Caudate nucleus
2. Language learning processes, language instruction and the nature and measure of language knowledge
2.1 Explicit vs implicit learning
2.4.3 Issues in GJT design
Although the GJT will not be used as a measure of implicit/explicit L2
knowledge in the experimental studies reported in Chapter 6 and 7, it will be used as a measure of morphosyntactic attainment. For this reason, it is useful to review the task in some more detail here, in particular with regard to issues relating to its design and use with adults and children.
In both adult and child L2 learning studies the grammaticality judgment test (GJT) represents one of the most widely deployed test instrument to measure language attainment. Grammaticality judgment tests can be administered auditorily and/or
activity (in the case of studies with young children). They can be manipulated in a number of ways to yield information about the knowledge of specific syntactic domains or grammar features, the type of knowledge attained, the level of knowledge
automatization and the subjective confidence or knowledge of the source of judgment. In their most common format GJTs expose participants to the same number of ungrammatical (ill-formed) and matched grammatical (well-formed) syntactic units to which a certain number of fillers may be also added to take the participant's focus away from the linguistic target of testing. Participants' performance is also compared to chance performance, to make sure that the correct responses are a reflection of genuine learning and are not simply the result of a successful guessing strategy.
As already discussed, timed and untimed GJTs have been considered to be more reliable indexes of implicit and explicit language knowledge respectively (Bialystok, 1979; Han, 2000; Loewen, 2009). For untimed GJTs some authors have suggested that correctness in the GJTs has to be interpreted differently for grammatical and
ungrammatical sentences. Correct judgment of ungrammatical sentences would depend more on the engagement of explicit language knowledge, as incorrect sentences invite reflection on the reason of the ill formedness, once the error is detected (R. Ellis, 1991). If the GJT is timed (the time allowed has ranged between 3 and 10 seconds per trial across studies), the correctness is in general expected to draw more on the learner's implicit knowledge, as sufficient time for reflection is not provided.
In addition to a limited time for response, computerized GJTs (such as the ones administered through E-Prime or similar software to program experiments) can also provide a latency measure (the time in milliseconds elapsing between the response request and the response). Further, GJTs can be designed to include additional features,
e.g. Likert scales for confidence ratings or source attributions for which latencies can also be measured.
An important point that has been raised with regards to GJT scores validity relates to the widespread use of dichotomous acceptability responses. In general, and specifically for children, it has been argued that the request of a response presenting a binary choice is biased towards a 'yes' or 'correct' response (McDaniel & Cairns, 1996). A common technique used to counterbalance this bias is the computation of so called A' (or d') scores, calculated considering the proportion of hits and false alarms in the data instead of simply reporting hits in the two categories (Saxton, Dockrell, Bevan, & van Herwegen, 2008).
Another alternative that has been proposed is the use of graded scales for grammaticality judgments. In adult studies graded grammaticality judgments have deployed magnitude or Likert-type scales with numerical values the participants are asked to select to express a more nuanced judgment compared to the binary option. More recently studies like Ambridge, Pine, Rowland, & Young (2008) have implemented this idea in a more child-friendly format presenting the points of the Likert scale on a test sheet as smiley faces (see also Theakston, 2004). In Ambridge et al. (2008) a five-point scale was used with two green smiling faces on the rightmost side and two red frowning faces on the leftmost side. The face in the middle was half green and half red with a neutral facial expression. The use of the color-coding in addition to the graded scale allowed for a combination of graded and binary acceptability judgment. The study deployed the same GJT paradigm comparing 5-6 year olds, 9-10 year olds and adults. However, since it was not clear whether the younger child group could provide a graded judgment, these children were asked to give a binary judgment first and specify a grade in the scale only subsequently. The authors were confident that both the older children
and the adults could easily perform the graded judgment task directly, without the need to provide a binary judgment first.
Metacognition and subjective confidence in children has recently received specific attention in the literature on implicit and explicit learning with the
extension/adaptation of assessment instruments used with adult participants to studies with children of primary school age (e.g., Bertels, Boursain, Destrebecqz, & Gaillard; 2015; Fritzsche, Kröner, Dresel, Kopp, & Martsch, 2012; Koriat & Ackermann, 2010). Investigating visual statistical learning in 9 year-old children and adults and in
association to a visual triplet-completion task, Bertels et al. (2015) used a binary
confidence rating with verbal labels; the participants were instructed to select "guess" if they felt they were answering at random or "remember" if they felt the choice was made on the basis of some form of recall (p. 3). In Koriat and Ackermann (2010) three groups of children of 8, 9 and 11 years of age were presented with a set of age appropriate general knowledge questions and asked to select one of two answers via a computer program. After each item they were asked how confident they were about their choice using the thermometer paradigm, whereby they had to slide a pointer on a scale, which was automatically converted into a percentage confidence score.
2.5 The nature of child language learning and language representations