Noticing: One Concept, Two Faces

The general research question we address in this article is whether second language (L2) learners pay more attention to unknown words during leisure reading and, if so, whether or not the increased amount of attention given to those words positively influences the likelihood of their retention. The impetus for the study was Richard Schmidt’s noticing hypothesis (Schmidt, Reference Schmidt1990, Reference Schmidt and Ellis1994, Reference Schmidt and Schmidt1995, Reference Schmidt and Robinson2001). In its original, strong form, this hypothesis took noticing to be the “necessary and sufficient condition for the conversion of input to intake” (Schmidt, Reference Schmidt1990, p. 129); that is, novel linguistic material (e.g., a new word) that is in the input will not be represented mentally (i.e., become intake) unless the L2 learner notices it. Intake is the representation of a stimulus in long-term memory, however temporary, and is a prerequisite for further processing and integrating the newly represented language material into the interlanguage system (Wong & Simard, Reference Wong and Simard2000). In later publications, Schmidt (Reference Schmidt and Ellis1994, Reference Schmidt and Schmidt1995, Reference Schmidt and Robinson2001) has upheld a weaker version of the hypothesis in which noticing is said to be at least facilitative of learning.

Although now a fixture in the SLA literature (e.g., de Bot, Lowie, & Verspoor, Reference de Bot, Lowie and Verspoor2005; Dörnyei, Reference Dörnyei2009; R. Ellis, Reference Ellis2008; Gass & Selinker, Reference Gass and Selinker2008; Lightbown & Spada, Reference Lightbrown and Spada2006; Ortega, Reference Ortega2009), the term noticing is not at all common usage in the neighboring fields of cognitive psychology and bilingualism (see Truscott & Sharwood Smith, 2011). Rather, in these fields, learning and memory phenomena are frequently studied from the perspectives of attention and awareness (or consciousness; e.g., Baars & Gage, Reference Baars and Gage2010; Baddeley, Reference Baddeley2007; Cowan, Reference Cowan1995). When such a view is adopted, it turns out that noticing as described by Schmidt (Reference Schmidt1990, Reference Schmidt and Ellis1994, Reference Schmidt and Schmidt1995, Reference Schmidt and Robinson2001) is actually a hybrid concept because it entails both attention and awareness. According to Schmidt (Reference Schmidt and Schmidt1995), “a low level of awareness, called here ‘noticing,’ is nearly isomorphic with attention, and seems to be associated with all learning” (p. 1). In a more recent publication, Schmidt (Reference Schmidt and Robinson2001) places greater emphasis on attention. He contends that “conscious and unconscious processes probably interact in all domains of language, but that there is little evidence for learning without attention (one reading of ‘unconscious’) in any of them” (p. 4).

Although intuitively appealing, Schmidt’s ideas about the practical equivalence of attention and awareness present a conceptual challenge. As Cowan (Reference Cowan1995) observes, “attention typically is used as a relative term, whereas consciousness typically is used as an absolute term” (p. 200). Put differently, although one can allocate different amounts of attention to a stimulus, people usually consider themselves to be either conscious of a piece of information or not. At a theoretical level, then, we are confronted with a mapping problem: It is as of yet unclear how the continuous variable of attention should be linked with the dichotomous (or at least discrete) variable of awareness, and vice versa.

In the field of SLA, it is empirical evidence of the role of awareness that has been presented in the majority of noticing studies to date (see the next section for a review). The focus of the present article, however, is noticing as attention. We use a quantitative measure of the attention given by L2 readers to novel words and gauge the degree to which this measure is associated with learning. This does not mean that we discard the role of awareness in learning. In fact, we hope that the procedure we describe will pave the way for the triangulation of attention and awareness measures. This, in turn, would help to evaluate whether or not it might be useful—for practical purposes as well as theoretical ones—to tease apart attention and awareness in the construct of noticing as it is applied in SLA. If attention and awareness turn out to be flip sides of the same noticing coin, we would still need to determine how the two mechanisms should be mapped onto each other.

The present study uses Robinson’s (Reference Robinson1995, Reference Robinson, Doughty and Long2003) model of memory, which builds on Cowan (Reference Cowan1988, Reference Cowan1995), to account for the constitutive role of attention in noticing. Robinson conceives of noticing as detection plus rehearsal in working memory. Detection is an attentional function (see Posner & Rothbart, Reference Posner, Rothbart, Milner and Rugg1992; Tomlin & Villa, Reference Tomlin and Villa1994) that amounts to the cognitive registration of stimuli in short-term memory. The initial memory encoding of a detected stimulus is further enhanced by means of subvocal rehearsal mechanisms. Rehearsal can take one of two forms (see Craik & Lockhart, Reference Craik and Lockhart1972; Craik & Tulving, Reference Craik and Tulving1975): either maintenance rehearsal, which consists of “data-driven, instance-based processing” (Robinson, Reference Robinson, Doughty and Long2003, p. 654), or elaborative rehearsal, which involves “conceptually-driven, schema-based processing” (Robinson, Reference Robinson, Doughty and Long2003, p. 654). Importantly, either type of rehearsal is “a consequence of the allocation of [attentional] resources to fulfil task demands” (Robinson, Reference Robinson1995, p. 297, citing Baddeley, 1986, p. 99) and is believed to give rise to awareness. Finally, the traces left by this additional processing (i.e., rehearsal) are what constitute long-term memory representations (i.e., intake). Given that rehearsing or recycling information mentally takes time, we can assume that, all other things being equal, the amount of time spent processing a novel linguistic element (such as a word [e.g., ox] or a morphological feature [e.g., the unusual plural in oxen]) is a measure of the amount of cognitive engagement with the material. In keeping with the noticing hypothesis, we can then predict that the longer stimuli are attended to, the more likely they are to leave attestable long-term memory traces.

Operationalization and Measurement of Noticing

Given its early position in the chain of learning subprocesses, the outcome of noticing (i.e., intake) is particularly susceptible to time-based decay. Without reactivation, the newly created memory trace will fade away. Thus, a delayed, product-oriented measure of intake may be invalid as a measure of noticing, as it may be unable to detect whether noticing occurred. This provides a strong argument for using online, process-oriented techniques to gauge noticing, which permit data collection while the participants are engaged with the L2 materials. The following techniques have been employed in SLA studies to measure noticing online: (a) note taking (Hanaoka, Reference Hanaoka2007; Izumi, Reference Izumi2002), (b) underlining (Izumi & Bigelow, Reference Izumi and Bigelow2000; Izumi, Bigelow, Fujiwara, & Fearnow, Reference Izumi, Bigelow, Fujiwara and Fearnow1999; Uggen, Reference Uggen2012), and (c) verbalizations (e.g., Alanen, Reference Alanen and Schmidt1995; Hama & Leow, Reference Hama and Leow2010; Jourdenais, Ota, Stauffer, Boyson, & Doughty, Reference Jourdenais, Ota, Stauffer, Boyson, Doughty and Schmidt1995; Leow, Reference Leow1997, Reference Leow2000; Rosa & Leow, Reference Rosa and Leow2004; Rosa & O’Neill, Reference Rosa and O’Neill1999; Sachs & Polio, Reference Sachs and Polio2007). We briefly review each method so as to explain how noticing has been operationalized within each line of inquiry and if the interpretation of findings distilled from these different methods has focused on noticing as attention, noticing as awareness, or an undefined mix of both.

Among the studies that have looked into the noticing function of output (see Swain, Reference Swain, Cook and Seidlhofer1995, Reference Swain, Doughty and Williams1998, Reference Swain and Hinkel2005),^{Footnote 1} two have operationalized noticing as self-reports through note taking. In Hanaoka (Reference Hanaoka2007), participants were instructed to take notes of whatever problems they noticed while writing a story and while subsequently comparing their stories to two native-speaker models. Izumi (Reference Izumi2002) asked English as a second language (ESL) students to note down during reading “any and every word” that they found “important, necessary or useful” (p. 553) for either text comprehension or text reconstruction purposes. Despite the similarity of the tasks and chosen methodology, the authors do not concur in their interpretation of the data. Whereas Izumi (Reference Izumi2002) frames his findings almost entirely in terms of attentional processes (see, e.g., pp. 552, 568), Hanaoka (Reference Hanaoka2007) does not attempt to distinguish in his measurement between attention and awareness (see, especially, p. 464).

Compared to taking notes, underlining the relevant parts of written input seems a faster and less intrusive task. Three studies employed the underlining technique to tap into the noticing function of output. Izumi and Bigelow (Reference Izumi and Bigelow2000), building on Izumi et al. (Reference Izumi, Bigelow, Fujiwara and Fearnow1999), requested that participants “underline the word, words or parts of words” in a text passage that they felt were “particularly necessary” (p. 250) to complete a subsequent writing or comprehension task successfully. Although the authors discuss their results exclusively in terms of their participants’ attentional foci (see pp. 255–257), they claim that their methodology is suitable for measuring Schmidt’s noticing construct (which also involves awareness) “because underlining was assumed to involve at least a minimum level of awareness” (p. 250). Uggen (Reference Uggen2012), a recent conceptual replication of Izumi and Bigelow (Reference Izumi and Bigelow2000), also took underlining to reflect first and foremost learners’ attentional processes.

Finally, the majority of studies that have attempted to gauge noticing online have utilized concurrent verbalizations, or think-aloud protocols (e.g., Alanen, Reference Alanen and Schmidt1995; Hama & Leow, Reference Hama and Leow2010; Jourdenais et al., Reference Jourdenais, Ota, Stauffer, Boyson, Doughty and Schmidt1995; Leow, Reference Leow1997, Reference Leow2000; Rosa & Leow, Reference Rosa and Leow2004; Rosa & O’Neill, Reference Rosa and O’Neill1999; Sachs & Polio, Reference Sachs and Polio2007). Think-aloud protocols are a learner’s real-time verbal report of the thoughts that he or she has while carrying out a particular task. Depending on the nature of the information (i.e., verbal or nonverbal) and the amount of detail to be verbalized, different types of verbalization have been distinguished (Ericsson & Simon, Reference Ericsson and Simon1993). All of the SLA studies cited have used the least intrusive Type I or nonmetalinguistic verbalization, in which participants are required to simply say their thoughts out loud while performing the primary language task. For example, in Leow (Reference Leow1997, Reference Leow2000), L2 Spanish students were instructed to speak their thoughts aloud while completing the preterit endings of mostly irregular (i.e., stem-vowel-changing) verbs in a crossword puzzle. Participants’ verbal reports served as the main measure to classify each individual as either more or less aware (Leow, Reference Leow1997) or aware or unaware (Leow, Reference Leow2000) of the irregular stem vowels, after which the pre- to posttest learning gains for each group were compared.

Verbal reports in noticing studies have been interpreted primarily with reference to awareness rather than attention (e.g., Hama & Leow, Reference Hama and Leow2010; Leow, Reference Leow1997, Reference Leow2000; Rosa & Leow, Reference Rosa and Leow2004; Rosa & O’Neill, Reference Rosa and O’Neill1999; Sachs & Polio, Reference Sachs and Polio2007; but see Alanen, Reference Alanen and Schmidt1995, for a counterexample). In particular, the verbal nature of think-aloud protocols seems especially suited to study the qualitative aspects of awareness, be they expressed as awareness levels (Leow, Reference Leow1997; Rosa & Leow, Reference Rosa and Leow2004; Rosa & O’Neill, Reference Rosa and O’Neill1999) or in terms of the content or “quality” of a verbal episode (Sachs & Polio, Reference Sachs and Polio2007, pp. 79–80). Furthermore, verbal reports have a long tradition in cognitive science as a behavioral index of conscious experience (see, e.g., Baars, Reference Baars1988, Reference Baars1997; Dehaene & Changeux, Reference Dehaene, Changeux and Gazzaniga2004; Dehaene & Naccache, Reference Dehaene and Naccache2001; Gazzaniga, LeDoux, & Wilson, Reference Gazzaniga, LeDoux and Wilson1977; and Weiskrantz, Reference Weiskrantz1997), given that “reportability [is] a key property of conscious representations” (Dehaene & Changeux, Reference Dehaene, Changeux and Gazzaniga2004, p. 1145).

The aim of this brief review of online measures (namely, note taking, underlining, and thinking aloud) has been to uncover the frequent disagreement among researchers about which cognitive process—attention or awareness—a given measure of noticing probes primarily (cf. differential interpretations of written notes by Hanaoka, Reference Hanaoka2007, and Izumi, Reference Izumi2002, or of verbal protocols by Alanen, Reference Alanen and Schmidt1995, and other researchers). Only for underlining do researchers seem to concur in their interpretation of the data as reflecting attention. Although attention and awareness co-occur to some extent on a practical, behavioral level (James, Reference James1890; Koch, Reference Koch2004; Posner, Reference Posner1994; Schmidt, Reference Schmidt and Schmidt1995; Wyart & Tallon-Baudry, Reference Wyart and Tallon-Baudry2008), we believe that researchers should be as explicit and precise as possible regarding which process—attention or awareness—a given measure aims to gauge. In this vein, we contend that, by virtue of their verbal nature, participant reports of their thought processes seem better suited to measure awareness than attention. Furthermore, if verbal reports are chosen as data, we recommend collecting information about participants’ conscious processing in spoken rather than written form, as the former is less time consuming and “physically demanding” (Hanaoka, Reference Hanaoka2007, p. 464) for the participants to produce and will, therefore, be less likely to interfere with the primary language task.^{Footnote 2}

For the final online measure of noticing—underlining—the consensus has been that it is indicative of attention. Izumi and Bigelow (Reference Izumi and Bigelow2000) argue that the ease and speed of underlining may benefit the completeness of the measure as an index of attention but may be detrimental to its precision, given that the “relative ease [of underlining] may mean that one underlines items not perceived as particularly important” (p. 270). Even if learners underline the forms to which they attend and do not underline the forms to which they do not attend, the act of underlining cannot inform us about different degrees of attention that participants pay to a given form. Put differently, underlining, at best, indicates the locus of attention, but it never quantifies attention. In what follows, we propose an alternative online technique for measuring noticing as attention—namely, eye-tracking, which, we think, not only performs well on the criteria of completeness and precision but also provides a measure of the amount of attention that participants allocate to target forms in the input.

An Alternative Technique: Eye-Tracking

Eye-tracking is the online registration of a participant’s eye-movement behavior, in particular (a) their eye fixations (i.e., where and for how long a participant is looking) and (b) their saccades or eye movements (i.e., where the eyes move next). The use of eye-tracking data in psychological research rests on the assumption of an “eye-mind link” (Reichle, Pollatsek, & Rayner, Reference Reichle, Pollatsek and Rayner2006, p. 4)—that is, the existence of a close relationship between the eyes and the mind. Under this assumption, overt attention (as manifested by the exact eye location) and covert attention (mental focus) are tightly linked, and cognitive processing is considered a major determinant of when and where the eyes move during complex task performance, such as reading (Just & Carpenter, Reference Just and Carpenter1980; Rayner, Reference Rayner2009).

In reading research, a host of lexical variables have been shown to influence fixation times. The “big three” predictors are frequency, contextual constraint or predictability, and word length (Kliegl, Nuthmann, & Engbert, Reference Kliegl, Nuthmann and Engbert2006, p. 13). Other potential confounds include word familiarity, age of acquisition, part of speech, concreteness (i.e., imageability), and frequency in spoken language (see Liversedge & Findlay, Reference Liversedge and Findlay2000; Rayner, Reference Rayner1998, Reference Rayner2009; Starr & Rayner, Reference Starr and Rayner2001, for reviews). The effect of word frequency has been observed consistently and across a wide variety of contexts: When other factors are held constant, readers spend more time fixating low-frequency (i.e., difficult) words than high-frequency (i.e., easy) words (e.g., Kliegl, Grabner, Rolfs, & Engbert, Reference Kliegl, Grabner, Rolfs and Engbert2004; Rayner & Duffy, Reference Rayner and Duffy1986; Rayner, Sereno, & Raney, Reference Rayner, Sereno and Raney1996; White, Reference White2008). Interestingly, especially with a view to using eye-tracking to capture attention, the frequency effect is attested even if the word on which the reader fixates disappears after 60 ms. This has been established in disappearing-text studies (Liversedge et al., Reference Liversedge, Rayner, White, Vergilino-Perez, Findlay and Kentridge2004; Rayner, Inhoff, Morrison, Slowiaczek, & Bertera, Reference Rayner, Inhoff, Morrison, Slowiaczek and Bertera1981; Rayner, Liversedge, & White, Reference Rayner, Liversedge and White2006; Rayner, Liversedge, White, & Vergilino-Perez, Reference Rayner, Liversedge, White and Vergilino-Perez2003) and has been instrumental in discriminating between models of eye-movement control that make different predictions about the influence of cognitive factors on fixation times (see Reichle et al., Reference Reichle, Pollatsek and Rayner2006).

In the disappearing-text paradigm, each word disappears (or is masked) 60 ms after the onset of fixation. Surprisingly, this manipulation does not disrupt the reading process: Participants read equally fast and with the same level of comprehension as in normal text conditions (Liversedge et al., Reference Liversedge, Rayner, White, Vergilino-Perez, Findlay and Kentridge2004; Rayner et al., Reference Rayner, Inhoff, Morrison, Slowiaczek and Bertera1981, Reference Rayner, Liversedge, White and Vergilino-Perez2003). Specifically, fixation times are still influenced by word frequency: Even though the word is no longer there after 60 ms, the eyes will remain in the empty space longer for low-frequency words than for high-frequency words. Rayner et al.’s (Reference Rayner, Liversedge and White2006) interpretation of these findings is that 60 ms is sufficient to encode the word; it approximates the duration of the “eye-to-brain lag” (Reichle et al., Reference Reichle, Pollatsek and Rayner2006, p. 5). Once the visual information has been transmitted to the brain, higher-level cognitive processing and programming of the next saccade can occur both in parallel to and in the absence of the fixated word. Hence, Rayner et al. (Reference Rayner, Liversedge and White2006) claim that lexical processing, rather than low-level visual or oculomotor factors, is the engine that drives the eyes through the text. In short, the findings of the disappearing-text studies lend strong empirical support to the eye-mind link and to the validity of eye movements as a measure of attention.

The experiment we report on in the present study was set up to measure the extent to which advanced L2 learners’ attention to novel words while reading (as measured by their eye-fixation durations) predicts their ability to recognize these words on an unannounced posttest. Although still a relatively novel technique in SLA, the number of studies using eye-tracking to investigate L2 phenomena has grown in recent years (e.g., Godfroid, Winke, & Gass, Reference Godfroid, Winke and Gass2013). At the time of acceptance, however, only two published studies had piloted eye-tracking procedures to investigate noticing (but see Godfroid & Uggen, Reference Godfroid and Uggen2013; Winke, Reference Winke2013). Godfroid, Housen, and Boers (Reference Godfroid, Housen, Boers, Pütz and Sicola2010) presented preliminary evidence of increased visual attention to novel words during L2 reading and discussed how eye-tracking may enrich current approaches to measuring noticing. Smith (Reference Smith and Hult2010) measured eight ESL learners’ noticing of “intensive recasts” (p. 85) of whatever lexical and grammatical errors they made during a synchronous computer-mediated chat session. Smith operationalized noticing as an “eye fixation on the target item of 500 ms or longer” (p. 87) because he claims that “fixations shorter than about 250 ms are of little interest” (p. 88). We believe that the proposed 500 ms cutoff point for establishing noticing is somewhat arbitrary and unsupported by the eye-tracking literature. Although a recent review does suggest that the average eye fixation of a native speaker (not a L2 learner) reading an English word is between 225 and 250 ms in length (Rayner, Reference Rayner2009), effects smaller than 250 ms abound. In a follow-up study, Smith (Reference Smith2012) replaced the 500 ms cutoff point by a heat map color—namely, the second coolest color in the spectrum, blue. As heat map colors are defined “relative to the length of the eye gaze recording” (Smith, Reference Smith2012, p. 63) rather than in relation to a control condition, this approach does not fundamentally address the criticism leveled against Smith’s (Reference Smith and Hult2010) operationalization of noticing. In particular, neither absolute fixation durations nor heat map colors take into account that the number and length of eye fixations are influenced by word length, word frequency, and predictability, among other factors. Therefore, Smith’s (Reference Smith and Hult2010, Reference Smith2012) approach has the undesirable effect that the noticing threshold will be reached more easily for some (e.g., long, infrequent, unpredictable) words than for other (e.g., short, frequent, predictable, etc.) items. In our study, we avoid this pitfall by working with different conditions, including a control condition, for which eye-fixation times can be compared. Noticing in our study, then, is operationalized in relative, rather than absolute, terms—that is, as an increase in fixation time relative to a baseline condition.

In the field of psychology, Williams and Morris (Reference Williams and Morris2004) pursued a question similar to ours. They examined the effects of different measures of word familiarity on eye-fixation times during reading. In their second experiment, native English-speaking college students read 48 target sentences in their first language (L1). These consisted of a neutral introduction, a target word, and a highly constraining posttarget context. There was one novel word condition in which readers had no prior knowledge of the target word (e.g., Sheila selected the MASDOR [capitals added] for her bouquet). After the reading session, participants had to select the correct meaning of the novel word from two possibilities on an unannounced vocabulary posttest. Williams and Morris (Reference Williams and Morris2004) established an association between online reading patterns and posttest performance. Readers tended to spend less initial processing time (i.e., shorter gaze duration) on those target words whose meaning they later identified correctly. However, reanalysis (as measured by second pass time) was associated positively with posttest accuracy. Although our own participants and materials differ from Williams and Morris’s, it is of interest to explore whether or not the processing times collected for our sample predict learning gains in a similar fashion.

Incidental Vocabulary Acquisition while Reading

The lexical targets for noticing in our study (i.e., pseudowords) were embedded into 9- to 13-line paragraphs, which participants were instructed to read at a leisurely pace.^{Footnote 3} Given that participants were initially unaware of the true aim of the experiment and of the ensuing vocabulary posttest, this experiment also fits into the body of research on incidental vocabulary acquisition during reading. Vocabulary acquisition is incidental—rather than intentional—when learners commit new words or phrases to memory “without having explicit instructions about a forthcoming memory test” (Mäntylä, Reference Mäntylä, Smelser and Baltes2001, p. 7262; see also Dörnyei, Reference Dörnyei2009; Hulstijn, Reference Hulstijn, Doughty and Long2003). In this case, lexical knowledge “accrues as a by-product of language usage” (Schmitt, Reference Schmitt2010, p. 29) such as “reading a novel simply for pleasure with no stated goal of learning new lexical items” (Schmitt, Reference Schmitt2010, p. 29). Although picking up new vocabulary as a by-product of reading can and does occur, this is typically a gradual, error-prone, and, hence, rather inefficient process (see Laufer, Reference Laufer, Foster-Cohen, García Mayo and Cenoz2005; Read, Reference Read2004). Therefore, if the aim is to expand learners’ vocabulary through reading, it seems desirable to enhance or complement reading activities by some form-focused component aimed at drawing the learners’ attention more explicitly toward the targeted lexical forms (e.g., Huckin & Coady, Reference Huckin and Coady1999; Hulstijn, Reference Hulstijn and Robinson2001; Laufer, Reference Laufer, Foster-Cohen, García Mayo and Cenoz2005; Nation, Reference Nation2001; Read, Reference Read2004; Schmitt, Reference Schmitt2008, Reference Schmitt2010).

One relatively nonintrusive way to accomplish this is by modifying the reading input so that it becomes more informative as to the meaning of the target word, thereby facilitating the lexical inferencing process. Watanabe (Reference Watanabe1997) tested the effect of three types of input modification on English as a foreign language (EFL) learners’ text comprehension and vocabulary learning while reading. The simplest form of input modification consisted of a synonym or paraphrase of a target word that was inserted between commas immediately after the lexical item (e.g., Each year in the U.S. about 7,000 infants die in their cribs, BABIES’ BEDS [capitals added], for no apparent reason). Watanabe conjectured that such appositive cues may make the context more helpful for vocabulary learning given the “(a) proximity of the illuminating context to the unknown word, (b) clarity of the connection between the context and the unknown word, (c) explicitness of the contextual information, and (d) completeness of the contextual information” (p. 288). However, contrary to expectations, learners in the appositive group did not exhibit significantly higher levels of vocabulary retention or reading comprehension than students who read the original, unmodified text. Watanabe attributed this zero effect to a lack of clarity of the semantic connection between the explanation and the word to be explained. Furthermore, he argued, even if participants do perceive the explanatory nature of the appositions, the question remains as to whether or not they will exploit their potential for vocabulary learning because, “if their purpose for reading is to understand a passage,” then “students may choose to ignore unfamiliar words and just read their explanations” (Watanabe, Reference Watanabe1997, p. 301, referring to Coady, 1993, p. 18).

In our study, we also enhanced the semantic informativeness of the context in which the novel words appeared by adding appositive cues in two out of the three critical conditions. Our study differs from Watanabe (Reference Watanabe1997), however, in two respects. First, appositions were signaled by the coordinating conjunction or rather than by commas. Second, the order of the defined expression (i.e., the novel word) and the definer (i.e., the known, existing word) was varied (see Quirk, Greenbaum, Leech, & Svartvik, Reference Quirk, Greenbaum, Leech and Svartvik1985, for a comprehensive account of appositions in English). Thus, our appositive conditions came in one of two forms, as illustrated by the following example: (Condition III) One demographical group stands out as being most likely to push PANIPLINES _DEFINEDOR BOUNDARIES _DEFINERand break rules, and (Condition IV) One demographical group stands out as being most likely to push BOUNDARIES _DEFINEROR PANIPLINES _DEFINEDand break rules (capitals added; see the Materials section for a description of Conditions I and II). Evidence from paired-associate learning and translation studies suggests that learners find it easier to link unknown words with known words than the other way around (e.g., Kroll & Stewart, Reference Kroll and Stewart1994; Mondria & Wiersma, Reference Mondria, Wiersma, Bogaards and Laufer2004; Webb, Reference Webb2008). It is an open question as to whether or not this pattern still holds when words are not presented in isolation but in a sentence context.

Participants

Twenty-eight female Dutch-speaking college students volunteered to participate in the study. They were all second- or third-year English language majors (age range 19–28) who had started to learn EFL from age 13 onward in secondary school in Belgium. Their level of proficiency in English was at the B2 (upper intermediate) or C1 (lower advanced) levels of the Council of Europe’s (2001) Common European Framework of Reference. All participants had either normal or corrected-to-normal eyesight.

Apparatus

Eye movements were recorded with the EyeLink II, a head-mounted eye-tracker manufactured by SR Research. The texts were presented in Courier New, 18-pt font, on a 19-in Dell computer monitor set up at 68 cm from the participants’ eyes. At this distance, 4.0 character spaces equaled 1° of visual angle. The use of corneal reflections in combination with pupil tracking (i.e., pupil + CR mode) allowed for more stable tracking of the participants’ point of gaze. Participants took the unannounced vocabulary posttest on the same computer after they had taken off the headband. Each participant was debriefed and informed about the true purpose of the experiment at the end of her session.

Materials

Participants read 20 short paragraphs in English (length: 83–122 words) while an eye-tracker recorded their eye movements. The reading passages consisted of authentic newspaper and magazine extracts that were adapted to suit the participants’ estimated proficiency level. Out of the 20 paragraphs that participants read, 8 paragraphs were fillers, and 12 were critical (i.e., they contained a new, unknown word in all but the control condition). See Appendix A for an example of two full paragraphs. On the computer monitor, text lines were displayed with two hard returns in between to assure accurate recording of participants’ gaze position. As a result, each paragraph took up three screens, and participants moved from one screen to the next by pressing a button; participants could not, however, return to an earlier part of the same text. The lexical target area always appeared roughly in the middle of the second screen (i.e., never as the first or final word of a line or at the top or bottom of the screen).

Great care was taken to ensure that the texts contained no unfamiliar items except for the targeted unknown words. The preparation of the materials consisted of two steps. First, we ran the paragraphs through Range (Heatley & Nation, Reference Heatley and Nation1994), a software program that categorizes the words of a given text into frequency bands. All words that did not belong to the 3,000 most frequent word families were considered for substitution by a more common paraphrase. Next, we pretested the adapted reading materials with first-year English-language majors at the same university.^{Footnote 5}

The first-year students were asked to translate a number of predetermined, potentially difficult words in the texts into Dutch and to mark (i.e., by underlining or circling) any of the remaining words of whose meaning they were not sure (see Godfroid, Reference Godfroid2010, for the complete pretest materials). Analysis of participants’ translation scores revealed that, apart from one word (i.e., synaesthesia), students’ receptive knowledge of all words was fairly good. In other words, although the students did not perform flawlessly, the error rates for most words were similar; only synaesthesia stood out, with only half of the students being familiar with it. Therefore, no changes were made other than to replace the synaesthesia paragraph with another text.

The various modifications that resulted from the frequency analysis and the pretest allow us to assume that all words in the paragraphs, including the existing words serving as appositive cues (e.g., boundaries), were known by the participants of the actual experiment, who had at least one more year of intensive formal English instruction than the participants in the pretest. Given that critical paragraphs were about 100 words long and contained one novel word, lexical coverage is estimated at around 99%; this constitutes a known-word density favorable to successful guessing from context (Nation, Reference Nation2001). In view of this high vocabulary coverage, it is also unlikely that participants’ failure to pay increased attention to a novel word in a paragraph resulted from their being distracted by some other unfamiliar lexical item in the same text.

Latin Square Design

Table 1 shows the four experimental conditions for a single critical paragraph. Each participant read all 12 experimental paragraphs but read each paragraph in only one of four versions. These versions are identical except for the realization of the target area. The paragraphs were presented to the participants according to a 4 × 4 Latin square design (Hinkelmann & Kempthorne, Reference Hinkelmann and Kempthorne2008; Kutner, Nachtsheim, Neter, & Li, Reference Kutner, Nachtsheim, Neter and Li2005), so that, in the end, each participant read three paragraphs per condition, in a counterbalanced fashion.

Table 1.

The four experimental conditions, illustrated with (an excerpt of) one paragraph

Note.

The target areas have been capitalized in Table 1 for ease of reference but appeared in regular print in the actual experiment.

Condition I is the control condition. The paragraph contains a known, existing target word (e.g., boundaries). In contrast, the pseudowords (e.g., paniplines) in the critical conditions II, III, and IV are hypothesized as targets for noticing and learning. The critical conditions differ from one another in the degree of contextual support that they offer to the novel word. In Conditions III (e.g., paniplines or boundaries) and IV (e.g., boundaries or paniplines), a known, appositive cue provides a direct, explicit hint as to the meaning of the novel word. Although appositive constructions normally imply a relationship of coreferentiality between the two nouns or noun phrases (e.g., Crystal, Reference Crystal1997; Quirk et al., Reference Quirk, Greenbaum, Leech and Svartvik1985), the interpretation of an appositive construction ultimately falls to the reader. Meyer (Reference Meyer1987) notes that units in apposition can be “co-referential, hyponymous, synonymous, or attributively related” (p. 103). In all of these cases, the juxtaposed known word is a strong clue that can facilitate the semantic integration of the novel word into the sentence.

To comply with our Latin square design, we ran a 4 (conditions) × 4 (paragraph sets) × 4 (groups) ANOVA with seven different participants nested within each of the groups. We included only main effects (i.e., no interaction terms). Condition and paragraph set constituted fixed and random within-subjects variables, respectively, and group was a fixed between-subjects factor. The significance level α was set at .05 for all analyses. In the next section, we present the four fixation time measures that served as the dependent variables.

Processing Measures

Eye-movement analyses were performed on participants’ first fixation duration, gaze duration, second pass time, and total time in the interest area. First fixation duration is the duration of the first fixation in an interest area when that area is encountered during forward reading (i.e., in English, left to right) for the first time. Given that first fixation duration is the earliest measure of processing time, it has been argued to reflect very fast cognitive processes (Rayner & Pollatsek, 1987, cited in Rayner, Reference Rayner1998). First fixation duration along with any other fixations made on a word during the initial visit make up gaze duration. Both first fixation duration and gaze duration are indices of initial processing, or first pass reading. If the eyes reenter an interest area after they had left, the sum of the fixations made during this second pass equals second pass time. Second pass time captures rereading episodes, which are indicative of the reader’s reanalysis of the interest area. It is conceivable that the addition of an appositive cue following the novel word (as in paniplines or boundaries) will induce higher second pass reading times for the novel word (Research Question 3). That is because the unexpected increase in contextual richness associated with encountering a useful cue after one has already processed the novel word may lead participants to revisit the novel word for (re)interpretation. In contrast, when the appositive cue precedes the novel word (as in boundaries or paniplines), there is no shift in informativeness at the novel word or just after. Accordingly, we expect second pass time for the pseudoword to be lower in Condition IV than Condition III.

Finally, the sum of all fixations over all passes is expressed in the measure total time. This is the primary measure that we use to assess the relationship between eye-fixation times (and, hence, attention) and vocabulary learning (Research Question 2).

Vocabulary Posttest

After the reading task, participants received an unannounced vocabulary posttest that assessed their ability to recognize the target items. The format of the vocabulary posttest is the result of careful pilot testing (see Godfroid et al., Reference Godfroid, Housen, Boers, Pütz and Sicola2010). It is essentially a gap-fill exercise under time pressure. For each target item, we followed a two-staged testing procedure. First, the participant read the exact same sentence she had encountered during the preceding reading task but in which the pseudoword (Conditions II–IV) or the corresponding existing word (Condition I) was replaced by a dotted line.^{Footnote 7} When ready, she pressed a button.^{Footnote 8} An array of 18 possible candidates for the missing word then appeared above the sentence. The participant had 15 s to scan the array and enter the number of the missing word. After that, a blank screen allowed the test taker to rest for half a minute, until the next sentence was presented. Sentences were generated in a random order, and, for each test item, the array was arranged completely anew. The whole procedure was repeated until all 12 items were completed.

The 15-s time limit only allowed participants to scan the various options. We reasoned that if participants had noticed the target word during the reading task, they would be able to quickly recognize it here too. If not, then 15 s would not suffice for them to reason the answer out, for example, by eliminating possibilities on the basis of grammatical number (i.e., singular vs. plural) or previously given answers. Thus, we would claim that the vocabulary posttest measures participants’ recognition of a target word’s written form within its original, meaning-bearing context (i.e., a form of receptive vocabulary knowledge). More than half of the distractors on the test were pseudowords that the participants had also encountered while reading. Mere recognition of an orthographic form as something that looked vaguely familiar did not, therefore, suffice to select the correct word. Rather, the novel word had to be matched with the right, meaning-bearing context in which it had been encountered.^{Footnote 9}