An eye-tracking study of learned attention in second language acquisition

This paper investigates the limited attainment of adult compared to child language acquisition in terms of learned attention to morphological cues. It replicates Ellis and Sagarra in demonstrating short-term learned attention in the acquisition of temporal reference in Latin, and it extends the investigation using eye-tracking indicators to determine the extent to which these biases are overt or covert. English native speakers learned adverbial and morphological cues to temporal reference in a small set of Latin phrases under experimental conditions. Comprehension and production data demonstrated that early experience with adverbial cues enhanced subsequent use of this cue dimension and blocked the acquisition of verbal tense morphology. Effects of early experience of verbal morphology were less pronounced. Eye-tracking measures showed that early experience of particular cue dimensions affected what participants overtly focused upon during subsequent language processing and how this overt study resulted in turn in covert attentional biases in comprehension and in productive knowledge. Naturalistic second language acquisition (L2A) tends not to reach nativelike ability. Although it may be sufficient for everyday communicative purposes, adultacquired language predominantly includes nouns, verbs, and adverbs with closedclass items, in particular grammatical morphemes and prepositions, that fail to be put to full nativelike use (Bardovi-Harlig, 1992; Clahsen & Felser, 2006; Klein, 1998; Schmidt, 1984; vanPatten, 1996, 2006). In comprehension, L2 learners prefer lexical to grammatical cues (Lee, 1999; Lee, Cadierno, Glass, & VanPatten, 1997; Musumeci, 1989). In production, novice L2 learners make temporal references mostly by use of temporal adverbials, prepositional phrases, serialization, and calendric reference, with the grammatical expression of tense and aspect © Cambridge University Press 2012 0142-7164/12 $15.00 Applied Psycholinguistics 2 Ellis et al.: Learned attention in second language acquisition emerging only slowly thereafter, if at all (Bardovi-Harlig, 1992, 2000; Meisel, 1987; Noyau, Klein, & Dietrich, 1995). One factor determining cue selection is salience: prepositional phrases, temporal adverbs, and other lexical cues to time are pronounced in the speech stream; they tend to be multisyllabic with at least one of the primary syllables receiving primary stress. Verbal inflections, in contrast, are often subsyllabic and unstressed (consider yesterday I walked). They are often unreliable as well; for example, in English the regular past tense morpheme can be pronounced /ed/ /d/ or /t/, but present tense verbs, as well as many other words, can also end with these phonemes/ letters. These factors make grammatical cues less easy to learn that lexical cues. However, they are equally of low salience and low reliability whether the language is learned as a first language (L1) or as a second language (L2). Something else must account for the grammatical limitations of L2A. Historically, the response to this question has been in terms of transfer and the ways that prior knowledge of an L1 biases attention in the learning of an additional language (James, 1980; Lado, 1957; Odlin, 1989). Contemporary theories of L2A that incorporate notions of transfer and learned attention include the competition model (MacWhinney, 2001; MacWhinney & Bates, 1989; MacWhinney, Bates, & Kliegl, 1984), input processing theory (vanPatten, 1996, 2006), and the associative–cognitive CREED (Ellis, 2006c, 2008). Associative learning research studies the range of transfer and inhibition effects that modulate learners’ attention to input as a result of prior experience. In particular, Kamin (1969) and Kruschke (2006) describe the phenomenon of blocking. Learning that a particular stimulus is associated with a particular outcome makes it harder to learn that another cue, subsequently paired with that same outcome, is also a good predictor of it. For example, if an animal learns that conditioned stimulus (CS) X is a reliable predictor of an unconditioned stimulus (e.g., that a light reliably predicts the onset of some painful stimulus such as a shock), then it will not become conditioned to another CS, Y, or learn that any other CS predicts that unconditioned stimulus (e.g., that a bell predicts the onset of the shock the same way the light did). The prior association essentially “blocks” further associations. Blocking is an effect of learned attention (Kamin, 1969; Kruschke & Blair, 2000; Mackintosh, 1975). It is a robust and ubiquitous phenomenon across animal and human learning (Moore & Schmajuk, 2008; Rescorla & Wagner, 1972; Shanks, 1995, 2010; Wills, 2005). Ellis (2006b) reviews this phenomenon as it might apply in L2A. There are many situations in natural language where cues are redundant (Schmidt, 2001; Terrell, 1991; vanPatten, 2006) and thus, as a consequence of blocking, one of the redundant cues might be less readily learned. If a learners’ L1 experience has led them to look for a certain type of cue to interpretation, they might use these cues where available in L2A. If they do, the principles of associative learning predict that this will be to the detriment of learning other cues that might also be relevant. For example, L1-derived knowledge that there are reliable lexical cues to temporal reference (words like gestern, hier, ayer, yesterday) might block the acquisition of verb tense morphology from analysis of utterances such as Yesterday I walked. Ellis and Sagarra (2010b) explored these phenomena of learned attention in two experiments. The first demonstrated short-term instructional sequence effects in Applied Psycholinguistics 3 Ellis et al.: Learned attention in second language acquisition adults learning temporal reference in Latin using the standard blocking experimental paradigm (Kruschke, 2006) but with linguistic content. Native English speakers learned a small number of Latin expressions and their English translations. There were three groups: adverb pretraining, control, and verb pretraining. In Phase 1, adverb pretraining participants learned two adverbs and their temporal reference, hodie “today” and heri “yesterday”; verb pretraining participants learned two verbs and their temporal reference, cogito “I think” present and cogitavi “I thought” past; the control group had no such pretraining. In Phase 2, all participants were shown sentences that appropriately combined an adverb and a verb (e.g., heri cogitavi “yesterday I thought,” hodie cogito “today I think,” cras cogitabo “tomorrow I will think”) and learned whether these sentences referred to the present, the past, or the future. If they chose incorrectly on any trial, they were given feedback as to the appropriate choice. In Phase 3, the comprehension test, all combinations of adverb (hodie, heri, cras) and verb tense marking (cogito, cogitavi, cogitabo) were combined, and the participants were asked to judge, without feedback, whether each sentence referred to the past, present, or future on a scale from 1 ( past) to 5 ( future). The logic of the design was as follows. In Phase 2 every utterance contained two temporal references—an adverb and a verbal inflection. If participants paid equal attention to these two cues, then in Phase 3 their judgments should be equally affected by them. However, if they paid more attention to adverbial ( /verbal) cues, then their judgments would be swayed toward them in Phase 3. The results showed that the three groups reacted to the cues in very different ways. In two-cue sentences (where both an adverb and an inflection cued temporal information), when these cues deviated, the verb pretraining group followed the verbal cue, the adverb pretraining group followed the adverbial cue, and the control group lay in between. Multiple regression analyses, one for each group, where the dependent variable was group mean temporal interpretation for each of the Phase 3 strings and the independent variables were the information conveyed by the adverbial and verbal inflection cues, estimated the differential cue use by each of the three groups. These were, in standardized ß coefficients, adverb group time = 0.97 adverb + 0.23 verb; verb group time = 0.12 adverb + 0.97 verb; control group time = 0.60 adverb + 0.72 verb. An additional design element allowed the inference that these results illustrate attentional biases to particular dimensions of cue (adverb vs. verbal inflection) rather than to particular words. In Phase 1, participants in the adverb and verb pretraining groups learned particular constructions relating to the present and the past. There was no reference to future at this stage. Thus, whereas subsequent responses relating to past and present judgments could reflect specific prior-learned associations, responses relating to future judgments could not. The Phase 3 results showed that the adverb and verb pretraining groups were as unalike and dissociated in their performance on cras “tomorrow” and cogitabo “I will think” items referring to the future as they were on the other past and present reference ones. Ellis and Sagarra (2011) extended the experimental paradigm to allow morphological cues in a more naturally complete state where the inflections reflected tense and person (cogito, cogitas, cogitat, “I think, you think, he thinks”; cogitavi, cogitavisti, cogitavit, “I thought, you thought, he thought”; cogitabo, cogitabis, cogitabit, “I will think, you will think, he will think”) in a design otherwise parallel Applied Psycholinguistics 4 Ellis et al.: Learned attention in second language acquisition to that of Ellis and Sagarra (2010b). Under these conditions, participants overall came to rely much more on adverbial cues, with performance in the control group of time = 0.93 adverb + 0.17 verb (compare control group time = 0.60 adverb + 0.72 verb in Ellis & Sagarra, 2010b). However, against this background, again the adverb pretraining individuals paid greater subsequent attention to the adverbial cues, as the verb pretraining individuals did to the inflectional cues. Learned attention effects can result in positive or negative transfer. Experience can lead us to pay more attention to cues we have known in the past, to process them more fluently or automatically, to be more open to particular evidence, to be positively disposed to arguments (+). It can also lead us to ignore new, foreign cues, to be blind to information, to be prejudiced against beliefs or explanations (−). Although these outcomes are often related, they are logically separable effects. In our language learning experiments, adverb pretraining might cause participants to pay more attention to adverbs in future language processing, but their future sensitivity to other morphological cues could be untouched by this history. Positive effects of frequency of exposure are those of entrenchment of the practiced cue (for a review of frequency effects, see Ellis, 2002a). Alternatively, increased sensitivity to adverb cues might be accompanied by a reduced sensitivity to morphological cues, that is, blocking as discussed in Ellis’ review of learned attention (2006b). Increased reliance on one cue, means, or method tends to diminish reliance on others, but not necessarily so. Not only are these logically separable effects, but they are also empirically separable. In these experiments, entrenchment effects are indexed by increased sensitivity of the adverb group to adverbs, and increased sensitivity of the verb pretraining participants to verbs cues, relative to control participants. Blocking effects are indexed by decreased sensitivity of adverb pretraining participants to verb cues, and decreased sensitivity of verb pretraining participants to adverb cues, relative to controls. A meta-analysis of the combined results of Ellis and Sagarra (2010b, 2011) demonstrated that the average effect size of entrenchment was large (+1.23) and that of blocking was moderate (−0.52). An important question is whether these learned attention effects in L2A are overt or covert. Overt attention is the act of directing sense organs toward a stimulus source; covert attention is the act of mentally focusing on one of several possible sensory stimuli (Wright & Ward, 2008). Overt attention involves physical orienting; covert attention is a neural process that enhances the signal from a particular part of the sensory input. There are various methodologies for tracking overt attention in L2A, such as think-aloud protocols, stimulated recall, and other measures of conscious focus (e.g., Bowles, 2010; Ellis, 1994; R. Ellis, 2005; Gass & Mackey, 2000). One direct measure that is gaining in currency is that of eye tracking, which is monitoring the timing and location of each fixation during visual study (Godfroid, 2012). Covert attention cannot be measured directly, but it can be inferred by analyzing behavioral cues, such as accuracy or fluency of processing, for evidence of cue-based learning. After initial preexposure to a cue (X) that sets the bias, is the differential attention to cues X and Y co-present in the subsequent input in Phase 2 overt or covert? Do people spend more time looking at the blocking cue X and less time looking at the blocked cue Y? Further, if there are differences in overt attention to the cues in Phase 2, do these predict what is learned from this experience, as indexed by Applied Psycholinguistics 5 Ellis et al.: Learned attention in second language acquisition measures of covert attention to these cues as it affects comprehension in Phase 3 and production in Phase 4? Not only are these questions theoretically important, but they also have important pedagogical consequences. If acquisition from Phase 2 input is prejudiced as a consequence of the learner simply not looking at the blocked cues in the simultaneous input, there is instructional scope in directing their attention to these so to increase gaze fixations and dwell time. However, if acquisition is prejudiced because fixated physical input fails to become intake (Corder, 1967), the challenge of successful intervention is greater. Kruschke, Kappenman, and Hetrick (2005) demonstrated that eye tracking could be used to inform the processes of blocking in associative learning. They showed that participants’ eye gaze in a later XY training phase of a blocking experiment (the equivalent of our Phase 2) was greater for the blocking cue than the blocked cue, and that individual differences in this ratio covaried with covert indices of blocking in a subsequent testing phase (r = .48): individuals who showed stronger overt gaze preferences in the learning phase tended to show stronger blocking. We therefore adopt eye tracking here to investigate the locus of learned attention effects in L2A. As in Ellis and Sagarra (2010b), the experiment involves the learning of a small number of Latin expressions and their English translations. It investigates the effects of successive learning of different types of cue for temporal reference, adverbs (hodie “today”; heri “yesterday”; cras “tomorrow”) and verbal inflections (cogito “I think”; cogitavi “I thought”; cogitabo “I will think”). Following Kruschke et al. (2005), it uses total gaze duration to each cue as an index of overt attention to that cue. It determines if the acquisition of one set of cues is impaired if another is already known as a reliable indicator of event time.