Desproporcionalidad de las penas

The GJ task comprised 70 English sentences. Twenty of these sentences were grammatical and 50 were ungrammatical. Only 38 of the ungrammatical sentences

were used as experimental items. The remaining 12 ungrammatical sentences and the 20 grammatical ones were control and distractor sentences.³⁸

All the test sentences, regardless of their grammaticality or function (experimental, control, or distractor), contained embedded clauses. The embedded clauses were of two types: (a) adverbial of time and (b) finite complement of V.

Examples of the test sentences are given below in 1–3:³⁹ 1. Experimental sentences:

a. Sentences with embedded null subjects in adverbial clauses:

*He drives whenever goes to work.

b. Sentences with embedded null subjects in complement clauses:

*John told me that found his money.

c. Sentences with missing 3^rd-person subject–verb agreement:

*Anne says that John sleep in class.

2. Control sentences

a. Sentences with overt embedded subjects in adverbial clauses:

The children played football until they left.

b. Sentences with overt embedded subjects in complement clauses:

The children think that she went to work.

c. Sentences with appropriate 3^rd-person subject-verb agreement:

Sara told me that her aunt often reads the newspaper after dinner.

3. Distractor items:

a. Sentences with null embedded objects:

*The car stopped before John filled with petrol.

b. Sentences with overt embedded objects:

The man says that the glass broke when his son played with it.

38 These 12 sentences will serve as experimental items for another study.

39 See appendix 1 for a complete listing of the test sentences presented in French version of the test.

As can be seen in 1–3, the distractor and control sentences resembled the experimental sentences in syntactic complexity. The grammatical control sentences covered the corresponding acceptable structures of the experimental ones in English. However, the ungrammaticality of all the distractor sentences, unlike the experimental sentences, results in the null realisation of the embedded object pronoun.

The experimental sentences that tested the learning and unlearning of the null subject and mismatched verb agreement were further broken down into different subgroups, each with different tokens. Those meant to test the learning and unlearning of overt subjects in English were divided into two structural types:

sentences with verbal-finite-complement or adverbial-of-time embedded clauses.

They were then divided into two subgroups, each with different tokens based on the position of the potential referential antecedent of the missing subject pronoun in the lower clause, as follows:

Subgroup 1 comprised 24 sentences with null embedded subject pronouns controlled by higher referential antecedents (pronominal subjects, proper nouns, or noun phrases) within the sentences in the main clause. Of these, 18 were sentences with embedded adverbial clauses while the other six were sentences with embedded complement clauses. This is illustrated by the following examples:

4. a. *The manager can see you when finishes his work.

b. *The student told me that finished his homework.

It can be seen that from the above examples in (4a, b) that both types of sentential structures share the same verb, finish, in the second clauses. This holds true with all other verbs used in the complement clauses in the test; in other words, every verb used in a complement clause is also used in an adverbial clause. The reason behind this is to see whether participants’ performance with complement clauses yields results similar to or different from those of the adverbial clauses (see the section 4.3.3, “Piloting and Trying Out the Instrument”).

Subgroup 2 comprised four sentences with null embedded subject pronouns in the adverbial clauses controlled by higher antecedents outside the sentences,⁴⁰ such as the following:

5. *I will not leave his office until pays me the money.

The sentences that were intended to test the knowledge of subject-verb agreement were also split into two sets with five tokens each, according to the animacy of the embedded subjects:

Type 1: Missing subject-verb agreement in sentences with animate imbedded subjects:

6. a. *Mary says that John sleep in class.

b. *Bill claimed that the professor always give too many low marks.

Type 2: Missing subject-verb agreement in sentences with inanimate imbedded subjects:

7. a. *Linda explained that the library open late at night.

b. *John though that the cinema often show films in the afternoon.

As can be seen, the embedded subjects and their verbs in some of the sentences in 6 and 7 are separated by an intervening adverb, such as in 6b and 7b but not in 6a and 7a. The purpose of this design was to determine which type of sentences the participants tended to accept or reject more often – that is, to draw a

40 Sentences with complement clauses that had sentences outside antecedents were avoided because in most cases these sentences require discourse contexts for the embedded subject to be interrupted to have external antecedents. It was found in the piloting that some participants, when correcting sentences such as that presented in (i), assumed that the missing embedded subject had an external referential outside the whole sentence.

These participants would correct the sentence “(i) *They believed that got better jobs” as

“They believed that we got better jobs.”

line between failure, optionality, and native-like attainment in L2A of morpho-syntactic features.⁴¹

So far, it can be observed from the above examples (1–7) that the experimental sentences investigating the acquisition of overt subject pronouns that include null definite 3^rd-person embedded subjects are more common than those including null definite 1^st- or 2^nd-person pronouns in the same syntactic position. There were only six sentences containing 1^st- or 2^nd-person covert pronouns, while the other 22 sentences included null 3^rd-person embedded pronouns. The reason for my interest in investigating the acquisition of 3^rd-person subjects is the cross-linguistic variations among the languages under investigation in the obligatory/optional realisation of this particular embedded subject pronoun (see section 2.3).

It is worth mentioning that the number of test sentences, the selected structures with either adverbial or complement clauses, and the number of items/tokens within each type or subtype of experimental sentences were not determined haphazardly.

These were composed based on information from previous research and initial empirical data elicited from the results of the pilot studies, as in the examples that follow:

i. The total number of sentences:

To eliminate (or at least reduce) the effect of fatigue on the judgement task, the total number of sentences to be judged was reduced to 70, following Cowan and Hatasa (1994), who argued against using more than 72 sentences.⁴²

41 Empirical research that has considered the accessibility of the universal inventory of features in adult L2A within the minimalist programme (MP) framework has yielded contradictory results. As a result, a number of hypotheses that account for the failure of or persistent problems in L2A features have emerged, such as, the missing surface inflection hypothesis (Prévost and White, 2000), the representational deficit hypothesis (Hawkins, 2005; Hawkins and Hattori, 2006), and the feature reassembly hypothesis (Lardiere, 2008, 2009).

42 It should be mentioned that this is not always the norm in acquisitional studies.

Researchers vary in the number of test sentences they give participants. Some studies

ii. The selected sentential structures:

My concentration on the two selected sentential structures (sentences with either complement or adverbial clauses) was grounded in the following facts:

a. The overt/null realisation of the subject pronouns in embedded clauses varies among the languages under investigation.

b. The structures of such clauses are neither so simple that a participant is aware of the purpose of the experiment nor so complex that there is a risk that the participant will reject them because of processing difficulties. This claim of relative ease of processing structures of such clauses, although the L2A of complex clauses in English has received little attention,⁴³ can be based on L1 acquisition research showing that such clauses are easily acquired and “need not present serious parsing problems for children”

(Bowerman, 1979, p. 294). For example, whereas several L1 acquisitional studies (e.g., Limber, 1973; Bloom, 1991; Diessel and Tomasello, 2001;

Diessel, 2004; and Kidd, Lieven, and Tomasello, 2005) have shown that children start to construct complement clauses early, at around age 2, others (Clark, 1970, 1973; Hood and Bloom, 1979; Bloom, Lahey, Hood, Lifter, and Fliess, 1980; Silva, 1991; Diessel, 2004) have noticed that children at age of three are able to make and comprehend sentences with adverbial clauses.

These latter studies have also observed that adverbial clauses of time introduced particularly by subordinating conjunctions such as before, after, when, until, and while are among the first to appear in children’s speech,

have been conducted with only 24 sentences (Gass, 1994); others have conducted tests with 282 sentences (Johnson and Newport, 1989). Because of the practicalities of the testing conditions (see section 4.4.4 regarding the test instructions and procedures), I decided not to include more than 70 sentences to avoid fatiguing the participants, even though providing them with longer tests may have increased the reliability of the test results as long as the variables affecting their judgements were controlled.

43 I am not aware of any study that has been done so far to examine the learning of such clauses in particular – unlike, for example, adjectival clauses.

before others kinds of adverbial clauses of condition, contrast, purpose, result, and so on.

iii. The subtotal of items/tokens within each type and subtype of the test sentences:

In the initial pilot study, the participants were far more likely to accept ungrammatical sentences with embedded null subjects in adverbial clauses than in complement clauses; therefore, more sentences with adverbial clauses compared to complement clauses were included (18 vs. 6) (for more details, see section 4.3.3). The remaining 10 experimental sentences were included to test issues related to licensing the null subject in second-language acquisition (SLA).

These 10 and the other 12 ungrammatical sentences with null embedded objects served as distractors from the sentences investigating the knowledge of null-subject parameter values as well as from one another. This way, the participants were unable to discern what the study was investigating, increasing the validity of the test. The other 20 grammatical sentences, referred to above as the control items, allowed comparison of the experimental sentences because a rejection of the ungrammatical sentences did not necessarily mean acceptance of the grammatical ones (i.e., successful acquisition).

This task required the participants to judge the test sentences on a 4-point scale:

clearly correct, clearly incorrect, possibly incorrect, and I don’t know. Learners’ intui-tions in SLA research are reported interchangeably using a variety of terms – grammatical, ungrammatical, acceptable, unacceptable, correct, incorrect, good, bad, etc. Though there are theoretical distinctions drawn between these terms – see Birdsong (1989) – the rationale for choosing the terms correct and incorrect to report intuitions in this study was the likelihood that L2 learners, especially those who are still taught via the use of grammar-translation methods, are used to such terms.

The participants were instructed to judge whether the given sentence was acceptable by indicating clearly correct or clearly incorrect, but only if they were confident in their perception of the sentence. If the participant felt there was an error in the sentence but was not certain, he or she was asked to judge the sentence as possibly incorrect.

If the participant had no idea about the answer, he or she was instructed to choose don’t know. (See section 4.3.4.) The scale is illustrated below:

1. Bill wondered where was Mary going shopping.

 Clearly correct  Clearly incorrect

 I don’t know  Possibly incorrect

It should be: ____________

The words “It should be” followed by a blank space were provided, following Schütze’s recommendation (1996), to prevent participants from making contextless judgements. In a case where the response was clearly or possibly incorrect, the participant was asked to underline the perceived or possible errors in the sentence and provide a correction in the given space.

This procedure enhanced the reliability of the gathered data because it allowed the researcher to consider the correct unexpected responses, such when a respondent rejected a sentence based on a reason that was not actually related to its grammatical incorrectness. In this way, the possibilities of random and careless responses were minimised.⁴⁴ Besides helping to address these cases,⁴⁵ the double-check procedure (making judgements and providing corrections where required as instructed) assured the researcher that the data provided a relatively accurate reflection the learner’s IL if the other performance factors were controlled for.

The 4-point scale in its new format presents several advantages compared to many other rating scales that have been used in linguistic or related research (for a detailed discussion about the different rating scales, see Schütze [1996] and Sorace [1996]). The 4-point scale attends to some of the inherent problems and limitations commonly associated with these kinds of rating scales. Before discussing the advantages, we will consider some of the problems associated with the other rating

44 This is because uncooperative participants might not correct all the sentences they mark as ungrammatical.

45 See section 4.3.6 for a detailed discussion about the sentences marking criteria.

scales. For the sake of illustration, it is useful to base the critical discussion on a few recent studies.

A binary, 2-point scale (e.g., grammatical vs. ungrammatical) is usually avoided in SLA research.⁴⁶ This is because L2 learners are usually asked to make judgements during stages of acquisition at which their knowledge about certain elements of the target grammar is incomplete or even totally absent; this status is typically referred to in the literature as “indeterminacy”.

Indeterminacy in a learner’s developing grammar has led some linguists (e.g., Coppetiers, 1987; Bley-Vroman, Felix, and Ioup, 1988) to include a third intermediate option in their scales: usually, the not sure selection. The analysis of this third option can be problematic, however. While Coppetiers’s solution is to consider not sure responses as meaning “correct”, Bley-Vroman et al. consider these responses as meaning “incorrect”. Both solutions create serious methodological problems that can affect the reliability of the scores. One of these problems concerns equalizing certainty with uncertainty (or doubt) regarding the student’s feelings about the judged sentences.

In other words, judging a sentence as grammatical or ungrammatical reflects a high degree of confidence about it, while judging it with not sure reflects much less confidence. Keeping in mind that the goal of psycholinguistic research is to measure IL, this forces us to ask what these not sure responses really reveal to us about grammar development. Other groups of researchers who have used 3-point scales have presented another solution to this analysis problem by excluding all of the not sure responses from the analysis. This solution, however, even if it may produce more reliable scores than the other two do, still falls short of producing accurate data, raising the same question as with regard to Coppetiers’s and Bley-Vroman’s solutions. What, exactly, do not sure responses tell us about grammar development?

Furthermore, another serious problem can arise when respondents give too many not sure responses.

46 Such two-response scales are relatively common in first language syntactic research.

Such problems and pitfalls associated with the 3-point scale have led some linguists (e.g., Schachter and Yip, 1990; Schachter, 1990) to use a 4-point scale ranging from, for example, clearly correct to possibly incorrect, in agreement with researchers who treat acceptability as a gradient concept (cf. Sorace, 1996;

Tremblay, 2005) in the sense that sentence acceptability depends partly on the strength of the learner’s preference regarding how to say it.⁴⁷ Yet the results analyses gathered using these scales are still problematic. Schachter (1990), for example, counted the possibly correct option as if it were the clearly correct option and the possibly incorrect as if it were the clearly incorrect option, allowing for no distinction between the possibly and clearly categories. Conversely, Schachter and Yip’s (1990) results reflect the separateness of these four options. Such unjustified analyses force us to consider the value of including options that cannot tell us anything about the learner’s IL since the options are not part of the analysis in any real way, such as in Schachter’s study. How do we, then, interpret the two possibly (in)correct options in light of Schachter and Yip’s analysis? In addition, what if none of the options applies, such as in the case where the given sentence is beyond the level of the learner, and therefore not part of his or her IL?

The same problems arise with five-point and other multipoint scales used in GJ tasks. In addition, these scales place a greater difficulty on the learner to choose from

47 More importantly, acceptability of a sentence depends on other factors, including the sentence’s grammaticality, the context in which it is uttered, and whether it is difficult to parse. Thus, it can be said that not every sentence, even if it is well formed, is considered acceptable by all learners (or even by all native speakers of that language), and not every ungrammatical sentence is considered unacceptable by all learners. Note that the notion of acceptability differs from the notion of grammaticality; acceptability refers to the native speaker’s or L2 learner’s intuitional judgement about a sentence, whereas grammaticality is a theoretical term used by linguists or grammarians to establish whether a sentence conforms to the requirements of the grammar of the given language.

Chomsky (1965) also stated that “acceptability is a concept that belongs to the study of performance, whereas grammaticalness belongs to the study of competence. . . . Grammaticalness is only one of many factors that interact to determine acceptability” (p.

11). For more detailed information about the different between these two notions, refer to Haegeman (1994) and Adger (2003).

multiple options in a task that is already highly effortful. Consider the following 7-point scale used in Gass (1994), where she asked her participants first to judge categorically the (un)grammaticality of test sentences and then assess their degree of confidence or doubt regarding each sentence they judged:⁴⁸

–3 –2 –1 0 +1 +2 +3

definitely incorrect unsure definitely correct

It is very possible that some (if not all) L2 learners find it difficult to differentiate between the middle options of Gass’s scale – for example, between +2 and +1 or – 2 and –1, or even between –1, 0 and +1.

Compared to other scales, however, data obtained from such multipoint scales would produce more complications. Such complications emerge from the same questions raised above with Schachter and Yip’s (1990) and Schachter’s (1990) studies, but in greater degree: How should a linguist interpret each symbol (number) that indicates more or less acceptability compared to the other ones? More precisely, how is it possible to map territories between all the symbols, especially the ones in the middle? Do these obtained scores reflect the learners’ abstract syntactic knowledge? And what can such a score tell us in general about the process of SLA?

Therefore, we can conclude that the results of previous studies discussed here, all designed to reflect L2 grammatical competence, are questionable, affected by a diverse number of factors including the type of measurement scale and response format used. Such a conclusion poses a serious challenge to linguists to find an alternative rating scale that can overcome these methodological problems and pitfalls. The only way to do so is by using a new rating scale that allows sharp lines

48 The scope of this section does not allow for further discussion of the problems connected with all of the multipoint scales used in research.

to be drawn between the learner’s certainty, doubt, and lack of knowledge reflected in his or her judgements.

The scale used in the present study managed to map the territory between the

The scale used in the present study managed to map the territory between the