The thematic verb-raising
studies:
Why they may be wrong, and
what we can do about it
Projects for SLA
Spring 2000
The many verb raising studies in L2 acquisition – starting with White (1989) and going through Eubank & Thipaut-Grace (1998) – generally used adverbs like rarely or always to indicate the position of the thematic verb itself. Hence, a sentence like (1a) below was assumed to involve a thematic verb in situ (i.e., not raised) while a sentence like (2a) below would be one with thematic verb raising. (Here we use English as a metalanguage to represent possibilities in several languages; for this reason, the sentence in [1b] below is not indicated as ungrammatical.)
(1) a. Harry rarely eats bananas.
b. Harry eats rarely bananas.
The most usual – and quite simple – theoretical framework underlying these studies involved a projection TP dominating VP, with the adverb adjoined to VP. Under this framework, the representation in (2a) below would be that for (1a), and that in (2b) for (1b). (Note that we abstract away from the VP-internal subject hypothesis in these representations.)
(2) a. [TP
Harry [T’ T [VP often [VP [V’ eats bananas ]]]]]
b. [TP Harry[T’ [T eatsi ][VP often [VP [V’ ti bananas ]]]]]
Of interest in such representations is that verb raising is either required or prohibited; for this reason, a sentence like (1a) is grammatical in English (no raising), but not in French (raising) while a sentence like (1b) is grammatical in French, but not in English.
The various L2 studies, but especially those of Eubank et al. (1997), Eubank & Grace (1998) and Eubank & Grace-Thipaut (1998), showed that L2 learners may optionally raise thematic verbs. Early on, it was thought that this kind of raising might be just a developmental stage; at some later stage, the raising effect would become obligatory in one way or another. However, the later studies, now including the study of Beck (1998), showed that the developmental understanding was not tenable. Rather, Beck argued, for example, that L2 knowledge incorporates a local impairment to the mechanisms that control verb raising.
A legitimate question, however, is whether the raising studies involving adverbs could be an artefact of assuming that adverbs would uniformly indicate the position of the thematic verb. After all, it is clear that adverbs like always may occupy several different syntactic positions, as shown by the English examples in (3a-d) below.
(3) a. Tom always eats spinach.
b. Tom has always eaten spinach.
c. Tom always has eaten spinach.
d. *Tom eats always spinach.
The example in (3a) is like example in (1a) above, and that in (3d) is like that in (1b). Of more interest are the two examples in (3b) and (3c). On the assumption that non-thematic verbs occupy T in representations like that in (2a) above, how does one represent the position of the adverb in (3c). Clearly, something is askew here; there must be an adverb position located to the left of T (and to the right of the subject). More generally, these examples suggest (i) that the simple theoretical framework assumed in most studies of verb raising are likely too simple, and (ii) that the use of adverbs as indicators of the position of the verb may have resulted in potentially uninterpretable findings.
What can one do to discover whether findings from the relevant studies to date were on the right track? One possiblity would be to re-examine the adverbs themselves. Note in this regard that the freedom of placement for adverbs suggested in (3a-c) above does not extend to all adverbs. Consider the English sentences in (4) through (6), for example.
(4) a. Mary deliberately bonked her sister on the head.
b. Mary always bonked her sister on the head.
(5) a. Mary has deliberately bonked her sister on the head.
b. Mary has always bonked her sister on the head.
(6) a. *Mary deliberately has bonked her sister on the head.
b. Mary always has bonked her sister on the head.
What these examples suggest is that while adverbs like always may occupy a position either to the right or to the left of T, adverbs like deliberately may not. It appears, then, that adverb placement is a rather complex story, more than likely having to do with how the semantics of given adverbs would be represented in extended representations involving a good deal more structure than the simple structures in (2) above might suggest. More generally, it appears to me that reanalyzing the “adverb” studies might involve determining (in some way) not only where thematic verbs are located with respect to adverbs, but also how learners represent the semantics of these adverbs – and how these representations are appear in linguistic theory in general, in the learners’ native languages, in the mature L2, and, of course, in the interlanguage. A mind-boggling project, indeed.
A rather different approach to the problem is to switch away from adverbs to some other marker that indicates the position of the verb. The most usual marker that comes to mind is (clausal) negation (e.g., English not). However, negation itself can be somewhat tricky because the negation particle functions as a free morpheme in some languages and as a bound morpheme (or at least as a clitic) in others. Hand in hand with this variation (and with related sources of superficial variations in word order, like verb raising), we find variation in the placement of negation vis-à-vis the thematic verb. (Witness, for example, the placement of not vis-à-vis thematic verbs in English as opposed to the placement of no- vis-à-vis thematic verbs in Spanish.)
A more promising approach – I hope – is to examine the floating quantifiers all and both. Here we follow the standard assumption that these quantifiers appear underlyingly with the subject constitutent, and that subjects themselves originate deep in the representation, perhaps from Spec,VP. What makes these quantifiers “float” is that they may or may not move to Spec,TP with the subject. Hence, even in English, we observe sentences like those in (7a-b).
(7) a. Both boys threw rocks.
b. The boys both threw rocks.
Crucial here is the assumption that the thematic verb does not raise to T overtly. As a result, even though the subject has moved to Spec,TP in both sentences, the quantifier has raised along with the subject only in (7a); it is in-situ in (7b).
Now, consider the case of a verb-raising language. We may use German as an example language here (since I know enough of that language to come up with example sentences). In particular, consider the sentences in (8a-b):
(8) a. Beide Jungen warfen Steine.
both boys threw rocks
b. Die Jungen warfen beide Steine.
the boys threw both rocks
Superficially, then, the German sentences appear somewhat similar to their English counterparts in (7a-b). The one exception is the placement of the verb in (8b), which now appears to the left of the floating quantifier rather than to its right. This is so because German, unlike English, requires thematic verbs to raise to a high position, at least as far as T in matrix clauses.
We can take advantage of the behavior of floating quantifiers to retrace the steps of some of the earlier verb-raising studies. More specifically, the plan is to conduct two studies on the matter, each employing a different experimental methodology. The one study would involve a fairly traditional grammaticality judgment task; the other study would involve a truth-value task. I will describe each of the two studies in a little more detail below.
In this study, we create stimulus sentences and distractors along the general lines of those developed by White (e.g., 1992), though now with both/all floating quantifiers. In line with the considerations described in Eubank et al. (1997), we test speakers of English whose native language is Chinese in order to remove the potential of transfer effects (though this assumption needs to be researched more in order to be sure). In this scenario, we examine results on stimuli like the one in (9) below.
(9) The boys threw both a rock.
For a stimulus like (9), if learners either require or even optionally allow the verb to raise to T overtly, then they should think that the sentence is a possible one: The verb raises to T, the subject raises to Spec,TP, but the quantifier remains behind. However, if they do not allow the verb to raise overtly under any circumstances, then the sentence should be ungrammatical for them. (Note, of course, that if learners think the sentence is possible, we would not have shown that optional raising obtains, but rather only that raising obtains, whether optional or obligatory; however, since this raising would presumably have come about neither on the basis of the native language nor on the basis of exposure to the target language, such a finding would certainly raise the same kinds of issues addressed by Eubank et al.)
For the study, we would need to develop a good number of stimuli like (9) along with an equal number of stimuli like (10).
(10) The children both drank a glass of water.
Further, because these sentences are so short, it will be necessary to create a significant number of distractors for this study. If we have, say, eight stimuli like (9) and another eight like (10), then we’d need at least double the total number of stimuli (i.e., 32 of them) divided evenly between grammatical and ungrammatical. My strong suggestion is that these distractors be of the same general kind, namely, using quantifiers (other than all or both) in ways that sound either possible (e.g., The girl read every book in the library) or impossible (Any man should not buy cigarettes). In any event, they should not have anything whatsoever to do with verb raising.
As for subjects and other related methodological procedures, we should follow fairly closely along the lines of the Eubank et al. and related studies. Of course, background library research will need to examine not only the previous verb-raising studies in L2 acquisition, but also the use of grammaticality judgments as research methodology (whether in L2 studies or else wise).
To approach the matter from a methodologically different direction, we should develop a truth-value study. In truth-value judgments, subjects are provided with a story and a stimulus sentence; they are then merely asked whether the stimulus is true or false according to the story. To carry this out, all stimuli must be syntactically ambiguous in that two distinct readings are potentially possible for either the experimental group (here, the L2 speakers) or the control group (here, the native speakers), but not both. This requirement turns out to be met by certain sentences with floating quantifiers of the relevant kind. Consider the sentence in (11) below.
(11) The people drove all the cars.
For native speakers of English, the sentence (presumably) has only one interpretation: All refers to the cars; it does not refer to the people. Suppose, however, that the relevant L2 speakers permit verb raising, as we might suspect from the previous “adverb” studies. If they do, then a sentence like (11) could have two interpretations, namely, one in which all refers to the cars, or another one in which all refers to the people.
To turn a “car” example like (11) into an experimental truth-value item, choose one of the two possible responses to the stimulus; suppose we choose false. We now write a story that matches the false response, in this case one in which it is not the case (i.e., it is false) that the people drove all the cars (but in which it is true that all the people drove). Consider, then, (12).
(12) Mary and John and their daughter Jane are rather wealthy; they own three cars! Last Monday Mary drove the pink car to the store in the morning, and John drove it to the mall in the afternoon. Jane drove the brown one to the movies in the afternoon. They didn’t drive the green car on that day.
Mary and John and Jane drove drove all the cars.
TRUE FALSE
Note in (12) that it is true that Mary and John and Jane all drove cars, but it is false that they drove all of their cars (because they drove only two of them). Now, because native speakers are expected to have only the interpretation in which all refers to the cars, they should respond with false. For the L2 speakers, however, if they permit verb raising, then they should be able to get two interpretations from the stimulus sentence, namely, all refers to the cars, or all refers to Mary and John and Jane. As a result, it should be possible to observe, among the L2 speakers, some false responses (like the native speakers) and some true responses. The critical part of the experiment would be to compare (statistically) the number of true vs. false responses across natives and non-natives (allowing us to set up a contingency table to which a chi-square can be applied).
This kind of all item seems fairly easy to create; I only made one total mess of it (in an earlier draft of this very document). My suspicion is that it will be more difficult to create items of this kind with both (though we should try just to make sure). In any event, the total number of experimental items of a type (e.g., all items with expected false from NS) will have to be fit into an overall counterbalanced design. If we have six all test items like (12), for example, then we’d also need six more all items as distractors that yield a true from NS; we’d also need around two times the total amount in other distractors (i.e., 24 non-all distractors), again evenly divided between true and false response expectations. (Note that distractors of any kind – all or otherwise – do not need to involve syntactically ambiguous sentences like those in the real test items.) As in the case of the grammaticality judgment study, these other distractors should probably also involve other quantifiers (e.g., every).
As above, we should follow fairly closely along the lines of the Eubank et al. and related studies on subjects and other related methodological procedures. In addition, background library research will need to examine not only the previous verb-raising studies in L2 acquisition (again), but also the use of the truth-value task as research methodology. (Note that the truth-value task was originally developed for use in first-language acquisition research.)