Timed Grammaticality Judgements by Vivian Cook

One of the main areas that Hans Dechert has illuminated is the importance of linguistic performance whether through his own writings, the conferences he organised, or the books he edited. The following paper makes some attempt to reconcile the Universal Grammar theory with performance criteria through the use of timed grammaticality judgements. Hence it will be regarded as heretical by those who refuse to take performance evidence into account; for others it may show how performance too can play its role in a UG oriented account of Second Language Acquisition, at least in the provision of other forms of data.

Universal Grammar (UG) theory, as put forward by Chomsky (e.g. Chomsky, 1986a; 1988), sees the speaker's knowledge of language as couched in terms of the principles and parameters theory (Chomsky, 1988; Cook, 1988a). The acquisition of the first language is a process of setting the values for parameters in response to 'triggering' evidence, rather than one of learning 'rules'. Acquiring the setting for a parameter such as the head parameter, which specifies the order of heads and complements in the phrase, requires comparatively minor evidence; as Chomsky (1987) puts it::

These facts can be determined from very simple sentences, for example the sentence "John ate an apple" (in English) or "John an apple ate" (in Japanese). To acquire a language, the child's mind must determine how the switches are set, and simple data must suffice to determine the switch settings, as in this case.

While recent years have seen a large growth in L2 learning research claiming to be based on the Chomskyan UG theory, surveyed in Cook (1993), there are nevertheless many methodological problems. Firstly, many of the areas of syntax tackled in L2 learning research are not, strictly speaking, part of the principles and parameters theory itself, for example the Principle of Branching Direction studied by Flynn (1984) or implicational universals such as the Acceptability Hierarchy looked at by Gass (1979) and Hawkins (1988). This is not to suggest that these areas of investigation do not have an important role in research into L2 learning of syntax in a broad sense nor that there is not ultimately a place for them within a principles and parameters framework. But their status in arguments about UG is different from aspects of syntax that are currently part of the principles and parameters theory and hence based on notions of learnability. To settle issues concerning UG means tackling areas of syntax that play a part in principles and parameters theory. There is a danger of thinking that any abstract syntactic point must have something to do with UG, when, in the strict Chomskyan sense, the only ones that are relevant are derived from a model with an appropriate goal.

Secondly, even the L2 research concerned with principles and parameters syntax as such has mostly dealt with the periphery of syntax where languages differ more and UG is less applicable rather than the core areas; there has been little or no investigation of such core concepts as the Projection Principle, the Theta Criterion, or Case Theory, with some honourable exceptions. Other research has looked at areas where the syntax itself is debatable, say German where L2 researchers assume German is SOV underlyingly, changing to SVO in certain main sentences (Meisel, 1986; Clahsen and Muysken, 1986; Jordens, 1988), while in LI acquisition Mills (1985) claims "Grammar books refer to a canonical word order of subject-verb-object from which they derive the other word orders." Thirdly, much of the research has relied on "rules" specific to one construction, for example the dative alternation (Mazurkewich, 1984) or preposition stranding (van Buren & Sharwood-Smith, 1985), rather than on the across-the-board principles and parameters that the theory requires. If one is to study movement within a principles and parameters framework, for example, it is necessary to take in not only Verb-movement affecting zero-bar categories, but also wh-movement and NP-movement which affect maximal projections (i.e. NPs and wh-phrases).

In X-bar syntax all the phrases of human language conform to a simple structure (Jackendoff, 1977; Chomsky, 1986a; Cook, 1988a). Each phrase contains certain functionally determined items - a head, a specifier, and complements. The maximal two-bar phrase (X") has a specifier and a one-bar head of a particular type (X'); this in turn has a head that is a lexical or functional category (X) and a complement.

Thus the maximal Noun Phrase in English (N"), say "his fear of the dark", consists of a specifier "his", and a head (N'), "fear of the dark"; this N' consists of a lexical category (N), "fear", and a complement, "of the dark", a prepositional phrase with internal structure of its own.

The variation in word order between languages is captured by the head parameter; when the head comes before the complements, the setting for the parameter is said to be "head-first"; when it comes after the complements, it is said to be "head-last". English, with a head-first setting, has Verb Phrases with Verb heads before complements, "ate a tomato"; Adjective Phrases with Adjective heads before complements, "easy to play"; Prepositional Phrases with Preposition heads before complements, "with a spoon"; and Noun Phrases with Noun heads before complements, "fear of the dark". A language with a head-last setting such as Japanese has the Verb head after complements in the Verb Phrase, "Nihonji desu" (Japanese am); the Preposition head after its complement NP, "Nihon ni" (Japan in), i.e. a Postposition. All languages are claimed to be consistent in the relative position of head and complements in phrases in that their position can be stated once and for all in the constituent structure of each language. A subsidiary claim is that specifiers occur on the opposite side of the phrase to complements; this may indeed be part of the same parameter, or may be a separate parameter, as discussed in Fukui (1986); Rothstein (1985) similarly makes the position of subject a parameter with in her Rule of Predicate-Linking. Though not all languages are entirely consistent in setting, all 126 languages in the sample in Hawkins (1983) for which the appropriate information is listed have statistically significant consistency of setting for the complements in the Verb Phrase, Noun Phrase, and Prepositional Phrase (p.<0.001, chi square test).

In Chomsky (1986b) the X-bar principles are extended to include the Complement and Inflection Phrases. The Inflection Phrase (IP) brings together abstract features of the sentence such as Tense and Agreement and is seen as having two levels, I" and I'. The N" specifier of I" is the subject of the sentence; subject is thus defined as a specifier in terms of the configuration of the phrase. The Extended Projection Principle makes the presence of a subject compulsory, even if it may be covert. The Verb Phrase is a V" acting as complement to I. Let us diagram this for "The drummer missed a beat."

The NP "the drummer" is the specifier of IP and hence the subject of the sentence; the head I' consists in turn of a head I, which consists of the abstract elements 'past' and 'singular' that become the Verb ending, "ed", and the complement V", "missed a beat". The IP thus conforms to the same framework as the other types of phrase: a head with zero, single and double bars (I, F, I"); a specifier, which is the subject; and a complement, which is the Verb Phrase. The variable position of the subject before or after the VP is an aspect of the head parameter governing the position of specifiers. In English the specifier "very" comes before the Adjective, as in "very nice;" the specifier "the" comes before the Noun, "the man"; and the specifier subject "he" comes before the VP, "he smokes." English specifiers consistently occur on the opposite side to complements. For more extensive discussion of the head parameter see Cook (1988a), Stowell (1981), and Huang (1982). Later developments such as Fukui (1986) or Larson (1988) have modified the principles to some extent by separating functional from lexical phrases and by treating the subject as occurring within the VP and moving into specifier position in the IP; these do not affect the head parameter itself so far as the complement/head and specifier/head order are concerned (Fukui, 1986, p. 60).

The head parameter at one level is a straightforward obvious point about the phrases of a language; at another level it is simply a label for certain consistencies with various explanations. It will be used here as a coverall term in this way; particular aspects of it are susceptible to different explanations.

In L1 acquisition broad claims have been made that the child does not make word order mistakes; for example Brown et al (1969, p.42) state that "the children we have studied scrupulously preserve sentence word order not only with respect to basic relations but also with respect to the order of articles, adjectives, auxiliary verbs, adverbs, and all other words." The Bristol transcripts from English children aged 15 months to 27 months confirm virtually no mistakes of order with Verb and complement order, Preposition and complement order, or Subject and VP order. In L2 learning; Meisel (1986) found that L2 learners of German and French prefer an SVO order from the start, unlike L1 learners. Consistent extrapolation strategies of word order from one phrase type to another have been found in children learning Micro Artificial Languages with different head parameter settings, though not quite those predicted by the head parameter itself (Cook, 1988b). Other L2 learning research that appears to concern word order is often either tackling some aspect of movement other than the order of elements in the phrase structure covered by the head parameter (Clahsen and Muysken, 1986; Jordens, 1988) or is concerned with an aspect of typological universals rather than the principles and parameters approach (Lujan et al, 1984).

The main L2 learning questions that can be explored through the head I parameter are the different types of access to UG (Cook, 1985; 1993). First language acquisition starts from the zero state of knowledge (S₀) and ends with a final steady state of knowledge (S_s); second language acquisition yields a terminal state of knowledge (S_t), which may be derived either from other mental faculties (a no-access model, i.e. non-availability of UG), or from the UG S₀ (a direct-access model), or from the LI S_s (an indirect-access model) (Cook, 1988b). A continuing theme in L2 learning research has been whether UG is still available to L2 learners, a variant on the question of whether L2 learning is like LI learning, discussed say from Ritchie (1978) to Bley-Vroman et al (1988). In the no-access model L2 S_sis related to other parts of the mind than the language faculty: however second languages are learned, it does not involve UG. Phrased in the terms of our research, this means seeing whether the head parameter is indeed available to L2 learners as it is to LI learners; relevant evidence would be consistency of response from L2 learners, comparatively error-free development, and extrapolation from one phrase type to another.

Supposing that UG is still available, access to it might still be indirect or direct. Much L2 learning research has supported the indirect-excess model in that L2 learners have been shown to be influenced by the LI S_ssettings rather than to start from the S₀ zero settings, e.g. subjacency (White, 1985) and the pro-drop parameter (White, 1986; Hilles, 1986). The indirect-access model would be supported if L2 learners with different Lls made different mistakes in the aspects of word order covered by the head parameter.

The two experiments to be reported used the same basic design. This adds a measurement of response time to the tradition of grammaticality judgments. The assumption is that the time involved in coming to a judgement may in itself reflect the difficulty of the task to the learner, regardless of whether the answer is right or wrong. So in this design Subjects are asked to say as quickly as they can whether a sentence is grammatical and the time they take is measured. This technique of timed grammaticality judgements yields two sets of scores: the judgements the Ss give and the response time taken.

The overall aim was to test whether L2 learners of English are consistent in the setting for the head parameter in the Preposition Phrase and whether learners with different Lls are similar or different through timed grammaticality judgements for Preposition Complement versus Complement Postposition orders. The direct and indirect access models predict consistency of response; the indirect-access model predicts differences between learners according to their LI setting. Experiment 1 compares timed grammaticality judgements for groups of learners whose Lls differ in having Prepositions or Postpositions, i.e. in having Prep NP or NP Post order in the Prepositional Phrase. The 20 test sentences are given in Appendix A; 10 had Prepositions and so were correct English, 10 had Postpositions and so were incorrect; in addition, there were, from the point of view of the present paper, 10 distractor sentences, 5 of which had an adjacency violation, 5 of which did not. The ten Preposition sentences were mirrored by ten Postposition sentences using the same vocabulary pool. Seven out of each set of 10 had PPs within the VP structure of the sentence, e.g.

Four practice sentences at the beginning were unconnected with the test sentences in structure.

Forty-two Ss learning English were tested; of these 21 spoke Lls that had Prepositions (Portuguese, Spanish, Arabic, Peulh, French, Italian, Czech, Polish), and 21 spoke Lls that had Postpositions (Chinese, Turkish, Urdu, Maithali, Japanese, Korean, Nepali, Sindhi); all were adults studying English at different institutions in England and were of intermediate standard in English, so problems of vocabulary were unlikely to affect the results.

The experiments reported here used a BBC microcomputer fitted with the word processing program WORDWISE and utilized a timing program EXMORE developed at the University of Exeter Psychology Department for carrying out psycholinguistic experiments (Mitchell & Barchan, 1984); this accurately times presentation and responses, records which key is pressed, and calculates the average timings for various conditions. Another L2 use of EXMORE for comprehension is reported in Cook (1990). The Subjects are asked whether a sentence is grammatical or not, and the time they take to answer is measured. They are presented with a sentence in the centre of the screen and with the request to "Press A if it is correct English, B if it is not correct English, and C if you are not sure;" the letters B and C were reassigned to the keys G and L to equalize the areas of the keyboard they had to cover; these keys were labelled accordingly and the other parts of the keyboard were obscured with a cloth. After the S had pressed the appropriate key, the message "Press Spacebar to go on" appeared; the program went on to the next sentence if the S had not answered after ten seconds. The sentences were presented in a randomized order that was the same for all Ss. The Ss were told that the experiment was to compare how learners of English from different nationalities treated various sentences; the computer would score their answers and would also record how long they took so that they should be as quick as possible. The experimenter took them through four practice sentences to see that they understood the instructions.

The timing results are given in Table 1; a null answer by the S counted as 10 seconds, as we saw above. Comparing the LI backgrounds, speakers of Lls with Prepositions were 0.535 second faster at judging Preposition sentences than at judging Postposition sentences (t test, one-tailed, repeated measures design, df 20, t=3.01, p.<0.01); speakers of Lls with Postpositions showed no significant difference for the two sentence types. Thus the LI had a marked effect on the result. Overall the two groups averaged 4.042 for Preposition sentences and 4.304 for Postposition sentences; while the groups were the same on Preposition sentences, the LI Preposition speakers were worse at Postpositions (t test, independent samples, one-tailed, df 40, p.<0.0l).

	N	Preposition Timings	Postposition Timings
All Ss	42	4.042	4.304
Prep N	21	3.926	4.461
N Post	21	4.158	4.148

The results for accuracy of judgement are given in Table 2. All answers were scored as wrong that were either wrong, or "not sure," or exceeded the time limit of ten seconds. The proportion of wrong answers was 15.5% overall; a typing mistake in the data for the experiment meant one of the Preposition sentences was incorrectly labelled and so had to be discarded, resulting in 9 Preposition sentences compared with 10 Postposition sentences. Overall no statistically significant differences emerged between judgements across sentence types or groups (chi square tests), though there was a slight advantage in correctness for the Ss with Postposition Lls in judging the sentences with Postposition (just as they were comparatively quicker at these). In other words, the differences in timing results are not reflected in clear differences in ability to judge correctness. The indirect access model is confirmed in so far as the timings of the grammaticality judgements differ according to the L1 of the learner; learners with a Preposition L1 are slower at evaluating whether Postposition sentences are correct in English.

Language	N	Preposition Percent Right	Postposition Percent Right
All Ss	42	84.2%	84.8%
Prep N	21	84.2%	81.0%
N Post	21	84.2%	88.6%

Experiment 2 used the same methodology, design, and instructions as Experiment 1 with different materials testing the order of Subject Verb and Object. The indirect-access model vis-a-vis the direct access model again predicts an influence from the LI; speakers with SVO Lls should treat the four sentence types differently from those with SOV or VSO Lls. The aspects of the head parameter tested here are the position of the Subject as specifier of IP and the position of Object as the complement of VP. The method was to compare timed grammaticality judgements for sentences with four different word orders, namely SVO, SOV, VSO and VOS, by groups with different word orders in the LI; SVO is then the only one of the four that is normally grammatical in English. VOS, found in 3% of the world's languages (Tomlin, 1986), has the same setting for the VP as English SVO (42%); SOV (45%) has the opposite setting to English; VSO (9%) is thought to be produced by movement of V from final position as it would otherwise violate the order principle (Randall, 1985), though this still results in a breach of the principle of Verb-Object Bonding (Tomlin, 1986). Sixteen of the 32 test sentences were distractors so far as the present experiment is concerned and tested the dative alternation. Four sentences had the normal SVO order for English, four SOV, four VSO, and four VOS:

John likes bananas. (SVO)
Bill apples likes.     (SOV)
Plays Sue tennis.     (VSO)
Drinks coke Joan.    (VOS)

The subject of the sentence was always animate, the object always inanimate. The test sentences are given in Appendix B.

The 38 Ss were drawn from the same institutions and population used in Experiment 1. Sixteen spoke SVO languages (Greek, Spanish, French, Polish, Serbo-Croat, Bulgarian, Indonesian, Amharic, Portuguese), 11 spoke VSO languages (Arabic), and 11 SOV languages (Urdu, Farsi, Japanese, Turkish, Bengali); no speakers of VOS languages were available.

The timing results for Experiment 2 are given in Table 3. The average time for all Ss for judging a sentence correct or incorrect with SVO was 3.934 seconds, with SOV 4.401 seconds, with VSO 3.971 seconds, and with VOS 3.629 seconds (all results; ANOVA, F=51, p<0.0l: SOV vs VOS; t test, df 37, p.<0.001). For all groups the VOS sentences were processed most quickly, the SOV sentences most slowly; the SOV and VSO sentences were in the middle. There appears then to be a common order across Lls in terms of processing time.

	N	A SVO	B SOV	C VSO	D VOS	Order for speed
SVO	16	3.566	3.954	3.555	3.124	VOS<VSO/SVO<SOV
SOV	11	3.891	4.260	3.812	3.543	VOS<VSO<SVO<SOV
VSO	11	4.560	5.276	4.817	4.541	VOS<SVO<VSO<SOV
all		3.934	4.401	3.971	3.629	VOS<SVO<VSO<SOV

The results for accuracy of judgement are given in Table 4. As before, all answers were scored as wrong that were either wrong, or "not sure," or exceeded the time limit of ten seconds. The proportion of wrong answers was 34.8%, larger than for Experiment 1. The SVO and SOV speakers have a similar order of accuracy from least to most mistakes VOS>SVO/ VSO>SOV; the VSO speakers not only have more mistakes overall but also a different order of SVO>SOV>VOS/VSO. The most difficult type for VSO speakers was sentences that had the same word order as their Lls, not true for either SVO or SOV speakers. The differences between the VSO and the SVO/SOV speakers were statistically significant (chi square test on the actual scores, df 7, p. <0.001). In other words, the essentially common order for processing time contrasts with the different orders for accuracy in this experiment. Again an effect from the LI is seen on one of the two measures and the Indirect Access model is supported.

	N	A SVO	B SOV	C VSO	D VOS	Order for accuracy
SVO	16	96.9%	92.2%	96.9%	100%	VOS>SVO/VSO>SOV
SOV	11	93.2%	93.2%	95.5%	100%	VOS>VSO>SVO/SOV
VSO	11	88.6%	79.5%	75.0%	75.0%	SVO>SOV>VOS/VSO

Clearly the LI setting for the head parameter plays a role in L2 learning. In Experiment 1 learners with Postposition Lls took equal time on both Preposition and Postposition sentences while learners with Preposition Lls were quicker on Preposition sentences than on Postposition sentences; thus the unfamiliar form of the Postposition sentences for the Preposition speakers seemed to mean that they took longer to judge it. Not only are there differences between the groups but these differences are related to the LI setting for the head parameter. In Experiment 2 we also find differences in accuracy for LI between SVO and SOV on the one hand and VSO on the other. Looked at as a matter of subject position, the SVO and SOV speakers (both with initial S) find all sentences equally easy; the VSO (with S medial) find the VSO and VOS (with final S) more difficult. Looked at from the point of view of Object position, the SVO and SOV speakers should be diametrically opposed, which they are not, and VSO speakers (if derived from SOV) should group with SOV, which they do not. Perhaps again this illustrates the odd status of VSO languages; it is the factor of movement which needs further research. Subject position appears more crucial than Object position.

The results from these two experiments provide further evidence against the no-access model of L2 learning and in favour of the indirect-access model in that learners are influenced by their LI setting sometimes in terms of accuracy, sometimes in response time; for PP, IP, and VP there are clear LI differences. Again this reinforces an argument for testing multiple consequences of the same parameter; while Postposition Lls had an effect on Experiment 1, Experiment 2 did not show all three Lls behaving in precisely the way the head parameter predicts. Like LI children, L2 learners have little overt problem with the setting, except inasmuch as their LI makes them slower or sometimes leads to mistakes. One of the aims of this research is achieved in that it has shown the relevance of a central principles and parameters area of syntax to L2 learning - not simply a construction-specific rule but a general parameter setting.

The second aim of this paper was to extend the methodology of L2 learning research. It is notoriously hard to decide what evidence counts as appropriate in UG research, as discussed in Cook (1993). Some researchers rely on observation of language produced by learners, say Hilles (1986) and Jordens (1988) in the L2 or Radford (1990) and Hyams (1986) in the LI; this runs into the problem of relating concrete E-language sentences with abstract I-language knowledge, as discussed in Chomsky (1965), or of confusing language development with language acquisition (Cook, 1988b), or of arguing only from the "positive" evidence of actually occurring forms, which may be an inadequate guide.

Perhaps the majority of researchers in the UG paradigm utilize pen and paper judgements of grammaticality, for instance Bley-Vroman et al (1988) or Van Buren & Sharwood-Smith (1985). The relationship between such judgements and actual knowledge has always been debated (Kellerman, 1985); as a research technique grammaticality judgements have fallen out of favour in psycholinguistics; none of the research surveyed in the introduction by Garnham (1985), for example, to my knowledge involves grammaticality judgements; one reason put forward by Elliot (1981) is the link between grammatical judgements and level of literacy; Nagata (1988) shows the variation in judgement that context or familiarity through repetition can bring; Carroll et al (1981) show its susceptibility to self-awareness factors in the Subjects. Chaudron (1983) reviewed the use of such metalinguistic judgements in L2 learning research, concluding optimistically that there is reasonably clear evidence that as learners develop towards target language proficiency, their ability to match the experimenter's "objective" norms improves.

The addition of timing to the grammaticality judgement task has two consequences. One is that it assumes that the relevant measures are not just the judgement itself but how long it takes; extra processing time may indicate difficulty, less processing time may indicate ease. In the present research the LI setting did not so much inhibit as liberate in that Japanese speakers were comparatively quicker at judging the ungrammaticality of Postposition sentences. The second consequence is that control of the time factor forces learners into more similar behaviour; with the usual untimed task, some learners may reply immediately and instinctively but others may take time and conscious effort, or indeed change their answers; comparison may be difficult, if this factor is not controlled: the variation of activity possible in judgement tasks is constrained by giving the learner a limited amount of time in which to answer. The timed grammaticality judgement technique used here not only provided an additional measure but also forced the students into snap decisions that could not be reflected on or indeed changed and that probably limited the effects of variable self-awareness discovered by Carroll et al (1981).

The timing measure turned out to be crucial for experiment 1; a conventional grammaticality judgement test would have shown no difference between groups, as reflected in the accuracy scores; only when the response time is taken into account are LI differences found. Experiment 2 partly showed the reverse; while the timings of judgements were constant across groups, degree of accuracy varied for different Lls. Again this would not have emerged without both measures being available; this time the timing results show that the matter is not so straightforward as a grammaticality judgement approach alone suggests. The experimental technique turned out to provide a useful alternative source of evidence. Grammatical sentences do not necessarily take less time than ungrammatical sentences to process; certain types of ungrammaticality are quicker to detect than grammaticality. The order of VOS>VSO>SVO>SOV from quickest to slowest is the same as the order from most frequent to least frequent in the world's languages so far as VOS and SOV are concerned, with VSO and SVO being in between but in the opposite order (Tomlin, 1984); that is to say, VOS, the most frequent language, is the quickest type of ungrammaticality to spot in English; SOV, the least frequent language of the four, is the slowest to spot. This may be related to perceptual sequence from left to right: to evaluate VOS and VSO as ungrammatical means simply noticing the leftmost word; to say whether SVO and SOV are grammatical means scanning the sentence further. Alternatively it has been sometimes proposed that a subject precedes the predicate regardless of the head parameter (Travis, 1984; Tomlin, 1986).

The type of experimental technique and measure afforded by computer control has then much to offer to L2 learning research; it seems odd that the measurement of response time has rarely been utilized in L2 research with syntax or comprehension, though used in other L2 areas. One of the problems with UG-related research has been that the important parameters are set quickly and so are not revealed on grammaticality judgements; hence the concentration in the research literature on parameters that are set relatively late. If, as the Chomskyan view suggests, the head parameter is set from a small sample of sentences, it will be fixed in the first few hours of L2 learning; differences between learners on conventional grammaticality judgements may show simply that all speakers know it regardless of LI, as we saw in Experiment 1. The timing results show however that there is still a measurable effect for intermediate learners according to LI, despite their evident knowledge of the right setting. Further experiments in this paradigm have used timed comprehension (Cook, 1990) and timed sentence comparison tasks.

One issue that the technique raises in explicit form is the relationship between language use, language learning, and language knowledge. The primary evidence for UG theory is what the person knows about language; discussion in principles and parameters theory normally ranges around a few test sentences that demonstrate the grammatical and ungrammatical features in question, called in Cook (1993) "single sentence evidence" it does not reflect an experimental investigation of what several speakers know. The reason why L2 learning has not relied on such data is presumably that, while native speakers of a language share similar S_ss and so should not be too different from each other, few L2 learners reach a plateau but achieve S_ts that differ in many ways. Deciding if a sentence is grammatical reflects a process of judgement, during which the knowledge of language plays some part, as it does in normal speech production and comprehension. In Cook (1991) a distinction is made between decoding a sentence, in which the setting is assumed, and codebreaking a sentence, in which a setting has to be inferred. In LI acquisition the child codebreaks a handful of sentences to get the appropriate setting, which can be used for decoding from then on. In L2 learning the learner starts from the LI settings and needs to codebreak sentences of the L2 to get a new decoding setting. But what our experiments have shown is that the LI setting is still 'there' in some sense when processing the L2, affecting the decisions or the response time. The two languages are not in watertight compartments with two versions of the head parameter set in different ways, as research with code-switching confirms (Nishimura, 1986); a parameter doesn't switch between the LI and L2 settings so much as veer, leaving some trace of the alternative LI setting. This is then clear support for the wholistic multi-competence view that the two languages in the mind of the L2 user from a single super-system (Cook, 1992). The pro-drop setting is still available to speakers of English as evidenced by null subject sentences in certain styles and situations (e.g. "Can't buy me love"). It may be that the setting for a parameter is more fluid than normally considered. Code-breaking may always be possible; that is to say, confronted with a sentence that violates the normal setting, native speakers still have access to the other setting; this would mean that a sentence such as "John milk drinks" will first be decoded as nonsense by the English speaker and then code-broken as SOV using other settings potentially available to the speaker. Evidence for this might be the ease with which English-speaking children acquired SOV languages in the MALs experiment (Cook, 1988b), in which 80% learned them to criterion from 30 sentences. The whole thrust of the UG argument on L2 learning is that the parameter can still be set differently in an L2; perhaps during normal LI processing something similar goes on in that other settings are still available for codebreaking sentences that are not susceptible to decoding.

A strength of the UG position is that it does not restrict itself to a single kind of evidence; "In principle, evidence ... could come from many different sources apart from judgements concerning the form and meaning of expressions" (Chomsky, 1986a, p.36). It seems dangerous that L2 learning research should rely so heavily on a single source - grammaticality judgments in questionnaire. Other research into the head parameter has used different sources of evidence. Work with Micro-Artificial Languages in Cook (1992) found a set of 4 extrapolation strategies with word order; one appeared to be that the learners ignored the position of the Object in adopting a setting for the PP. In the real language L2 experiments used here, learners showed no signs of an overall preference for Postpositions; the strategy may then be a product of the MAL situation encouraging the students to look for 'anti-English', i.e. settings as different as possible to those in their LI, or may reflect behaviour in the early stage of acquiring the setting for a language rather than the more established stage the learners had reached here. The computer simulation of PAL (Program that Acquires Language) (Cook, unpublished) highlights inter alia the need to separate decoding passes from codebreaking passes. The present paper confirms that L2 learning too can be considered from the head parameter perspective and raises further questions about the importance of parameter setting in language performance that need further investigation.

Bley-Vroman, R.W., Felix, S., & loup, G.L. (1988). The accessibility of Universal Grammar in adult language learning. Second Language Research, 4,1-32.

Brown, R., Cazden, C, & Bellugi-Klima, U. (1969). The child's grammar from I to III. In J.P. Hill (Ed.), Minnesota symposia on child psychology (Vol. 2, pp. 28-73). Minneapolis: University of Minnesota Press.

Carroll, J.M., Bever, T.G., & Pollack, C. R. (1981). The non-uniqueness of linguistic intuitions. Language, 57,368-383.

Chaudron, C. (1983). Research on metalinguistic judgements: A review of theory, methods, and results. Language Learning, 33, 343-377.

Chomsky, N. (1965). Formal discussion: The development of grammar in child language. In U. Bellugi & R. Brown (Eds.), The acquisition of language (pp.35-39). Lafayette, IN: Purdue University Press.

Chomsky, N. (1986a). Knowledge of language: Its nature, origin and use. New York: Praeger.

Clahsen, H., & Muysken, P. (1986). The availability of universal grammar to adult and child learners - a study of the acquisition of German word order. Second Language Research, 2,93-119

Cook, V.J. (1985). Chomsky's universal grammar and second language learning. Applied Linguistics, 6,1-18.

Cook, V.J. (1988a). Chomsky's universal grammar: An introduction. Oxford: Basil Blackwell.

Cook, V.J. (1988b). Language learners' extrapolation of word order in phrases of micro-artificial languages. Language Learning, 38,497-529.

Cook, V.J. (1990). Timed comprehension of binding in advanced L2 learners of English. Language Learning, 40,557-599.

Cook, V.J. (1991). Second language learning and language teaching. London: Edward Arnold.

Cook, V.J. (1992). Evidence of multicompetence. Language Learning, 42,557-591.

Cook, V.J. (1993). Linguistics and second language acquisition. Basingstoke: Macmillan.

Cook, V.J. Universal grammar considered as a parser that acquires language. Online

Flynn, S. (1984). A universal in L2 acquisition based on a PBD typology. In F.R. Eckman, L.N. Bell, & D. Nelson (Eds.), Universals of second language acquisition (pp. 75-87). Rowley, MA: Newbury House.

Flynn, S. (1988). Second language acquisition and grammatical theory. In F. Newmeyer (Ed.), Linguistics: The Cambridge Survey (pp. 53-75). Cambridge: Cambridge University Press.

Fukui, N. (1986). A theory of category projection and its application. Unpublished dissertation, MIT.

Gass, S. (1979). Language transfer and universal grammatical relations. Language Learning, 29, 327-344.

Hawkins, R. (1988). Do second language learners acquire restrictive relative clauses on the basis of relational or configurational information? The acquisition of French subject, direct object and genitive restrictive relative clauses by second language learners. Second Language Research, 5,156-188.

Hilles, S. (1986). Interlanguage and the pro-drop parameter. Second Language Research, 2, 33-52

Hyams, N. (1986). Language acquisition and the theory of parameters. Dordrecht: Reidel.

Huang, C.-T.J. (1982). Logical relations in Chinese and the theory of grammar. Unpublished dissertation, MIT.

Jackendoff, R. (1977). X'-Syntax: A study of phrase structure. Cambridge, MA: MIT Press.

Jordens, P. (1988). The acquisition of word order in Dutch and German as LI and L2. Second Language Research, 4,41-65.

Kellerman, E. (1985). Dative alternation and the analysis of data: A reply to Mazurkewich. Language Learning, 35,91-101.

Larson, R.K. (1988). On the double object construction. Linguistic Inquiry, 19, 335-291.

Lightbown, P.M., & White, L. (1988). The influence of linguistic theories on language acquisition: Description versus explanation. Language Learning, 37,483-510.

Lujan, M., Minaya, L., & Sankoff, D. (1984). The universal consistency hypothesis and the prediction of word order acquisition stages in the speech of bilingual children. Language, 60,343-371.

Mazurkewich, I. (1984). The acquisition of the dative alternation by second language learners and linguistic theory. Language Learning, 34, 91-109.

Meisel, J.M. (1986). Word order and case marking in early child language. Evidence from simultaneous acquisition of two first languages. Linguistics, 24,123-183.

Mills, A.E. (1985). The acquisition of German. In D.I. Slobin (Ed.), The crosslinguistic study of language acquisition (Vol. 1, pp. 141-254). Hillsdale, NJ: Erlbaum.

Mitchell, D.C., & Barchan, J. (1984). EXMORE: Exeter Module for on-line reading experiments. University of Exeter, Psychology Department.

Nagata, H. (1988). The relativity of linguistic intuition: The effects of repetition on grammaticality judgements. Journal of Psychological Research, 17,1-17.

Nishimura, M. (1986). Intrasentential codeswitching: The case of language assignment. In J. Vaid, (Ed.), Language processing in bilinguals: Psycholinguistic and neuropsychological perspectives (pp. 123-143). Hillsdale, NJ: Erlbaum.

Radford, A. (1986). Small children's small clauses. Bangor Research Papers in Linguistics, 1,1-38.

Randall, J.H. (1985). Indirect positive evidence: Overturning generalisations in acquisition. Mimeo.

Ritchie, W. (1978). The right roof constraint in an adult acquired language. In W. Ritchie (Ed.), Second language acquisition: Issues and implications (pp. 33-63). New York: Academic Press.

Rothstein, S. (1985). The syntactic forms of predication. Bloomington, IN: Indiana University Linguistics Club.

Schmidt, M. (1980). Coordinate structure and language universals in interlanguage. Language Learning, 30, 397-416.

Tomlin, R.S. (1986). Basic word order: Functional principles. London: Croom Helm.

Travis, L. (1984). Parameters and effects of word order variation. Unpublished dissertation, MIT.

Van Buren, P. & Sharwood-Smith, M. (1985). The acquisition of preposition stranding by second language learners and parametric variation. Second Language Research, 1, 18-46.

White, L. (1985). The acquisition of parameterized grammar: Subjacency in second language acquisition. Second Language Research, 1,1-17

White, L. (1986). Implications of parametric variation for adult second language acquisition: an investigation of the pro-drop parameter. In V.J. Cook (Ed.), Experimental Approaches to Second Language Acquisition (pp. 55-72). Oxford: Pergamon.

Timed Grammaticality Judgements of the Head Parameter in L2 Learning	Vivian Cook SLA Home
In G. Bartelt (ed.) The Dynamics of Language Processes: Essays in Honour of Hans W. Dechert, Gunter Narr 1994, 15-31