What is processing?

A definition of processing is central to any work about input processing, regardless of the level of the learner under study. As in other publications (e.g., VanPatten Reference VanPatten1996, Reference VanPatten and VanPatten2004, Reference VanPatten, VanPatten and Williams2007, Reference VanPatten, Piske and Young-Scholten2009, Reference VanPatten, Gass and Mackey2012; VanPatten & Rothman, forthcoming), I take processing to be one of two things. The first is the connection between form and meaning made during real-time comprehension. This aspect of processing has to do with linking the sound sequence /kæt/, for example, to the meaning of cat (however that is represented semantically in the linguistic system – see Carroll, this volume). It also has to do with linking meanings to morphological structure such as dishonest means the opposite of honest, and talked refers to a past time frame as opposed to talk/talks.¹

A second aspect of processing also has to do with meaning but is related to syntactic structure: parsing. Processing as parsing is the real time computation of sentence structure. During parsing, relationships among words are computed and syntactic structure is projected and built during the course of listening to (reading) a sentence. This process is related to meaning as parsing involves basic sentence meaning (e.g., who did what to whom?) as well as the resolution of ambiguity (e.g., which antecedent is linked to this pronoun/verb?). In short, parsing is related to sentence interpretation.

What is critical to processing, be it morpho-lexical or syntactic processing, is a focus on meaning. That is, the definition offered here links processing to how people (or in the present case, language learners) link morphophonological units in the input to meaning.² Storage and the development of grammar are separate processes, but what is clear from work in linguistics and psycholinguistics is that even though there are internal constraints on grammar (i.e., those imposed by Universal Grammar), language acquisition happens in the context of communication. Thus, processing is linked to what learners do during acts of communication; that is, how they process for meaning and what guides or constrains this processing.^{3, 4}

Before continuing, it is fitting to address the issue of explicitness/implicitness in processing. I call the reader’s attention to the fact that the above definition of processing is agnostic with regard to whether learners process language explicitly or implicitly. This agnosticism has much to do with two somewhat related issues. The first is that the explicit/implicit processing distinction invites a slippery slope, with tremendous difficulty in articulating the actual distinction (e.g., Hulstijn Reference Hulstijn2005; Rebuschat, forthcoming; Truscott & Sharwood Smith Reference Truscott and Sharwood Smith2011). What would make processing explicit? Knowing that you’re processing something? Or knowing what you’re processing? Or knowing what you’re processing, how you’re processing it, and what connections you are making? It is not my place to resolve this issue here, but only to point out why a definition of processing may have difficulty with explicitness/implicitness as a construct. A second related issue is that in the end, a good deal of language acquisition is implicit and what is stored in the linguistic system is implicit in nature; this is held by most researchers regardless of theoretical orientation (e.g., generativists, emergentists; see White, forthcoming, and Ellis Reference Ellis2005, as two examples). This raises the question “To what extent can processing be explicit if the product is implicit?” To put this another way, “What is the relationship between the ‘what’ of processing and the ‘what’ of the internalized linguistic system?” (See VanPatten & Rothman, forthcoming, for some discussion.) The matter of explicitness will surface in a moment as we review some of the ideas presented in the chapters in this volume.

With the above defining characteristics of processing in mind, the first and most important question to ask is how the researchers in the present volume define processing. Equally important is how processing is measured. Although I will discuss research methods in the next section, it is difficult to completely separate measurement/assessment from the definition of processing. Thus, the reader will find some degree of redundancy between this section and the next. For the first chapters in this volume (Han & Sun, Park), it is not at all clear that the authors define or conceptualize processing the way it is here. In fact, one could question whether these two chapters are focused on processing or on something else (e.g., explicit strategies for tackling text). What emerges explicitly or implicitly (no pun intended) in both studies is that the real focus is on noticing, not processing. What is noticing? Noticing is a construct that seems to have some elasticity, or at least has evolved since its original conception (e.g., Schmidt Reference Schmidt1990, Reference Schmidt and Robinson2001). What I think is fair to say is that noticing is not synonymous with processing. Noticing entails some level of awareness (which can vary from definition to definition of the term), while processing does not. What is more, noticing requires only that a particular datum be, well, noticed; there is no part of the definition that says that a morphophonological unit is linked to meaning (and ultimately, function).⁵ However, a definition of processing requires that linguistic data are linked to meaning during real-time processing. In short, one can notice something in the input, but not process it (see VanPatten Reference VanPatten and VanPatten2004 for a full discussion). Finally, noticing has never been meant to apply beyond the word level. That is, to my knowledge, there is no noticing that is related to syntactic computation of sentences.⁶

In arguing that noticing and processing are not the same thing, am I arguing that they are completely dissociated? As one reviewer queried, isn’t noticing a part of processing? The answer is that as currently defined, noticing need not be a part of processing. That is, processing does not necessarily entail that learners are aware of something while processing; indeed, awareness may be epiphenomenal to processing, as discussed in Truscott and Sharwood Smith (Reference Truscott and Sharwood Smith2011). What I mean is that a definition of processing does not require awareness of what is being processed. This hypothesis is, of course, researchable, although it is not clear to me how one would examine such a hypothesis at this point in time. Even if we accepted that noticing is part of processing, to limit oneself to noticing as a construct is to limit what one can say about processing more fully – that is, about connecting form to meaning. It is to this issue I now turn attention.

In raising the issue of noticing versus processing, then, I question the extent to which we can discern from Han and Sun, as well as Park, the degree to which learners have made connections between surface properties of language and their meanings. For example, Han and Sun claim to find evidence that does not support the hypothesis that learners begin input processing by searching for words (morphophonological units that are lexical in nature); that is, they argue against the model of input processing articulated in VanPatten (Reference VanPatten and VanPatten2004, Reference VanPatten, VanPatten and Williams2007, and elsewhere). Their claim is based on the self-reports of their participants – self-reports that seemingly run contrary to the hypothesis. But even if learner reports do contain suggestions they are not concerned primarily with words, the task they were engaged in (self-reporting) delimits what they can actually say about linking form with meaning. (I will address this in more detail in the next section.) What is more, the one learner that Han and Sun cite to support their claim says this: “I couldn’t find any clue for the meaning, and the letters are totally different from my language and have few similarity to English” (emphasis added). What I find telling from this one introspection is how clear the learner was about her task: ‘find the meaning.’ Linking meaning to surface elements is what processing is about, and in spite of Han and Sun’s concern with noticing, processing seems to pop up in learner comments. Close examination of their protocols in Tables 1.3 and 1.4, for example, suggests a drive to get words from the input, not a concern with formal features of language. This is particularly true when one eliminates vague reports such as “information about English” and “English, French, Creole” – which tell us little about what they were doing – and focuses on more specific statements such as “The words that match with the words that I know,” “Compared the words with English words,” and “Guess and memorize some words similar to English.” What one sees after culling the reports is learners’ push to find lexical items to bootstrap themselves into the content or meaning of the passage. This is precisely what I would predict,⁷ and I find it odd that Han and Sun see something else in their data.

Rast et al. focus on how learners accrue, over a brief period of time, knowledge about nominal morphology in Polish. To this end, the study is less about processing and more about product, especially given the assessment tasks they used (see next section). In addition, they also confound noticing with processing as revealed by their statement regarding an inclusion of a test of individual differences: “individual differences which may play a role in the noticing [...] and intake of new language forms” (emphasis added). Such tests of individual differences often bias toward explicit learning, and as noted above, there is some (not too clear) relationship between explicitness and noticing. Nonetheless, some interesting findings emerged in their study from which one can make some inferences about processing. For the present discussion, I will limit my comments to the results of the grammaticality judgment task (GJT).

Rast et al. found that learners showed evidence of knowledge of certain case endings on the GJT; that is, with very limited exposure, they had picked up particular morphological features in the input and had begun incorporating them into their grammars. There did appear to be frequency and transparency effects; the more frequent and the more transparent the word, the more likely the learners were to be correct with morphological endings. I interpret this finding as partial support for the fact that learners are actively seeking out lexical items as opposed to grammatical features in the initial stages of processing for acquisition. I also find this to support the idea that morphological features may be initially processed as part of the lexical item and not as separate entities (for some discussion, see VanPatten, Keating & Leeser Reference VanPatten, Keating and Leeser2012 as well as VanPatten & Rothman, Reference VanPatten, Rothman, Laval, Arche and Benati2014). In short, if one works backwards and attempts to infer process from product, in Rast et al.’s results, lexical acquisition is driving or at least anchoring the processing of morphological inflections. Although I have argued against rule learning in other publications (VanPatten Reference VanPatten, Sanz and Leow2011, VanPatten & Rothman, Reference VanPatten, Rothman, Laval, Arche and Benati2014), these findings make sense if one believes in productive morphological rules: in order to attach an inflection to something, you have to have that something to attach it to. And that something is a word (lexeme, to be more technical). The effects of frequency and transparency, however, can also be used to suggest that learners don’t yet have productive rules; instead, the morphological endings to which they show some sensitivity in the GJT are directly connected to words in their beginning lexicons. Thus, the frequency and transparency affecting lexical processing during acquisition in turn affect morphological processing. In the next section, I will address the issue of measurement and assessment, as I understand that in coming to the present conclusion, it appears that I am accepting GJTs as appropriate measures of underlying knowledge.

There is little to address in Carroll’s study on word learning, as implicit in her design is the idea that processing connects meaning with form (i.e., her description of what it means to learn and know a word). Because her study focuses directly on word learning, which as mentioned previously is an important if not critical first step in language acquisition as processing, it does not directly test any hypotheses related to input processing. However, she does report the interesting finding that the role of frequency in the input is attenuated by the nature of the word being learned; in short, not all words are created equal, and some may require much less exposure to be processed and learned compared to others. As she correctly points out, this is problematic for hard-core emergentists/connectionists for whom frequency often tends to be absolute. From an input processing perspective, she also provides evidence not supportive of the Sentence Location Principle reported in Barcroft and VanPatten (Reference Barcroft, VanPatten, Glass and Pérez-Leroux1996), and VanPatten (Reference VanPatten1996, Reference VanPatten and VanPatten2004); that is, Carroll finds sentence-initial position is not privileged compared to sentence-final position and sentence-medial position. Although her results do suggest that sentence location is not a factor for the learning of names, by her own account we must restrict this finding to the study at hand as not all words are created equal. What may be true for names may not be true for other kinds of words (e.g., names vs. non-names, cognates vs. non-cognates, concrete vs. abstract, lexical vs. functional). What is more, the Sentence Location Principle was meant to account for the processing of surface grammatical features, so we should be cautious in extending her findings on the learning of names to the realm of processing morphological and other forms related to functional features of a language.

Sagarra’s study in this volume does not quite address ab initio learners as do the other studies, but what it shares with others is its vague definition of what processing is. At times, Sagarra refers to “sensitivity/insensitivity” to morphological form, without such sensitivity being linked to processing as the connection between form and meaning/function. Explicit in her study is that L1 biases for language processing should effect initial processing (as I define it) in the L2. In her study, such processing would be the linking of verbal morphology with the semantic notions of person and number, with that idea that such processing is L1 based. As such, her study would presumably test whether something like the Lexical Preference Principle (e.g., learners begin processing an L2 using lexical as opposed to grammatical cues when both encode the same meaning; see VanPatten Reference VanPatten, VanPatten and Williams2007) is universal or not, and would be a welcome experiment. Assuming Sagarra agrees with the definition of processing (for acquisition) that I outline, it is difficult to determine whether her study can speak to initial processing as her study focuses on learners who had already studied some Spanish. Her study is more in line with that reported in VanPatten et al. (Reference VanPatten, Keating and Leeser2012), of which it is basically a conceptual replication with the exception of the methodology used; we used self-paced reading, while Sagarra used eye tracking. In that study we were interested in whether learners who already possess some kind of grammar are sensitive to morphological violations, which is a question quite different from how learners get morphology in the first place (i.e., how processing happens in the initial stages to begin forming grammars). Thus, Sagarra’s study is more in line with the connection of underlying knowledge with real-time implementation during comprehension as opposed to initial input processing. I will address the use of eye tracking in the next section.

Moreno (this volume) attempts to address the issue of simultaneous attention to meaning and form by reviewing a number of studies. One is the original VanPatten (Reference VanPatten1990) study, subsequently replicated by Wong (Reference Wong2001) and Greenslade et al. (Reference Greenslade, Bouden and Sanz1999), as well as a related study by Leow, Hsieh, and Moreno (Reference Leow, Hsieh and Moreno2008) and another study by Han and Peverly (Reference Han and Peverly2007). Because the first three studies tend to dovetail in their evidence that learners’ processing of lexical items takes precedence over their processing of grammatical form when focus is on meaning, I will focus on the other studies.

In Leow et al., the methodological innovations introduced were enough to render it a different kind of study from the first three. There are three major points deserving of comment: (1) the use of written as opposed to aural texts; (2) the use of concurrent think-alouds; and (3) the assessment task. It is not clear to me that reading a text and marking it up is the same as listening to a text and pressing a button when you hear something (as pointed out by Leow et al. in their original article, and tested by Wong in her study). More importantly, what is missing from their study is another control group that did not think aloud. That is, because they introduced think-alouds across the board, the fact that they did not get results similar to those in the previous studies can be due to this one design feature. Only a pure control group that paid attention only to meaning and did not think aloud would let us know this. Finally, Leow et al. used a multiple-choice test as opposed to a free recall, because they questioned the validity of free recalls as a measure of comprehension as opposed to memory. First, free recalls have been used extensively in both first and second language reading and listening research and have been tested against other measures, with the findings of this research supporting free recalls as a valid measure of comprehension. In fact, they are sometimes considered more valid because they are free of researchers’ manipulation of ideas in other kinds of tests, such as the one used by Leow et al. (see, for example, the discussion in Heinz Reference Heinz2004). Multiple-choice tests of comprehension are not always the best way to test comprehension in an experimental study because it is difficult for researchers to construct such tests so that they don’t actually inflate what the learner actually understood (i.e., the learner can reason out from the test itself what the answer may be), and such tests can easily fail to capture the process of comprehension itself (e.g., Bernhardt Reference Bernhardt1991). Although Moreno does not report on the actual comprehension scores, the mean scores across the conditions in the original Leow et al. study was 4.8–10 (SD 1.61), with a range of scores among the different conditions from 4.36 to 5.20. This compacted range of scores that hovers around 50 percent suggests to me that the test itself may have induced responses such that all groups performed similarly. This is, of course, speculation on my part, but it is reasoned speculation and one that Moreno does not consider. In short, I think it is premature to argue against an internal processing hierarchy for novices that privileges content lexical items to get meaning from a passage in the earliest stages of acquisition.

The other study that Moreno reviews is the precursor to the Han and Sun study in this volume, namely Han and Peverly (Reference Han and Peverly2007). Given that the two studies are nearly identical, my comments above apply to the original Han and Peverly study, namely, it is a study about noticing and not a study about processing. Thus, it has little to say about the simultaneous attention to form and meaning, even though Han and Peverly ask the learners to perform a free recall. (I will have more to say about tasks in the next section.)

One point on which I wholeheartedly agree with Moreno is that future research must revisit methodology and assessments. In the next section, I will comment on this matter.

To conclude this section then, my main points are the following:

1. Any study on the processing of language by beginners/novices must include a definition of processing in which both form and meaning are linked. Noticing is not the same as processing.

2. We need to have clear distinctions between strategies for tackling a task and actual processing of sentential elements so that we don’t confuse the two.

3. We also need to distinguish between process and product and be clear that if we are dealing with product, we can only “work backwards” and infer what happens during processing.

4. Based on the studies presented in the present volume, it is premature to reject the major principles of input processing as outlined in VanPatten (Reference VanPatten and VanPatten2004, and elsewhere).

What is the nature of tasks and research design in this research?

Critically, any study – not just those regarding input processing – must make clear how experimental tasks relate to actual processing. Currently, most processing research uses some kind of real-time measure, such as self-paced reading, self-paced listening, eye tracking, and so on (see, e.g., Jegerski & VanPatten Reference Jegerski and VanPatten2013). Other offline measures are possible, as long as they are able to measure the extent to which learners link meaning and form. To this end, Carroll’s study provides an example of an offline study in which both accuracy and latency measures were used and are appropriate to see if learners have mapped particular lexemes onto their meanings/referents. Still, we need to be aware that these are product measures. She also measured, however, trials-to-criterion (e.g., Henry, Culman & VanPatten Reference Henry, Culman and VanPatten2009), or how many items it took before learners began to process sentences correctly.

Of interest here is whether GJTs and self-reports/introspections are measures of anything related to processing. The answer is “It depends on the research questions and on one’s definition of processing.” Because these studies are less about processing and more about noticing, the measures used can be considered appropriate tasks. However, as suggested in the previous section, such tasks fall short as tests of any hypotheses about actual processing. That is, if we take processing to be how learners connect surface sentential properties with meaning, then where in GJTs and self-reports do we see a measurement of these connections? The answer is we don’t. What is more, GJTs have been found to cause learners to interact with sentences differently than if they were reading purely for meaning (e.g., Leeser, Brandl & Weissglass Reference Leeser, Brandl, Weissglass, Trofimovich and McDonough2011), turning the reading of sentences into explicit searches for form. As for self-reports, as mentioned in the previous section, they tend to yield information about explicit strategies for tackling a task but not about underlying principles that guide processing. In the present case, as I argued, the strategies actually coincide with the Primacy of Meaning Principle’s major corollary that early-stage learners’ processing mechanisms are on the hunt for words and word boundaries in the initial stages (the processing of formal elements coming in later). At first blush this may seem to be in contradiction to my claim that self-reports are dubious. But what happens when we get out of this particular aspect of processing and get into something like the processing of temporal reference? I draw the reader’s attention to the fact that not a single participant in the Han and Sun or Park study says anything about trying to figure out when something occurred or is occurring. This may be due to the nature of the text depending on how the participants were oriented to it prior to performing the task (i.e., they may expect the entire thing to be in the present tense because they know it’s a personal ad). But what about the First-Noun Principle? Again, I note that not a single participant reports being on the lookout for who’s doing what to whom. This is because something like the First-Noun Principle tends to fly under the radar of awareness and explicitness; syntactic processing tends to happen (if not exclusively) outside of awareness. In short, self-reports such as those used in some of the chapters in the present volume yield results that are restricted in terms of what they can research, or, if we wanted to be as critical as possible, irrelevant to actual processing.

Something similar can be said for Park’s use of having participants read a text and mark “anything they notice” in it. This may be a fine measure for noticing, but it does little to tell us about processing. A person may consistently mark something in a text but have no idea what it means or what its function is. Marking up a text, then, is not necessarily indicative of processing (i.e., how learners connect form and meaning). (For additional comments on methods and design issues in this kind of research, see Leeser Reference Leeser, Jegerski and VanPatten2013.)

The use of eye tracking in Sagarra’s study is interesting and is a long-standing method in L1 and L2 processing and parsing more generally. However, as pointed out earlier, her study is not about initial processing and how learners come to link meaning and form, so it is not clear how the study speaks to the issues of the present volume. The question here, then, is whether something like eye tracking (or other online measures) can be used to look at initial processing. Sagarra’s study (and a very similar one that I conducted with Gregory Keating in 2007 in which we looked at the tension between lexical cues and grammatical cues to tense) might be useful if we worked backwards from the findings. Sagarra essentially finds a reliance on lexical cues as opposed to morphological cues to get meaning (cf. the Lexical Preference Principle), apparently modulated by, but not completely attenuated by, effects of the L1. If we project back to what the learners must have been doing in the very first moments of learning/processing Spanish, before they had built up any lexicon, my hypothesis is that we would see even more significant gaze and duration times on lexical cues as opposed to verbal morphological cues as learners struggle to figure out what words mean. For the purposes of the present volume’s concerns, this is the experiment that needs to be conducted: an online study in which learners are asked to process for meaning (not to notice) when they essentially have no grammar (including lexicon) for the L2.

Stimuli, of course, are important in any research design. In the present volume, we have several studies that used written paragraph-level text and others that used aural sentences during training as well as sentence-length stimuli in assessment. The first question that occurs to me is why we would want to test absolute novice learners with paragraph-level stimuli. Aside from the occasional learner who might troll the Internet looking for something in a second language to read, why would we ask people who know no language to read a text? Let’s imagine a task not used in the present volume anywhere, a task in which learners read each sentence of the text and then report on what they think that sentence means. This would certainly tell us in a sentence-by-sentence format how they are processing language. The problem, of course, is that with absolute beginners, most of the time one would get nothing from the participants as the level of language is beyond what they can do, save the occasional cognate word recognition if the L1 and L2 orthographic systems are the same. But as we know, the tasks used with paragraph-level texts in the present volume did not focus on meaning but instead on something else. That is, at no point were learners asked to demonstrate that they understood anything (the lone exception being Carroll’s study).⁸ Instead, they were asked to think aloud, self-report, judge a sentence, or mark up the text to show what “they noticed.” Some researchers did use free written recalls afterward, which on the surface appears to have learners focus on meaning, but two aspects of how recalls are implemented are important. The first is that the question that the participants were asked in the Han and Sun study does not seem to really focus on sentential meaning within the passage, as the question was “What was this passage about?” How does one approach this question? If I answer “It was about the news,” is that enough? Second, did participants know they were being held responsible for meaning? Clearly they were asked to “notice” things, but were they told something like this: “After you have read this passage, you will be asked to recall as much information about the content of the passage as you can.” Such task orientations surely change the focus of the participant. I note here that the mean recall scores in the Han and Sun study were between 0.48 and 1 out of eighteen idea units. Depending on how learners were oriented to the passage, two possible conclusions emerge: (1) they weren’t focused on meaning (they didn’t really understand they would be tested on content), or (2) their “noticing” did indeed interfere with a focus on meaning, either scenario suppressing the recall scores in the end, and neither scenario offering evidence against the basic principles that guide input processing.

To drive the point home, the issue of meaning in the tasks we use in processing research is crucial. Reminding the reader, I underscore here that processing involves a relationship between form (morpho-lexical, sentential) and meaning. A processing task without a focus on meaning cannot be a processing task. This has always been a cornerstone of both input processing research and its derivative, processing instruction. In addition, for this reason, contemporary approaches to processing and parsing include a focus on meaning. This is why researchers include comprehension questions in self-paced reading/listening, eye tracking, and other types of methods. Comprehension questions after stimuli sentences keep the participant focused on meaning.

I have been rather critical of the research designs in several of the studies in this volume, but in all fairness to the researchers, tapping learner processing at the very earliest stages is not the easiest thing to do. We run the risk of using tasks that, seemingly appropriate for the level of the learner, may induce behaviors that wouldn’t happen if the learners weren’t in an experimental setting. In short, we must be aware of the following potential issue: the nature of the stimuli driving the task rather than a true processing task driving the nature of the stimuli. My suggestions for future research in this area include the following:

1. For novices, stimuli should be sentential and not textual.

2. Stimuli should be accompanied by context (visual or others to aid in interpretation of the stimuli).

3. Comprehension of sentences should be tested.

4. Depending on what the research is interested in, there could be simultaneous focus on meaning and form or sole focus on meaning. A sole focus on form makes little sense from a processing perspective.

5. Researchers should focus as much attention on aural stimuli as possible. Or, researchers should include comparisons of aural and written stimuli so that we can better separate out the effects of modality.

6. Researchers should consider more implicit measures in their research design, such as reading/listening times to avoid a bias for explicit knowledge and explicit behaviors, as well as the possibility that more explicit measures would delimit the data about actual processing for the researcher (again, see Leeser Reference Leeser, Jegerski and VanPatten2013).

Concluding remarks

As mentioned at the outset of these comments, work on earliest stages of acquisition ought to be more central to L2 research. Although it has proved fruitful to examine outcome and to compare L2 systems with L1 systems, only by examining acquisition at various stages can we piece together developmental processes. Important questions underlying this research include the following: To what extent are the “things” that guide early stage acquisition the same as those that guide later stage acquisition? Do underlying processing strategies and processes for acquisition change over time? What makes early-stage learning “early” and advanced-stage learning “advanced” other than the content of the linguistic system? These questions suggest, then, the importance of looking at beginning language learners. As my comments in this chapter indicate, we need to proceed with some caution and with some clarification. If our focus is on the nature of input processing, we need to come to agreement about what the term processing means, and we need to engage in discussion about appropriate research designs and tasks to use with novice language processors. The present volume, then, sets the stage for such discussion, and I think the editors are helping us move in that direction.