, the difference in the predictions about which forms we’d expect to see in the child data goes away. The Legendre et al. system still makes predictions about the frequency at which we’d expect to see missing T, Agr, etc., something Wexler doesn’t attempt to do. On the other hand, Wexler predicts the correlation between non-null-subject languages and the existence of root infinitives in the child language, which Legendre et al. don’t attempt to do.
As for which language, English or French, we might expect to find more root infinitives in, the idea I was relying on when thinking about this is basically that it’s just easier to get a root infinitive in English—if even one of the TP and AgrP projections is missing, you get a root infinitive (at least in the present tense). In French, both have to be missing in order to get a root infinitive. This is a fact about the morphological analysis more than anything else, something which is basically shared by both the Wexler and Legendre et al. proposals.
I got a worried email about the comprehensibility of the Hsu & Hsu article about statistics and methodology, and I want to try to clarify what the goal is here in the one-week crash course in statistics and methodology that we just went through.
Do not worry if you don’t understand everything that Hsu & Hsu are saying, or if you don’t feel confident that you’d be quite up to doing serious statistical analysis of the data you’d end up with at the end of your experiment. They have whole semester-long courses to teach you that, in fact several-semester sequences. A background in statistics wasn’t required for this course, and I have no expectations that anyone who came into the course without statistics background will leave the course an expert. The goal was simply to give you a “feel for statistics” as much as is possible in a couple of readings and a single 3-hour lecture. My hope is that what you will get out of this, if you haven’t seen much statistics before, is an idea of what kinds of things you do with statistics and what sorts of things you’d want to learn someday if your future research requires it. Plus, some basic understanding of what they mean in articles that we read when they say an effect is significant, with a small “p value” (that is, a small chance that the observed effect arose by random chance).
So, if you were reading the Hsu & Hsu article and were overcome by a wave of dread and/or terror, please put the article down, breathe a bit, and be reassured that I was only asking you to make a reasonable attempt at getting the points they are making. Do not go out and buy statistics textbooks and try to teach yourself statistics before the semester ends (unless this simply appeals to you and you wish to do this for reasons of your own).
To the extent that statistics will be useful in the experiments you do and write up, I will try to help you see what there is to do and to do it, but in general only the most basic things will be needed (and, frankly, my own level of expertise is not that high in this area).
So, sorry if this caught some of you off-guard! What we’re really after here is the big ideas, not the technical details, as far as statistics goes. Seeing enough of how statistics is done to get an idea of how an experiment can be best designed to make use of statistics is I think helpful, but that’s about as far as I intended for this to go.
Just to report on my own perusal of the schedule for the fast-approaching BUCLD 30, I wanted to highlight a few talks as possibly of interest, having to do with things we’ve talked about in class (or will talk about in class) at least partially.
When we get to the meeting before the conference (not next time, the time after that), we can discuss some of the things that will be relevant in these (or other) talks, as preparation for the conference itself. If you have any control over what times are blocked out for you due to volunteering duties, you might want to consider (some of) these talks as you plan when to work.
Friday, Nov. 4
Saturday Nov 5
Sunday, Nov 6.
By the way, somehow I think I failed to mention this, but I treat the experimental design as ungraded requirement—in the advertised “grading scheme” on the syllabus, it’s worth 10% of your final grade, but really that just means that not turning something in costs you 10%. I’m not going to grade it so much as I’m going to try to provide constructive commentary for the next step in the process. So, you can basically think of it as being worth one point (one relatively heavily weighted point), a point you get by handing the experimental design in, and lose by not handing it in.
Part of the reason I’m doing it this way is that I want you to feel free to “take risks” in your design proposal, or allow for you to put down a couple of different ideas.
Incidentally, if you are interested in doing your experimental design (and possibly the course project) on a topic that concerns second language acquisition, there are a couple of things to consider.
We’re already looking at the first 2 chapters of the White textbook for next week, although you might want to skim ahead a bit as well in the next couple of chapters, just to get an idea of what issues arise in this kind of research.
To get started, one thing that you’ll probably want to think about is where the two languages in question differ. You want to be a little bit careful about this as well, because the type of difference that you should be looking for is the kind of difference that we actually have some kind of theoretical understanding of. The traditional parameters such as verb-raising, null subject, wh-movement are good examples. (That is, do the languages differ with respect to whether they allow null subjects? Do they differ with respect to whether the verb moves to T? Do they form wh-questions by moving all, one, or no wh-words?) We pretty much know what’s going on in each language on these dimensions, and so if you have two languages that differ here, you can begin to make predictions and test them.
Another thing that you might try to consider is what second language learners with this pairing of native and target language often “have trouble” with. This might give you ideas, though again I want to stress that we only want to try to deal with things that we have some kind of a theoretical understanding of. Something like the ordering between verb and adverb, or the ordering between adjective and noun, or sensitivity of wh-questions to syntactic islands, these we understand well enough to say something about.
The other thing you want to think about is what question you’re hoping to answer. One question that is not of particular interest is: “Is it possible to learn Russian?” (which one might try to answer by testing native speaker judgments and comparing them to second language speaker judgments). That is, we want a much narrower question than that. For example, in second language learner’s grammatical representation of the target language, does the verb move? What are the implications of this with respect to having a second language with parametric settings that differ from those of one’s native language?
Once you have an idea of the area you want to look at, you’ll want to formulate a hypothesis or two about what you’ll find. Then, try to think of things that might be predicted if one or the other hypothesis is actually true, and plan to try to see if those predictions are met. A relatively simple example might be that if the verb in fact raises to T, it should move past not only negation but adverbs as well. So, you’d want to check both to increase your confidence that you are actually seeing verb movement to T when the verb precedes negation (rather than something like a “negative suffix” in negative sentences).
Just a couple more clarifying comments, with respect to the one actually difficult question on the homework.
First of all, let me state succinctly what the constraints are:
Structural Parallelism requires that the overt pronoun and the elided pronoun are interpreted as being the same type. So, it is satisfied when both pronouns are interpreted as bound, or when both pronouns are interpreted as referential. It is not satisfied if one pronoun is interpreted as bound and the other is interpreted as referential.
Referential Parallelism only comes into play when both pronouns are interpreted as referential; so, if either or both of the pronouns are interpreted as bound, Referential Parallelism is irrelevant. But when it is relevant, it says that both pronouns should be interpreted as referring to the same individual.
Principle B regulates binding relations, so it will only be relevant when one or both of the pronouns is interpreted as a bound variable. If a pronoun is interpreted as a bound variable, then its binder must be “far enough away” in the usual sense (the binder has to be outside the smallest DP or IP that properly contains the bound pronoun).
Principle P (or Rule I, Rule I*, etc.) says that bound variable interpretations are preferred. Or, put another way, you don’t use more indices than you have to. This is the principle that can be overridden by adults in special contexts (the “identity debates” or the “Evans-style” contexts, etc.).
Now, you are given a particular scene and a sentence, and you’re asked to determine whether kids will (at least sometimes) agree that the sentence describes the scene. An adult would not, but the question is whether there is an interpretation of the sentence that the kid might allow that would come out as true in that scene.
The evidence (Thornton & Wexler’s study) is that kids always respect Principle B and Structural Parallelism, but are relatively willing to violate Principle P and Referential Parallelism.
Back to the problem at hand, you have a sentence. On its own, this sentence has many, many possible interpretations. Each pronoun might be a bound variable, or it might be referential. If it’s referential, it might refer to Mama Bear, Goldilocks, or someone else. The other pronoun has the same options, and there’s the additional question of whether, when the two pronouns are referential, they refer to the same individual. So, many, many possible interpretations.
However, you also have a scene, and so you can narrow down your focus to just the interpretations that might come out “true” for that scene. The scene is one in which MB is washing herself, and GL is washing herself. So, how might that come about?
Here’s a little head start on how the thinking goes…
One possibility would be:
MB λx[x touches x], GL λx[x touches x]
Here, both pronouns are interpreted as bound variables. The sentence would evaluate as true of the scene. It satisfies Structural Parallelism, and Referential Parallelism is irrelevant. It satisfies Principle P. But it violates Principle B. And kids always respect Principle B, so they would not consider this a possible representation. So, if the kid is going to agree that this sentence accurately describes the scene, it will not be on the basis of the interpretation given above.
Another possibility would be
MB λx[x touches x], GL touches GL
Here, the overt pronoun is interpreted as a bound variable, and the second one is interpreted as referential, referring to GL. But this violates Structural Parallelism (the two pronouns are not interpreted as being the same type), and the overt pronoun violates Principle B. So, this one is also not among the representations a kid will be considering.
And so forth…
If at the end, you’ve considered all of the ways this sentence might come out “true” of the scene in question, and none of them would have been considered by a kid (who obeys Principle B and Structural Parallelism, but can at least sometimes disregard Principle P and Referential Parallelism), then the prediction is that the kid would not accept the sentence. If, however, at least one of the interpretations is possible given the constraints that kids obey and ignore, then the kid might sometimes accept the sentence.
Powered by WordPress