Within a formal approach, the AJ task can be viewed as an experimental paradigm designed to measure intuitive assessments on the formation of different strings of linguistic units (KELLER, 1998). As opposed to the informal procedure, the formal AJ needs a representative sample of a given group of speakers, as well as a set of sentences that represent distinct lexicalizations of the construction under scrutiny. Basically, participants are exposed to a set of sentences – target-sentences, control-sentences and distractors – and after viewing/ listening to each of them, they use some kind of scale to indicate how acceptable each sentence is. This methodology allows the application of statistical tests for the verification of hypotheses. Despite their suitability from the point of view of experimental methodology, the formal judgments of acceptability also present methodological aspects that may still need to be improved, and which are still open to debate. Among such aspects, the choice of measurement scale for trial conduction plays a leading role.

There are at least four types of scales that can be used to perform the judgment of acceptability tests: nominal scale, ordinal scale, interval scale and ratio scale (BARD et al, 1996.). As argued by kelleR (1998), the scale used in the elicitation of judgments has crucial importance, since it determines what kind of data will be obtained and what kind of mathematical operations will best suit the ensuing statistical inference procedure.

The nominal scale is used to label different items. The objective is that participants determine whether different items are the same or different with respect to certain properties. it would be possible to use such a scale, for example, to sort words in relation to which language they belong. Needless to say, the items labeled in this test cannot be ordered and, moreover, the study is limited in regard to mathematical operations. It is not uncommon to find nominal scales to describe the acceptability of linguistic constructions (acceptable X unacceptable, ✔x *, etc.), but such an operation hinders the perception of gradient differences between items, as these can only be revealed if a large number of participants are recruited and the statistical frequencies are computed.

The ordinal scale focuses on two properties: order and equivalence. This scale is used to sort different items according to the amount or intensity of a particular property. despite the fact that it is adequate to test scalar properties, this kind of scale does not assume that there is equality in the interval between the groups, as in the classification adopted by some linguists – “√” “?” “??” “?*” “*” “**” – in introspective judgments. Thus limitations regarding the application of statistical tests must be taken into consideration.

The interval scale, dissimilarly, seems appropriate for judging acceptability tests, precisely because it allows the use of several analytical tools in data treatment. This scale not only assumes that the groups are ordered, but it also assumes that the intervals between them are equal. Thus, it is possible to compare the differences between pairs of items in relation to a given property. Most studies that utilize this scale to carry out an AJ task do so with a likert Scale (lS) of 5 or 7 points. Usually, the minimum value of these scales represents total unacceptability, while the maximum value is the total acceptability. The midpoint is neutral while the other points express partial acceptability or partial unacceptability. According to FukudA et al. (2012), AJ tasks with the LS are user-friendly and generate refined results, which can receive statistical treatment. however, the author also points out that the lS cannot express all distinctions in acceptability as perceived by the participants.

The ratio scale, though, is argued by BARD et al. (1996) to be the most informative scale. While this scale has the same benefits as the interval scale, it also has more freedom and flexibility in the implementation of judgments and, consequently, more fine-grained data. Whereas points on the interval scale – usually represented by numbers – have fixed boundaries, on a ratio scale there are no pre-established lower limit values across integers, making it possible to express very subtle differences between items. Therefore, it is argued that the ratio scale does not place any restrictions on the representation of the acceptability, allowing participants to express all personally perceived differences irrespective of how minute they are.

However, the use of such a scale to measure the acceptability of different constructions is not easy to implement. in response to this challenge, Bard et al. (op. cit.) propose a method originating from psychophysics research called the “magnitude estimation” (Me), which aims to take, to the interval scale, the fine-grained data of the proportional scale (KELLER, 1998). We now turn to a description of both the details and the controversy surrounding this proposal.

The ME was proposed by Stevens (19561, apud BARD et al., 1996) in order to measure more efficiently people’s perception of different continuous physical stimuli, such as brightness and sound. These perceptions are traditionally measured by a numerical judgment, but can also be represented by lines whose lengths reflect an individual’s perception. with the Me, participants compare stimuli proportionally, instead of classifying, ordering, or labeling them as in most other scales. The judgments are made based on a comparison with a standard stimulus, i.e., the standard stimulus has the function of being the basis for estimating the magnitude of all other stimuli.

There are two variants of the Me with respect to presentation of the standard stimulus, or modulus, against which comparisons are to be made (FEITOSA, 1996; SORACE, 2010). In the first, the participants receive the standard stimulus with a fi[ed value – the modulus – and should estimate the value of the other items based on a comparison with the modulus. The modulus should preferably have an intermediate value and may or may not be visible throughout the task. in the second variant, participants attribute the value of the standard stimulus, whose availability throughout the task is also optional. imagine, for example, that the items to be compared are different intensities of the same sound. First, each participant assigns a numerical value to the first item, which will be used as the experiment modulus. if the participant assigns the value 50 for the standard stimulus and he/ she perceives the second sound as being 10 times more intense, he/ she should assign the second sound the value 500. Similarly, if the third stimulus seems to have half the intensity of the module, then it must be assigned the value 25. instead of classifying the items, the participants compare them in relation to the standard stimulus. Since there is no restriction on the numbers that can be used to express this comparison, participants in the linguistic experiment as explained earlier could not only tell whether a sentence is better or worse than another, but could also point out the numerical degree to which it is better or worse.

It should be noted that one of the peculiarities of the Me is also one of its advantages: the judgments made with this method are relative, i.e., the judgments are made through comparisons. Most other methods require an absolute judgment or, in other words, the judges must use their own points of reference to judge the intensity of a certain property – such as acceptability – in the items being tested. According to Sorace (2010), from a psychometric point of view, the participants tend to be better at relative judgments than at absolute ones.

According WESKOTT & FANSELOW (2011), the extension of this methodology to areas other than perceptual psychology began in the 1980’s with studies in the social sciences. Despite being a debated methodology in this area, it remains a methodological option for researches on subjective attitudes. Bard et al. (1996) states that a considerable number of studies have demonstrated that social opinion can also be analyzed by methods and quantitative analysis. Sorace (2010) goes on to state that ME can be applied to validate social scales, which yields a powerful quantitative measure of social opinions.

The ME has also been extended to linguistic studies. More specifically, it has been used to conduct AJ tests. This method has several aspects that make some linguists see it as gold standard for conducting this kind of task. According to Sorace & Keller (2005), the Me allows linguistic acceptability to be treated as a continuous property and is able to reveal subtle differences in judgments. kelleR (1998) agrees, stating that this method has been applied successfully by different linguists to investigate various linguistic phenomena. Moreover, Bard et al. (1996) and Sorace (2010) argue that normal adults are able to perform this kind of task reliably. They also claim that if the instructions are clear, even less experienced participants are able to provide consistent estimation of magnitude of sentences’ relative acceptability. According to these authors, even though the test may seem non-traditional at first sight, participants have the ability to express the proportionality between the sentences’ difference in acceptability just as they can express differences in sound intensity.

Sorace (2010) argues that high sensitivity; no restriction of judgment values; relative and gradient judgments; and liability to powerful statistical tests are some advantages of the Me over competitor forms of the AJ task. FuRTheRMoRe, SoRACe & kelleR (2005) advocate the use of the ME for its ability to generate fine-grained data, which enables the investigation of important linguistic aspects such as the differences between soft constraints and hard constraints.

The superiority of the Me compared to other methods, however, is still a questionable point. As noted by FUkUdA et al. (2012), problems that may be present in other scales, such as non-uniform distances between distinct points and limited representation of perceived differences do not seem to be present in the Me method. however, the learnability of a Me task may be severely jeopardized due to its counter- intuitiveness and the mathematical reasoning requirements it imposes on participants.

Therefore, studies aimed at contrasting the data from the Me with others – such as the lS – are of great importance for the validation of AJ methods. BARD et al. (1996) argue that the ME provides more informative data than those extracted from experiments with 7-point scales. WESKOTT & FANSELOW (2011) put this claim to test with a series of three experiments, which aimed to compare the degree of informativeness of a 2-point scale, a 7-point scale, and the ME. In none of the tests in the authors’ study was the ME significantly more informative than the other scales. in fact, there were data that pointed to the greater informativeness of the 7-point scale. Thus, the authors consider the claim that the Me is the best method for AJ tasks doubtful.

Fukuda et al. (2012) conducted a study divided into 3 e[periments that aimed at contrasting three AJ task methods: a binary yes/no task, a multiple-point likert scale and the Me. in general, all methods appear to be sensitive enough to capture the different degrees of acceptability with respect to specific conte[ts, such as: inversion and non-inversion structure in wh-questions; the that-trace effect; sub-e[traction from objects, subjects and wh-subjects; and the effect of subject type on wh- questions without inversion. There was data, however, that indicated that only the Me was not able to represent the difference in acceptability between wh-object sentences with and without “that”. in this case, either the Me was less sensitive than other methods or the other methods generated an inferential error.

Some authors have pointed out to aspects of the Me that may cause it not to be as superior as other methods as previously thought (FEATHERSTON, 2009; WESKOTT & FANSELOW, 2011; SPROUSE 2008, 2011). Sprouse (2008) argues that the constant repetition of the module can affect the processing mechanisms of the participants due to a priming effect. Moreover, SpRouSe (2011) suggests that participants do not seem to use the default sentence as a unit of measure. Such influence can cause changes in the patterns of acceptability assigned by the participants.

Weskott & Fanselow (2011) and Sprouse (2011) present psychophysical studies showing that people do not seem to provide the levels of accuracy advocated by the proponents of the Me method regarding the mental arithmetic involved in this task. These studies have primarily tested if particpants’ magnitude estimation of sound intensity presented commutativity and multiplicability, which are the basic assumptions of the Me. Commutativity is the property that makes the order in which successive adjustments are made irrelevant – p ★ (q ★ x) ≈ q ★ (p ★ X) – while multiplicability is the property that makes the result of two successive adjustments equal to one adjustment that is numerically equivalent to the product of the two mentioned adjustments – p ★ (q ★ x) ≈ r ★ x, when p.q = r (SPROUSE, 2011). The results of these studies indicate that the production of the magnitude of sound intensity has commutativity, but not multiplicability. These findings are interpreted by Weskott & Fanselow (2011) as causing several implications to the use of the Me in linguistic studies, especially because according to the authors the scale underlying sound intensity is less complex than the one involved in sentence AJ tasks.

Sprouse (2011) conducted a study consisting of two exeriments to investigate whether Me judgments presented commutativity. The results indicate that less than 20% of participants presented this property in their judgments, i.e., over 80% of the participants performed the task differently from the manner they were instructed. Among the justification presented by the author, there is the absence of a 0-point for the acceptability scale, which is required in this type of judgment. The author suggests that there is no reason to believe that the Me provides more significant data than other methods of AJ.

The present study brings a contribution to this debate by exploring the behavior of two AJ task paradigms in the context of l2 knowledge investigations. we now pass over to the description of the resultative construction of english, and to an exploration of the reasons why it may pose a learnability challenge to speakers of Brazilian portuguese.

According to Goldberg and Jackendoff (2004), the english resultative construction has been a major focus in research related to the interface between syntax and semantics. According to the authors, in this type of construction there are two sub-events. The verb phrase determines a sub-event, whereas the resultative construction as a whole determines the other. in order for both sub-events to occur in harmony, it is necessary that the arguments licensed by the verbs and the arguments licensed by construction share some syntactic positions. in (4), for instance, the sub-event determined by the construction is “BILL CAUSE [TULIPS BECOME FlAT]”, in which “Bill” occupies the subject position and “flat” occupies one of the internal argument positions. Such sub-event occurs through the second sub-event “BILL WATER TULIPS” whose subject coincides with the one in the first sub-event and therefore the subject position is shared. The NP “tulip” occupies the other internal argument. Thus, all thematic roles are performed harmoniously.

(4) Bill watered the tulips flat

As presented by Goldberg and Jackendoff (2004), both the semantics and the syntax of the resultative construction can vary. in turn, the authors propose that the resultatives should be considered a family, which can be divided into different sub-constructions. in this study, our focus will be only in the subconstruction instantiated by (4), which is formed by a subject (NP), a direct transitive verb (VP), a direct object independently selected by the verb (NP), and a resultative predicate expressing property (Ap).

This resultative sub-construction must be telic in order to be grammatical, but not all Aps are able to provide a sentence with this property, as maintained by Wechsler (2001). According to the author, a sentence with a durative verb, such as (5), is atelic. Adding an Ap to the sentence may make it telic (6) or keep it atelic (7). Wechsler goes on to contend that a telic event needs to have three aspects: an affected theme, a property scale, and a bound, related as follows: “Some property of the affected theme argument changes by degrees along a scale due to the action described by the verb, until it reaches a bound.” (WECHSlER, 2001, p.7). The resultative construction, therefore, requires APs that are capable of constituting a bound.

(5) he wiped the table

(6) He wiped the table dry / clean

(7) *He wiped the table wet / dirty

The only differences between (6) and (7) are the APs that constitute the resultative predicate. Both sentences have a theme – table, which is transformed by an action – wipe – and acquire a scalar property. In (6) the APs can be considered a definite bound because they are the maximum point of their own scale since “dry” and “clean” represent respectively 0% dirt and 0% liquid. The APs in (7) are different because neither “wet” or “dirty” represent the maximum value of their own scale. They basically represent any value above 0% and, in turn, cannot impose a bound for the action. in sum, in order for a sentence instantiating the resultative construction to be telic/ grammatical, it needs an Ap that represents the maximum point of a scale property.

Some studies have investigated the presence of the resultative construction in Bp. For FOLTRAN (1999), for instance, the BP resultative construction is e[emplified by sentences like (8) and (9). MARCELINO (2000), differently, argues that in Bp the resultative predicate is formed by the adverb “bem” (well) followed by an Ap that represents the change that the object NP has undergone. Moreover, this AP should contain the diminutive suffi[ “-inho”, as illustrated in (10).

(8) ele fez o chá fraco. he made the tea weak

(9) Ele construiu a casa muito grande.

he built the house very big

(10) Joana picou o papel bem picadinho.

Joana cut the paper well cut(DIMINUTIVE)

Lobato (2004) also argues for the existence of the resultative construction in BP. Besides the structure shown in (10), according to the author, there are at least four major groups within the resultative construction in Bp, which are illustrated below:

(11) Deus criou os homens fracos.

God created the men weak.

(12) A manteiga congelou torta.

The butter froze crooked.

(13) João pintou a casa torta.

John painted the house crooked

(14) Ele cortou o cabelo curto.

he cut the hair short.

MARCELINO (2014) compares these potential resultative structures in order to verify if they have the same properties of the english resultative sentences. The author proposes that these two languages differ from each other in two main aspects.

First, the english resultative sentences are formed by an activity verb, but the sentence has an accomplishment reading due to the telicity generated by the resultative predicate. In turn, telic modifiers such as “in an hour” can be added to these sentences (15). However, if the resultative predicate is removed from these resultative sentences, they become atelic and it is no longer possible to add telic modifiers such as “in an hour” (16). In BP, however, the sentences that have been presented as instances of the resultative construction do not have an activity verb. instead, they are formed by an accomplishment verb which makes the sentence telic regardless of the presence of the Aps. Thus, the presence of the Aps does not have an impact on the grammaticality status of these so-called BP resultatives, as illustrated in (17) and (18).

(15) He hammered the metal flat in an hour

(16) *He hammered the nail in an hour.

(17) Ela costurou a saia bem justinha em uma hora. She sewed the skirt well tight(DIMINUTIVE) in an hour

(18) Ela costurou a saia em uma hora.

She sewed the skirt in an hour

(MARCELINO, 2014)

Second, MARCELINO (2014) shows that BP and English respond differently to the how-test. Notice that in English the resultative predicate cannot be utilized as an answer to a how-question, i.e., a question about the manner the action was performed (19). In BP, dissimilarly, the APs can be utilized as answer to the how-question (20).

(19) How did he hammer the nail? (*flat) (slowly/ rapidly)

(20) Como ela custurou a saia? (bem justinha)

How did she sew the skirt? (well tight(DIMINUTIVE))

(MARCELINO, 2014)

MARCELINO (2014) argues that in this case we are dealing with two different constructions: the “adverbial resultatives” and “true resultatives”. The sentences that have been proposed as instances of the resultative construction in Bp are, in reality, “adverbial resultatives” because instead of denoting a resultant state, their AP modifies the resulting state, which is indicated by the verb itself. “Adverbial resultatives” can also be found in English in sentences such as (21). Sentences such as “he hammered the metal flat” are instances of the “true resultatives”. They have an activity verb and their telicity depends on the presence of the resultative predicate. “True resultatives” are licensed and productive in english, but not in Bp.

(21) How did he cut the meat? (thick)

he cut the meat thick

From a second language acquisition perspective, verification of the development of representations that support the interpretation of the English resultative by L2 users of English whose L1 is Brazilian portuguese will be of theoretical interest. The contrast between the resultative construction in english and in Brazilian portuguese, as well as the subtle semantic requirements for the licensing of such construction in english, imposes non-trivial cognitive operations to the second language learner.

Let us assume, together with MARCELINO (2007, 2014), that the “true” english resultative emerges from a positive setting of the compounding parameter. Such assumption entails that evidence of l2 learning of the english resultative by native speakers of Brazilian portuguese would constitute evidence suggestive of behavior that resembles parameter re-setting in l2 acquisition. whether or not second language learners do re-set L1 parameter values has been a core empirical question within generative studies of second language acquisition. Such question has been theoretically framed as related to the ultimate question of whether second language acquisition is guided or not by access to principles of Universal Grammar (for a review see WHITE, 2003).

In the case of the english resultative construction the task the second language learner would face outreaches the level of morphemic compositionality, however. As discussed above, there are complex semantic configurations related to event structure that need to be computed for a grammatical resultative to ensue. Therefore, not only would the learner need to figure out the right mapping of syntactic structure to semantic reading, but also he/she would need to figure out the right event structure that supports the construction.

Another empirical question in second language acquisition research is whether or not fine semantic configurations are prone to be learnable. SlABAkovA (2006, 2008) defends that there is no maturational constraint that would impede l2 learners from acquiring syntax and semantics of a non-primary language. To the researcher, the bottleneck of second language acquisition is functional morphology proper. So, according to his perspective, the learnability of the resultative construction is predicted.

Based on these facts about the resultative construction in english and Bp, as well as based on the challenges it imposes for second language acquisition, the present study sought to test the behavior of the two AJ tasks as psychometric instruments to probe the following two learnabiliy questions:

(i) Do bilinguals of BP and English with high proficiency in their L2 accept the English resultative construction?

(ii) do they learn the resultative predicate rules, therefore rejecting unlicensed APs?