Efficient comprehension of sentences requires rapidly and continuously accessing and integrating different sources of information in real time. Psycholinguists have developed detailed models and theories to account for the processes involved in on-line sentence comprehension as well as a number of sophisticated experimental designs for studying these processes. But how about real-time sentence processing in bilinguals? The study of bilingual sentence processing has received considerable attention and has led to a remarkable growth of experimental studies over the last 10 years. The focus of these studies has been on late bilinguals, i.e., on second-language (L2) learners who learned a non-native language after early childhood, as adolescents or adults. These studies have revealed both similarities and differences between native (L1) and non-native (L2) sentence processing. Several proposals have been made to account for the experimental findings, but the significance and nature of native vs. non-native differences in sentence processing has remained controversial. Some researchers have claimed that L1 and L2 sentence processing are essentially the same and that observed performance differences between native and non-native sentence comprehension are due to peripheral factors, e.g., decoding problems, working memory limitations, slower processing speed, difficulties with lexical access and retrieval, or a reduced ability to predict during L2 processing (e.g., McDonald, 2006; Hopp, 2016; Kaan, 2014). Others have posited more substantial differences between L1 and L2 processing. One prominent proposal is Clahsen and Felser's (2006a, b) Shallow-Structure Hypothesis (SSH). Assuming multi-stream models of language processing (e.g., Ferreira & Patson, 2007) with two routes from form to meaning, a heuristic one that employs surface-form information, lexical and semantic cues, and an algorithmic route that relies on a full grammatical parse, the SSH holds that L2 processing relies less on grammatical and more on non-grammatical information sources, in comparison to L1 processing of syntactic (and morphological) phenomena.

A growing body of research has investigated bilingual sentence processing. How to account for differences in native (L1) and non-native (L2) processing is controversial. Some explain L1/L2 differences in terms of different parsing mechanisms, and the hypothesis that L2 learners adopt ‘shallow’ parsing has received considerable attention. Others assume L1/L2 processing is similar, and explain L1/L2 differences in terms of capacity-based limitations being exceeded during L2 processing. More generally, the role that working memory plays in language acquisition and processing has garnered increasing interest. Based on research investigating L2 sentence processing, I claim that a primary source of L1/L2 differences lies in the ability to retrieve information constructed during sentence processing from memory. In contrast to describing L1/L2 differences in terms of shallow parsing or capacity limitations, I argue that L2 speakers are more susceptible to retrieval interference when successful comprehension requires access to information from memory.

In his article, Parsing and working memory in bilingual sentence processing, Cunnings invites us to consider the hypothesis that important differences in how L1 and L2 populations process sentences stem from differences in how these speakers store and retrieve linguistic encodings in memory during the course of sentence comprehension (Cunnings, 2016). Specifically, he proposes that L2 speakers are more susceptible to similarity-based retrieval interference than their L1 counterparts, and L2 speakers weight discourse cues more heavily than syntactic cues when resolving open linguistic dependencies via memory retrieval. I find these to be interesting hypotheses that merit further investigation, especially in light of the prominence that these issues currently enjoy in L1 processing research. Nonetheless, I sound a note of caution: these claims go quite a bit beyond what is currently known, either theoretically or empirically. This makes support for these claims weak at present, but happily, this state of affairs offers clear directions for future research.

Memory is an integral part of language processing. Given this, a better understanding of how people learn, represent and process language requires considerations of the principles of memory that support language comprehension. Cunnings’ paper (Cunnings, 2016) does just this. The core of his proposal is that second language (L2) processing that is non-target like can be explained in terms of memory operations rather than by invoking a shallow processor (cf. Clahsen & Felser, 2006).

The thesis in this paper is that L2 speakers differ from L1 speakers in their ability to do memory storage and retrieval about linguistic structure. We would like to suggest it is possible to go farther than this thesis and develop a computational-level theory which explains why this mechanistic difference between L2 and L1 speakers exists. For this purpose, we believe a noisy channel model (Shannon, 1948; Levy, 2008; Levy, Bicknell, Slattery & Rayner, 2009; Gibson, Bergen & Piantadosi, 2013) could be a good start. Under the reasonable assumption that L2 speakers have a less precise probabilistic representation of the syntax of their L2 language than L1 speakers do, the noisy channel model straightforwardly predicts that L2 comprehenders will depend more on world knowledge and discourse factors when interpreting and recalling utterances (cf. Gibson, Sandberg, Fedorenko, Bergen & Kiran, 2015, for this assumption applied to language processing for persons with aphasia). Also, under the assumption that L2 speakers assume a higher error rate than L1 speakers do, the noisy channel model predicts that they will be more affected by alternative parses which are not directly compatible with the form of an utterance.

Cunnings (2016) proposes that differences between native (L1) and second language (L2) sentence processing can best be explained in terms of susceptibility to effects of interference and an overreliance on discourse level cues during memory retrieval. Cunnings’ argument that difficulty in retrieval operations may provide a better explanation than a syntactic deficit account for explaining certain L1-L2 differences is convincing. However, the proposal for the ‘overuse’ of discourse is too broad and needs to be refined in terms of the specific contexts and conditions under which learners have difficulty. We also believe that difficulty with cue-based retrieval is still a characterization of the symptoms of differences between L1-L2 processing, and does not necessarily address the source of the variability.

Cunnings (Cunnings) offers an interpretation of L2-L1 sentence processing differences in terms of memory principles. We applaud such cross-domain approaches, which seem likely to significantly elucidate the neurocognition of language. Cunnings attributes sentence processing differences between (adult) high proficiency L2 and L1 speakers to an increased susceptibility to similarity-based retrieval interference, rather than to qualitative L2-L1 processing differences (cf. Clahsen & Felser, 2006). On his account, both L1 and L2 sentence processing depend upon a ‘bipartite’ working memory, which involves maintaining items active by focusing attention on long-term memory representations (Cowan, 2001).

In an impressive review of recent research on the L2 processing of syntactic and anaphoric dependencies and ambiguities, Cunnings (Cunnings) convincingly demonstrates that the evidence from studies conducted within the last decade is not compatible with the Shallow Structures Hypothesis (SSH) which posits that L2 readers engage in structurally less detailed parses than native speakers. He also concludes that invoking capacity limitations in working memory in L2 learners cannot account for the full range of findings. Instead, he puts forth a novel approach to non-native processing in the context of cue-based models of memory retrieval. Cunnings assumes (i) L2 learners are more susceptible to interference, and (ii) L2 learners weight retrieval cues differently from native speakers.

Cunnings' keynote offers a new perspective on L2 processing by casting L1–L2 differences in terms of the memory system. One advantage of this approach is that it makes use of cue-based memory retrieval, a framework that has given rise to a wealth of research on L1 processing. However, there remain some questions about the evidence-base and predictions of the account, as well as conceptual challenges to its implementation.

Cunnings (2016) provides welcome insights into differences between native speaker (NS) sentence processing, adult non-native speaker processing (NNS), and working memory capacity (WMC) limitations. This commentary briefly raises three issues: construct operationalization; the role of first language (L1); and context.

Over the years, models proposed for second-language (L2) processing have been remarkably parallel to those proposed for Broca's aphasia. Differences between agrammatic and unaffected language processing have been explained, e.g., in terms of lack of detailed syntactic structure building (Grodzinsky, 1995), resource deficits (Haarman, Just & Carpenter, 1997), slow syntactic processing (Burkhardt, Avrutin, Piñango & Ruigendijk, 2008), or slowed lexical access (Love, Swinney, Walenski & Zurif, 2008). Each of these approaches have their homolog in L2 processing (e.g., Clahsen & Felser, 2006; McDonald, 2006; Dekydtspotter, Schwartz & Sprouse, 2006; Hopp, 2013, respectively). It is therefore not surprising that Cunnings's proposal (Cunnings, 2016) parallels another idea in aphasia and aging research, namely that deviations from healthy young adult monolingual sentence processing can be attributed to an increased susceptibility to interference (e.g., Sheppard, Walenski, Love & Shapiro, 2015).

Based on a detailed review of existing studies of high-proficiency second-language (L2) learners who acquired the L2 in adolescence/adulthood, Cunnings (Cunnings, 2016) argues that Sorace's (2011) Interface Hypothesis (IH) and Clahsen and Felser's (2006) Shallow Structure Hypothesis (SSH) do not explain the existing data as well as his memory-based approach which posits that memory-retrieval processes in the L1 and L2 do not pattern alike. Cunnings proposes that L1 and L2 processing differ in terms of comprehenders’ ability to retrieve from memory information constructed during sentence processing. He concludes that L2 processing is more susceptible to interference effects during retrieval, and, most relevantly for this commentary, that discourse-based cues to memory retrieval are more heavily weighted in L2 than L1 processing.

Grammatical processing in a second language (L2) that is learned in adulthood has been shown to differ from processing in a first/native language (L1). Clahsen and Felser's (2006) landmark article provided the first comprehensive account of these differences. According to their shallow structure hypothesis (SSH), L2 learners, unlike L1 speakers, do not compute abstract, hierarchical representations during online sentence comprehension; instead, they rely on lexical, semantic, and pragmatic information to build ‘good enough’ representations. However, native-like processing is attainable – with sufficient L2 proficiency – for word-level processing and morphosyntactic feature processing between locally related words. Clahsen and Felser's article spurred a prolific volume of research over the last decade. Some findings support the SSH, whereas others favor the competing claim that L1/L2 differences result from capacity-based limitations.

Analyzing L2 sentence processing in terms of cue-based memory retrieval is promising. But this useful general framework has yet to become a specific theory of L1-L2 differences.

Cunnings’ keynote article outlines a novel approach to native/non-native differences in on-line language comprehension by proposing that L2 speakers are more susceptible to cue-based retrieval interference than natives. Cue-based, parallel access approaches to processing have been prominent in monolingual studies for around 15 years now, but have barely been applied to L2/bilingual processing. We are particularly excited about the possibilities that this approach offers for understanding L1, L2 and bilingual processing, as well as individual differences. In this commentary, we focus on two issues: 1) whether the existing evidence for cue-based retrial mechanisms in L2 processing support a deficit model, as Cunnings seems to claim, and 2) how individual differences may explain both similarities and differences in L1 and L2 processing.

Cunnings’ proposed theory can explain why second language (L2) learners have difficulty resolving certain types of dependencies (i.e., backward-looking dependencies) but not others (i.e., forward-looking dependencies). However, his theory should be more explicit about the mechanism underlying late L2 learners’ and native speakers’ different weightings of retrieval cues, and research framed within his theory should strive to tease apart age-of-acquisition effects from bilingualism effects.

The target article makes a strong case that L2 comprehenders recover linguistic representations that are qualitatively similar to those recovered by L1 comprehenders. Moreover as they attempt to link non-adjacent elements, they do so using the same basic mechanism: cue-based retrieval in a content-addressable memory (Van Dyke & Lewis, 2003). In this commentary, I will not address the empirical adequacy of the argument, but instead consider some interesting theoretical challenges it poses for our understanding of working memory in sentence processing.

The primary aim of my target article was to demonstrate how careful consideration of the working memory operations that underlie successful language comprehension is crucial to our understanding of the similarities and differences between native (L1) and non-native (L2) sentence processing. My central claims were that highly proficient L2 speakers construct similarly specified syntactic parses as L1 speakers, and that differences between L1 and L2 processing can be characterised in terms of L2 speakers being more prone to interference during memory retrieval operations. In explaining L1/L2 differences in this way, I argued a primary source of differences between L1 and L2 processing lies in how different populations of speakers weight cues that guide memory retrieval.

This study investigates the relationship between intra-sentential codeswitching restrictions after subject pronouns, negative elements, and interrogatives and language-specific syntactic structures. Data are presented from two languages that have non-cognate lexicons but share identical phrase structure and syntactic mechanisms and exactly the same grammatical morphemes except for pronouns, negators, and interrogative words. The languages are the Quichua of Imbabura province, Ecuador and Ecuadorian Media Lengua (ML), consisting of Quichua morphosyntax with Spanish-derived lexical roots. Bilingual participants carried out un-timed acceptability judgment and language-identification tasks and concurrent memory-loaded repetition on utterances in Quichua, ML, and various mixtures of Quichua and ML. The acceptability and classification data show a main effect for category of single-word switches (significant differences for lexical vs. interrogative, negative, and for acceptability, pronoun) and repetition data show significant differences between lexical vs. interrogatives and negators. Third-person pronouns (which require an explicit antecedent) also differ significantly from lexical items. Logical-semantic factors may contribute to code-switching restrictions.