Download full-text PDF

The role of discourse-level expectations in non-native speakers' referential choices

Conference Paper (PDF Available)  · January 2014with91 Reads

Conference: Conference: Boston University Conference on Language Development, At Boston
Amy Schafer at University of Hawaiʻi at Mānoa
  • 17.99
  • University of Hawaiʻi at Mānoa
Theres Grüter at University of Hawaiʻi at Mānoa
  • 17.39
  • University of Hawaiʻi at Mānoa
Hannah Rohde at The University of Edinburgh
  • 22.2
  • The University of Edinburgh
BUCLD 38 Proceedings
To be published in 2014 by Cascadilla Press
Rights forms signed by all authors
The Role of Discourse-Level Expectations in
Non-Native Speakers’ Referential Choices
Theres Grüter, Hannah Rohde, and Amy J. Schafer
Expectation is a powerful mechanism in native-language processing.
Listeners child and adult use information from various sources to create
expectations about what is likely to come next. A variety of recent studies have
probed anticipatory effects by measuring comprehenders’ referential
expectations, i.e., their guess about who or what the speaker will mention next.
That work has found evidence that comprehenders are sensitive to cues such as
the lexical semantics of a verb, which can create expectations regarding an
upcoming referent in the current or next sentence (Altmann & Kamide, 1999;
Pyykkönen & Järvikivi, 2010). Similarly, there is evidence that comprehenders
are sensitive to the morphosyntactic properties of determiners which restrict the
possible nouns to follow (Dahan et al., 2000; DeLong et al., 2005; Lew-
Williams & Fernald, 2007). Expectations like these are not only characteristic of
language processing in native speaking adults, but also in children (Borovsky et
al., 2012). It thus appears that the ability to narrow down the potentially infinite
range of upcoming information through the generation of expectations is likely
to be a critical factor in explaining how we process language at the speed and
with the success we typically do, at least in our native language.
The use of expectations, however, is not present in all contexts across all
populations. Recent work indicates a decline in expectation generation in older
adults (Federmeier et al., 2002), suggesting that “predictive processing may not
be the best or even a viable strategy for all individuals at all phases of the
lifespan and/or in all processing situations” (Federmeier, 2007, p. 495). This is
because predictive processing comes with the risk of failure, that is, the
possibility that built-up expectations are not fulfilled. In such cases, the
processor will have to react swiftly and flexibly, and it will require sufficient
resources to do so and still keep up with the ongoing flow of information. This
may not always be possible, as for older adults with lower working memory
capacities (Federmeier et al., 2002). Non-native speakers may be another such
case; recent work points to limits in their expectation-driven processing at
Theres Grüter, University of Hawaii at Mānoa (UHM),; Hannah
Rohde, University of Edinburgh; Amy J. Schafer, UHM. Many thanks to Alexis Toliva,
Michelle Adams, and Mónica Vidal for assistance with data collection and annotation.
This research has been supported by grants from the National Science Foundation (BCS-
1251450) and the UHM Research Facilitation Fund, for which we are grateful.
lexical and morphosyntactic levels (Kaan et al., 2007; Lew-Williams & Fernald,
2010; Martin et al., 2013).
What emerges from this recent work is a generalization that non-native
speakers have Reduced Ability to Generate Expectations. We will refer to this
as the RAGE hypothesis. If RAGE influences non-native processing, one can
ask how non-native speakers’ ability to engage in predictive computations varies
across different levels (phonology, morphosyntax, semantics, world knowledge,
pragmatic inference, etc.) and whether RAGE interacts with other factors known
to affect second language (L2) processing, such as proficiency and first language
(L1) background. This paper focuses on a domain that has received little
attention in the literature on L2 processing, namely expectations at the discourse
level. While the syntax-discourse interface has been a topic of much recent
investigation in the L2 literature (see Sorace, 2011, for review), the role of
expectations in L2 processing of discourse phenomena such as coreference has
remained largely unexplored. Here we consider cross-sentence coreference and
ask whether and to what extent non-native speakers use cues known to guide
native speakersexpectations about who will be mentioned next in a discourse.
Using a method and context manipulation introduced in prior work (see
Sections 1 and 2), we conducted a story-continuation experiment to test whether
native and non-native participants make different use of available cues in
coreference processing. The results are consistent with the RAGE hypothesis:
Non-native speakers show native-like sensitivity to a cue that is available at the
point of coreference interpretation but show weaker sensitivity to a cue whose
effect requires predictive computation. An additional task rules out the
possibility that our non-native speakers lacked the requisite L2 knowledge to
understand the predictive cue. Their native-like performance in understanding
the interpretive properties of the cue thus contrasts with their ability to use the
cue to generate expectations.
1. Expectation-driven processing
Research on expectations relies on a variety of different methods to measure
comprehenders’ biases about upcoming information. A compelling case for
anticipation comes from eyetracking studies in which listeners can look to
potential upcoming referents in a visual-world scene. For example, the
selectional restrictions that a verb imposes on its direct object (e.g., The boy is
eating...) induce anticipatory looks to semantically plausible referents (e.g., a
cake rather than a toy; Altmann & Kamide, 1999). Likewise, the lexical
semantics of certain verbs (The butler frightened the guitarist or The guitarist
feared the butler) induces looks to a referent who is favored for re-mention via
causal reasoning (the causally implicated butler; Pyykkönen & Järvikivi, 2010).
Visual-world eyetracking also reveals that gender-marked determiners in French
and Spanish lead native speakers of those languages, both adults and young
children, to anticipate a noun of the same grammatical gender class (Dahan et
al., 2000; Lew-Williams & Fernald, 2007). ERP studies have been used to probe
expectations as well (DeLong et al., 2005): Readers’ ERP responses to a
determiner differ depending on whether its properties are compatible with the
phonological features of an expected upcoming noun (e.g., The day was breezy
so the boy went outside to fly a/an...). In all of these studies, anticipatory effects
emerge before the anticipated referent is mentioned. More indirect measures of
expectation rely on processing difficulty at the point when an unexpected word
is encountered. Difficulty may be observed in reading-time slowdowns or in
ERP components associated with surprisal (e.g., Kutas & Hillyard, 1984).
Expectations can also be probed via offline studies, by asking participants
to write a continuation following a prompt. Story-continuation tasks reveal
comprehenders’ biases about which referent among a set of referents they expect
to be mentioned again in a subsequent sentence (Arnold, 2001; Kehler et al.,
2008; Stevenson et al., 1994). A number of offline coreference results have been
replicated with tasks that measure online processing (e.g., Ferretti et al., 2009;
Koornneef & van Berkum, 2006), indicating that various effects of predictive
processing can be captured using online as well as offline tasks. In this paper,
we employ the story-continuation method to probe native and non-native
speakers’ coreference expectations.
While expectations are an important factor in native language processing,
their role appears to be diminished in non-native processing. Recent eyetracking
studies show that non-native speakers of Spanish make reduced or no use of
gender-predictive determiners (Grüter et al., 2012; Lew-Williams & Fernald,
2010), and non-native speakers of English fail to show native-like ERP
responses to phonologically-predictive determiners (Martin et al., 2013). Kaan et
al. (2007) report reading-time results showing that non-native speakers of
English do not use an extracted wh-phrase to predict a syntactic gap in contrast
to native speakers. These differences appear to be modulated by general
processing skills such as lexical access speed (Hopp, 2013) as well as by
properties of the speakers’ L1 and their L2 proficiency (Dussias et al., 2013).
What remains an open question is whether non-native speakers’ RAGE
persists at the discourse level. On one hand, the existing work on lexical and
syntactic expectations suggests that RAGE is pervasive. A discourse-level task
like tracking coreference may extend that difficulty, not only because
comprehenders must integrate a variety of linguistic and world-knowledge cues,
but also because there are no categorical rules dictating how those cues should
combine (unlike, for example, a grammatical gender cue). On the other hand,
coreference is one of the most fundamental and universal elements of
comprehension since it determines who or what a discourse is about. As such,
non-native speakers have abundant practice with it from their L1. If resources
are spread thin during non-native processing, coreference may stand as one of
the best candidates for resource allocation given its importance to understanding
the speaker’s message. The present study aims to delimit the scope of RAGE in
non-native language processing by focusing specifically on the discourse
2. Coreference processing
Tracking coreference across sentences depends on a variety of factors
some related to properties of the available referents such as their grammatical or
thematic roles, others related to properties of the unfolding discourse and the
events being described (Arnold, 2001; Caramazza et al., 1977; Kehler et al.,
2008; Stevenson et al., 1994). Here we focus on two discourse-level cues that
have been used in previous coreference studies with native speakers: event
structure and referential form. Both cues influence native speakers’ biases about
who a subsequent sentence will be about, but they do so in different ways.
Event structure is a cue that can be conveyed via a verb’s grammatical
aspect. Rohde et al. (2006) elicited story continuations following transfer-of-
possession sentences, as in (1), with either a perfective or imperfective verb. The
Source referent is in subject position; the Goal is the indirect object.
(1) John
handed/was handing a book to Bob
. He _______________
Perfective aspect, which describes a completed event, was shown to favor
continuations that described what happened next or as a result. Imperfective
aspect, which describes an event as ongoing, was shown to favor continuations
that elaborated or explained the transfer event. The different ways that the
continuations relate to the context sentence (their discourse coherence relation)
yield different coreference biases: Source continuations are more frequent in the
imperfective condition than the perfective, presumably because elaborations and
explanations favor the referent at the start state of an event, i.e., the Source,
whereas the referent associated with the end state or result of a transfer event is
likely to be the Goal. As such, verb aspect guides coreference biases via a fairly
complex predictive computation which depends on the comprehender’s ability
to build a mental model of the transfer event being described, to reason about
the coherence relation between the context sentence and a likely continuation,
and to predict which referent will be re-mentioned in that continuation.
Referential form, on the other hand, has been shown to influence
coreference biases via another discourse-level mechanism: information structure
(Rohde & Kehler, to appear). Specifically, the presence of a pronominal form
even the ambiguous one in (1)favors continuations about the subject referent
more so than a non-pronoun prompt. This follows from the fact that pronouns
are the preferred referential form for re-mentioning the topic of the discourse
(e.g., Gundel et al., 1993), and in a short passage like (1), the discourse topic is
typically assigned to be the subject of the context sentence. Referential form
hence only influences coreference biases via prompt type at the onset of the
continuation; no predictive computation is required beforehand.
In this paper, we examine the role of expectations in guiding referential
choices for native and non-native speakers. Note that this goes beyond previous
L2 work in which the focus has either been on non-expectation-driven
processing of coreference (e.g., Roberts et al., 2008; Sorace, 2011) or on
expectations at non-discourse levels. Using the story continuation paradigm
from Rohde et al. (2006), we test whether non-native speakers show sensitivity
to manipulations of event structure (perfective vs. imperfective) and prompt type
(pronoun vs. free). Under the RAGE hypothesis, we predict non-native speakers
to show a weaker effect of aspect on their referential choices than native
speakers. At the same time, we can expect native and non-native speakers to be
similar in their sensitivity to the referential form in the prompt. This is because
the aspect cue requires a predictive computation about how the upcoming
discourse will unfold (appearing at a point before any coreferring element has
been encountered), whereas referential form is a cue available only at the point
of coreference interpretation (when the participant encounters the prompt at the
beginning of the continuation and must make a decision about who that
continuation will be about).
Given that our study tests non-native speakers of English whose L1 is either
Japanese or Korean, it is worth noting that the predictive cue of interest, verb
aspect, induces the above-mentioned coreference effects for both Japanese and
Korean speakers in their native languages. Two recent studies confirm the effect
of aspect on coreference biases in transfer-of-possession contexts in Japanese
and Korean, despite differences in the inventory of pronominal forms (notably
the availability of null pronouns) between these languages and English. Ueno
and Kehler (2010) report that native Japanese speakers write more Source
continuations following imperfective than perfective context sentences, although
this pattern is restricted to overt pronoun prompts. Kim et al. (2013) report a
similar effect of aspect for native Korean speakers, in both overt and null-
pronoun conditions. Given that Japanese and Korean speakers show expectation-
driven effects in their native language, any reduction in these effects that we
observe in the current study in English cannot be attributed to L1 transfer and is
therefore likely to be a property of non-native language processing.
3. Method
3.1 Participants
All participants were recruited from the University of Hawai‘i community.
Participants in the native-speaker (L1) group (N=39) indicated that English was
the only language used systematically in their childhood homes. Participants in
the non-native speaker (L2) group (N=48) were primarily international or
exchange students at the time of testing (23 L1-Japanese; 25 L1-Korean). They
were first exposed to English between the ages of 8 and 13 years in school in
Japan/Korea; their length of exposure to English in the U.S. varied considerably,
ranging from 2 months to 16 years, yet most had spent less than one year in the
U.S. at the time of testing (median = 6 months).
Three measures of English language proficiency were collected from L2
participants: (i) performance on a written cloze test, (ii) self-ratings of their
English language ability, and (iii) performance on the Versant English Test, a
commercially available assessment of oral fluency (Pearson, 2011). L1
participants also completed (i) and (ii). Cloze test scores and self-ratings were
significantly higher in the L1 than in the L2 group (p<.001). No substantial
differences were found between the L1-Japanese and L1-Korean subgroups on
any measures. For the experimental measures, we therefore report only
aggregated results from all L2 participants regardless of L1 background here.
3.2 Materials and procedure
Participants completed two experimental tasks: a written story continuation
task adapted from Rohde et al. (2006) and a truth value judgment task designed
to assess participants understanding of verb aspect in English. The first task
(3.2.1) is the measure of interest, addressing our key research question: Do non-
native speakers show sensitivity to manipulations of event structure and
referential form? For outcomes from this task to be interpreted meaningfully,
however, independent evidence is required to establish that participants
understand the basic semantics of grammatical aspect in English, namely that
perfective denotes completed events, whereas imperfective describes ongoing or
incomplete events. The second task (3.2.2) was included for this purpose.
3.2.1 Story continuation task
The experiment employed a 2× 2 design, varying grammatical aspect of the
verb in the context sentence (perfective/imperfective) and referential form of the
continuation prompt (pronoun/free), as in (2). In the pronoun-prompt condition,
an ambiguous pronoun was provided as the first word of the continuation. In the
free-prompt condition, there were no restrictions on the continuation.
(2) a. Patrick gave a towel to Ron. (He) _________ [perfective]
b. Patrick was giving a towel to Ron. (He) ______ [imperfective]
Twenty critical items with transfer-of-possession verbs (5 per condition) and 20
fillers were distributed in one of four counterbalanced lists. As in (2), the Source
of the transfer-of-possession event was always the syntactic subject of the
context sentence, and the Goal was the indirect object. Participants were
instructed to imagine a natural continuation of the story, and write the first
continuation that came to mind, avoiding humor.
3.2.2 Knowledge-of-aspect task
The goal of this task was to assess whether L2 participants consistently
associate perfective and imperfective aspect with completed and incomplete
events respectively. In English, the interpretation of the imperfective be ing as
incomplete occurs with verbs of all event classes. In Japanese, however, the
imperfective marker te i- denotes incomplete events when combined with most
verbs, yet yields a resultative reading with others, specifically achievement
verbs (Gabriele, 2009). Prior work on aspect in L2 acquisition has shown that
Japanese learners of English, even at advanced levels of proficiency, do not
consistently rule out resultative interpretations of achievement verbs in English
(Gabriele, 2009). Given that transfer-of-possession verbs are typically
considered achievement verbs, potential null effects on the story continuation
task in the L2 group could thus be attributed to learners’ deriving the same
resultative interpretations from the context sentences in both the perfective and
imperfective condition, i.e., not distinguishing between the two. The knowledge-
of-aspect task, inspired by a story compatibility task originally designed by
Gabriele (2009), was included so that this possibility could be ruled out.
Participants read stories describing events that were either complete or
incomplete. Following the story, they were asked to judge the truth of a (written)
test sentence uttered by an observer (the cartoon character Pikachu) at a
particular point in time, by clicking true, false or not sure, as illustrated in (3).
(3) Story beginning:
Patrick and Ron are at the pool together. [picture of towel]
This is the towel that Patrick will give to Ron.
At 4:00, Ron is done swimming and ready to shower.
Story end, completed condition:
At 4:05, Ron disappears into the showers with the towel in his hand.
Story end, incomplete condition:
At 4:05, Patrick grabs the towel for Ron and walks over to the side of
the pool.
Test sentence:
At 4:05, Pikachu says: “Patrick is giving the towel to Ron.
For native English speakers, the test sentence in (3) is false in the completed
condition and true in the incomplete condition. The task consisted of 10
experimental items with an imperfective-marked transfer-of-possession verb
following a completed (k=5) or incomplete (k=5) event, as illustrated in (3). The
transfer-of-possession verbs were the same as those used in the story
continuation task. An additional 12 items were included as controls to ensure
that native and non-native speakers respond similarly in this task when no
relevant interpretive differences exist crosslinguistically.
4. Results
We begin by reporting the results from the knowledge-of-aspect task, as this
task constitutes a critical prerequisite for the interpretation of the results from
the story continuation task. In general, participants in all (sub)groups showed a
good understanding of verb aspect: They judged sentences with imperfective
aspect predominantly as ‘true’ in contexts with incomplete events and ‘false’ in
contexts with completed events; their judgments were reversed for perfective
aspect. T-tests were used to compare the percentage of ‘true’ responses for the
two different event types. Critically, L2 participants’ percentage of ‘true’
judgments for sentences with imperfective aspect differed significantly
depending on the completedness of the event (complete: 12%, incomplete: 82%;
t(47)=19.5, p<.001), indicating that they consistently associate the imperfective
with incomplete events, like L1 participants (complete: 2%, incomplete: 87%;
t(38)=32.5, p<.001). We thus conclude that the L2 participants understand the
interpretive consequences of aspect when combined with transfer-of-possession
verbs in English. We now turn to the question of whether they use this
knowledge to create expectations about next-mentions in a discourse.
Following protocols established by Rohde and colleagues in previous work,
two trained judges annotated continuations for intended reference of the
syntactic subject (Source, Goal, ambiguous, other), referential expression chosen
for the subject in the free prompt conditions (pronoun, name, other), and
coherence relation between the two sentences (not reported here). Examples of
continuation types are given in (4).
(4) Context: Patrick gave/was giving a towel to Ron. (He) _________________
a. He made sure to give him a clean dry one. (Source continuation)
b. He said “Thank you.” (Goal continuation)
c. He did not notice the puddle of water on the floor. (ambiguous)
d. The towel was still warm from the drying machine. (other)
Responses were classified as ‘ambiguous’ if both judges indicated that the
continuation was ambiguous or if one judge assigned a classification of Source
while the other assigned a classification of Goal. Ambiguous responses
accounted for 3.8% of the data in the L1 group, and 3.9% in the L2 group.
‘Other’ responses (12.8%/12.3% of L1/L2 data) and ambiguous responses were
excluded from further analysis. We thus report outcomes in terms of percentage
of Source-continuations out of Source- and Goal-continuations combined.
Figure 1 and the text report subject means. Analyses of variance were conducted
on the percentages; we report significance for effects and interactions that
achieved p<0.05 in both the by-participants and by-items analyses.
L1-pro L1-free L2-pro L2-free
% Source
Figure 1. % Source reference by aspect, prompt type, and group
A 2 (aspect) × 2 (prompt type) × 2 (group) mixed ANOVA showed the
predicted main effect of aspect (F1(1,84)=13.7, p<.001, F2(1,19)=11.0, p<.005),
whereby imperfective yielded more Source references than perfective (42.8%
vs. 30.0%). Aspect did not interact reliably with group (F1(1,84)=5.1, p<.05;
F2(1,19)=2.3, p=.15). Follow-up analyses within each group indicated that the
effect of aspect was robust in the L1 group (F1(1,39)=19.9, p<.001;
F2(1,19)=9.7, p<.01), but non-significant in the L2 group (F1(1,45)=1.2, p=.28;
F2(1,19)=1.6, p=.22), consistent with the predictions of the RAGE hypothesis.
As in previous work with native speakers (Rohde & Kehler, to appear), a
significant main effect of prompt type was observed (F1(1,82)=112.1, p<.001,
F2(1,19)=285.5, p<.001), whereby pronoun prompts yielded more Source
references than free prompts (52.3% vs. 14.8%). Prompt type did not interact
reliably with group (F1<1; F2(1,19)=6.1, p<.05). Follow-up analyses within
each group indicated that the effect of prompt was robust in both the L1 group
(F1(1,39)=61.1, p<.001; F2(1,19)=162.5, p<.001) and L2 group (F1(1,45)=51.4,
p<.001; F2(1,19)=176.4, p<.001), indicating that L1 and L2 participants were
equally sensitive to information structural properties of referential expressions in
English. This interpretation is further supported by an analysis of referential
forms chosen by participants in the free-prompt condition: Both L1 and L2
participants chose pronouns far more often when referring back to the Source
(L1: 65% pronouns, 34/52; L2: 82%, 32/39) than to the Goal (L1: 4%, 10/241;
L2: 20%, 63/308). There was no prompt type × aspect interaction (F1(1,84)=2.8,
p=.10; F2<1), nor a 3-way prompt type × aspect × group interaction (Fs<1).
Finally, the 2×2×2 ANOVA yielded a significant main effect for group
(F1(1,84)=7.1, p<.01; F2(1,19)=18.0, p<.001), driven by an overall bias for Goal
continuations in the L2 compared to the L1 group. This effect was not predicted,
yet fits with recent evidence from a story continuation task conducted with
children aged 5 to 6 years, where an analogous Goal bias was observed (Kehler
et al., 2011). We concur with these authors that this effect is most likely a
recency bias, given that the Goal is mentioned just before the prompt. These
results suggest that both children and adult L2 learners are more strongly
affected by recency in their referential choices than adult native speakers are.
5. Discussion
The main goal of this study was to investigate the extent to which non-
native speakers create expectations about who will be mentioned next in a
discourse based on linguistically encoded information about event structure, i.e.,
grammatical aspect, in the preceding context. We also asked to what extent non-
native speakers’ referential choices in our story continuation task were affected
by biases associated with different referential forms (pronouns vs. names, or
more specifically, pronoun prompts vs. unconstrained prompts that allowed
participants to produce names in their continuations). Our findings indicate
different answers to these two questions. Non-native speakers made appropriate
form choices in the free prompt condition, and showed the same sensitivity to
the prompt manipulation as the native speakers. Yet their referent choices were
less affected by the grammatically encoded event structure cue (aspect) in the
previous sentence than those of native speakers. This was the case even though
native and non-native speakers performed equivalently on an independent task
assessing knowledge of grammatical aspect in English.
Looking first at the prompt effect, findings from both groups of speakers
show more Source references with a pronoun than with a free prompt. This is
consistent with previous results and the widely accepted analysis that pronouns
are the preferred form (in languages like English) for reference to a recently
mentioned topic (e.g., Gundel et al., 1993). Since the prompt effect appears
equally robust for non-native and native speakers, we have no indication that
non-native speakers have reduced knowledge of the associations between
pronouns and topic maintenance and between fuller referring expressions and
shifts in reference. Critically, choices induced by the form of referring
expression (the prompt type) do not obviously depend on expectations. These
choices are forced at the point when forms are encountered, when a processing
decision regarding subject reference is necessary, and hence they draw on
information already present in the discourse. For example, when encountering
‘he’ (as in the pronoun prompt condition), a comprehender must access salient
entities consistent with the pronominal form to fully integrate the linguistic
material and proceed with language processing. Similarly, when encountering
the beginning of a new sentence (as in the free prompt condition), a processing
decision about the subject referent must be made at that point. Although it is
possible that these form-related biases are influenced by prior expectations, they
are forced when the prompt in our task is encountered, similar to how processes
such as lexical retrieval and structural integration are forced in reaction to
incrementally incoming cues that helps the comprehender derive meaning. Since
expectations are not critically involved in this process, the similar pattern for
native speakers and non-native speakers is fully consistent with the RAGE
Turning to the aspect manipulation, we assume based on previous research
(e.g., Kehler et al., 2008) that native speakers use aspect to build a mental model
that represents either an ongoing or completed event. These different types of
events then lead to different expectations about what kind of information will
appear next in a discourse, such as the probability of an explanation of the event
versus a description of the result of it. These different coherence expectations in
turn drive predictions for reference. Therefore, the aspect alternation does not
directly make the Source or Goal more salient in the speaker’s discourse
representation. Rather, the aspect effect is mediated by expectations about
coherence. If a non-native speaker understands the aspectual distinction between
imperfective and perfective (as our knowledge-of-aspect task shows), and
understands the associations between event structure and coherence patterns
(which are analogous in Korean and Japanese; see Section 2), but nevertheless
does not anticipate a coherence relation for upcoming sentences, s/he will
encounter the pronoun or free prompt without a coherence-mediated bias for
Source or Goal referencewhich could yield a reduced effect of aspect on
referential choice. Our findings are thus consistent with the predictions we
derived from the RAGE hypothesis, which states that non-native speakers have
reduced ability to generate expectations, that is, reduced ability to engage in
proactive processing, while their abilities in information integration, or reactive
processing, may be more closely aligned with those of native speakers.
We take these findings as a first indication that expectation generation at the
discourse level is reduced in a non-native language, consistent with evidence
from recent studies looking at anticipatory processing at the lexical and syntactic
level. At the same time, we must emphasize that our conclusions about
anticipatory processing are indirect, as they rely on assumptions about different
processing decisions involved in an offline task in which what we see is only
participants’ final choices. Based on evidence from story continuation tasks, we
cannot definitively exclude the possibility that the event-structure cue provided
by grammatical aspect is processed only at the point when the referential
decision has to be made, rather than the cue being used proactively when
encountered to incrementally update expectations. In addition, there are multiple
dimensions along which our two factors differ, including the point at which they
occur relative to decisions about reference, the strength of their effect within
native speakers, and the degree to which they draw on knowledge of real-world
event structures. Additional research will be necessary to tease apart these
various dimensions.
The results we have presented here from a written story continuation task
can thus provide preliminary support for the RAGE hypothesis at the level of
discourse processing. Further investigation is required to corroborate these
findings and probe the nature and scope of RAGE in non-native language
processing more fully. The intuition underlying the RAGE hypothesis is that a
limited capacity (L2) processor is stretched to its limits by processes that are
immediately required for dealing with incrementally incoming information, such
as lexical access and structural integration. In other words, it is fully occupied
by reactive processing, with little or no resources left for taking up non-essential
cues to update expectations, or severely limiting the scope of expectations to
immediately upcoming choices, such as the next word in a sentence. Testing
whether this intuition is on the right track will require the use of methodologies
that allow more control over and insight into incremental computations as
information is encountered. Moving from written stimuliwhich allow for
variation in reading rate and assignment of implicit prosodyto oral stimuli will
be a first step in this direction. Combining the presentation of such oral stimuli
with analyses of participants’ eye gaze patterns to a visual-world scene that
depicts competing referents could reveal expectation-driven effects in real time
(see e.g., Pyykkönen & Järvikivi, 2010). Finally, the inclusion of non-native
speakers with a broader range of proficiency levels than those in the present
study will be necessary to probe the interaction of RAGE with general L2
proficiency, and to determine whether RAGE is a developmental phenomenon,
or a more persistent characteristic of processing a non-native language. Studies
exploring these additional dimensions are currently under way in our laboratory.
Altmann, Gerry T. M., & Kamide, Yuki (1999). Incremental interpretation at verbs:
restricting the domain of subsequent reference. Cognition, 73, 247-264.
Arnold, Jennifer E. (2001). The effects of thematic roles on pronoun use and frequency of
reference. Discourse Processes, 31, 137-162.
Borovsky, Arielle, Elman, Jeffrey L., & Fernald, Anne (2012). Knowing a lot for one’s
age: Vocabulary skill and not age is associated with anticipatory sentence
interpretation in children and adults. Journal of Experimental Child Psychology,
112, 417-436.
Caramazza, Alfonso, Grober, Ellen, Garvey, Catherine, & Yates, Jack (1977).
Comprehension of Anaphoric Pronouns. Journal of Verbal Learning and Verbal
Behaviour, 16, 601609.
Dahan, Delphine, Swingley, Daniel, Tanenhaus, Michael K., & Magnuson, James S.
(2000). Linguistic gender and spoken-word recognition in French. Journal of
Memory and Language, 42, 465-480.
DeLong, Katherine A., Urbach, Thomas P., & Kutas, Marta (2005). Probabilistic word
pre-activation during language comprehension inferred from electrical brain activity.
Nature Neuroscience, 8, 1117-1121.
Dussias, Paola E., Valdés Kroff, Jorge R., Guzzardo Tamargo, Rosa E., & Gerfen, Chip
(2013). When gender and looking go hand in hand: Grammatical gender processing
in L2 Spanish. Studies in Second Language Acquisition, 35, 353-387
Federmeier, Kara D. (2007). Thinking ahead: The role and roots of prediction in language
comprehension. Psychophysiology, 44, 491-505.
Federmeier, Kara D., McLennan, Devon B., De Ochoa, Esmeralda, & Kutas, Marta
(2002). The impact of semantic memory organization and sentence context
information on spoken language processing by younger and older adults: An ERP
study. Psychophysiology, 39, 133146.
Ferretti, Todd R., Rohde, Hannah, Kehler, Andrew, & Crutchley, Melanie (2009). Verb
aspect, event structure, and coreferential processing. Journal of Memory and
Language, 61, 191-205.
Gabriele, Alison (2009). Transfer and transition in the SLA of aspect: A bidirectional
study of learners of English and Japanese. Studies in Second Language Acquisition,
31, 371-402.
Grüter, Theres, Lew-Williams, Casey, & Fernald, Anne (2012). Grammatical gender in
L2: A production or a realtime processing problem? Second Language Research, 28,
Gundel, Jeanette K., Hedberg, Nancy, & Zacharski, Ron (1993). Cognitive status and the
form of referring expressions in discourse. Language, 69, 274307.
Hopp, Holger (2013). Grammatical gender in adult L2 acquisition: Relations between
lexical and syntactic variability. Second Language Research, 29, 33-56.
Kaan, Edith, Dallas, Andrea, & Wijnen, Frank (2010). Syntactic predictions in second-
language sentence processing. In Jan-Wouter Zwart & Mark de Vries (Eds.),
Structure preserved. Festschrift in the honor of Jan Koster (pp. 207-213).
Amsterdam: John Benjamins.
Kehler, Andrew, Hayes, Emily, & Barner, David (2011). Pragmatically-driven biases in
children’s pronoun interpretation. Poster presented at the 24
annual CUNY
conference on human sentence processing, Stanford, CA.
Kehler, Andrew, Kertz, Laura, Rohde, Hannah, & Elman, Jeffrey L. (2008). Coherence
and coreference revisited. Journal of Semantics, 25, 1-44.
Kim, Kitaek, Grüter, Theres, & Schafer, Amy J. (2013) Effects of event-structure and
topic/focus-marking on pronoun reference in Korean. Poster presented at the 26
annual CUNY conference on human sentence processing, Columbia, SC.
Koornneef, Arnout W., & van Berkum, Jos J. A. (2006). On the use of verb-based
implicit causality in sentence comprehension: Evidence from self-paced reading and
eye tracking. Journal of Memory and Language, 54, 445465.
Kutas, Marta, & Hillyard, Steven A. (1984). Brain potentials reflect word expectancy and
semantic association during reading. Nature, 307, 161163.
Lew-Williams, Casey, & Fernald, Anne (2007). Young children learning Spanish make
rapid use of grammatical gender in spoken word recognition. Psychological Science,
18, 193-198.
Lew-Williams, Casey, & Fernald, Anne (2010). Real-time processing of gender-marked
articles by native and non-native Spanish speakers. Journal of Memory and
Language, 63, 447-464.
Martin, Clara D., Thierry, Guillaume, Kuipers, Jan-Rouke, Boutonnet, Bastien, Foucart,
Alice, & Costa, Albert (2013). Bilinguals reading in their second language do not
predict upcoming words as native readers do. Journal of Memory and Language,
69, 574-588.
Pearson (2011). Versant English Test: test description and validation summary. Palo
Alto, CA: Pearson Knowledge Technologies. (
Pyykkönen, Pirita, & Järvikivi, Juhani (2010). Activation and persistence of implicit
causality information in spoken language comprehension. Experimental Psychology,
57, 5-16.
Roberts, Leah, Gullberg, Marianne, & Indefrey, Peter (2008). Online pronoun resolution
in L2 discourse: L1 influence and general learner effects. Studies in Second
Language Acquisition, 30, 333-357.
Rohde, Hannah, Kehler, Andrew, & Elman, Jeffrey L. (2006). Event structure and
discourse coherence biases in pronoun interpretation. In Ron Sun (Ed.), Proceedings
of the 28th Annual Conference of the Cognitive Science Society (pp. 617-622).
Mahwah, NJ: Lawrence Erlbaum Associates.
Rohde, Hannah, & Kehler, Andrew (to appear). Grammatical and information-structural
influences on pronoun production. Language and Cognitive Processes: Special Issue
on Production of Referring Expressions.
Sorace, Antonella (2011). Pinning down the concept of interface’ in bilingualism.
Linguistic Approaches to Bilingualism, 1, 1-33.
Stevenson, Rosemary J., Crawley, Rosalind A., & Kleinman, David (1994). Thematic
roles, focusing and the representation of events. Language and Cognitive Processes,
9, 519-548.
Ueno, Mieko, & Kehler, Andrew (2010). The interpretation of null and overt pronouns in
Japanese: Grammatical and pragmatic factors. In Stellan Ohlsson & Richard
Catrambone (Eds.), Proceedings of the 32nd Annual Meeting of the Cognitive
Science Society (pp. 2057-2062). Austin: Cognitive Science Society.
  • ... A current theoretical hypothesis states that whereas native language processing tends to be highly anticipatory, with comprehenders continuously predicting upcoming information, non-native speakers show a "Reduced Ability to Generate Expectations" (RAGE hypothesis, Grüter & Rohde, 2013;Grüter, Rohde & Schafer, 2014). According to this hypothesis, rather than relying on predictive processing as occurs in L1 comprehension, L2 comprehension primarily relies on a posteriori integration and thus tends to be relatively less proactive than L1 processing. ...
    ... These results contribute to our understanding of the nature of L1 and L2 processing. Recent theoretical proposals hold that L2 processing tends to be less proactive than L1 processing, and that this factor may account for many differences observed between the L1 and the L2 across linguistic domains (RAGE hypothesis, Grüter & Rohde, 2013;Grüter et al., 2014;Grüter, Lew-Williams & Fernald, 2012;Hopp, 2013;Martin et al., 2013). Underlying this notion is the belief that typical L1 comprehension is highly proactive, in that good comprehenders continuously predict upcoming information on the basis of incrementing lexical, semantic and morphosyntactic cues. ...
  • ... This might lead to delays, less accurate performance, and/or less efficient processing of L2 sentences which, as a result, may interrupt successful prediction in bilinguals. This prediction is in line with previous work that states that bilinguals have weaker prediction mechanisms ( Grüter et al., 2012;Grüter, Rohde, & Schafer, 2014;Lew-Williams & Fernald, 2010;Martin et al., 2013). Grüter et al. (2014), for example, proposed the so-called RAGE (Reduced Ability to Generate Expectations) hypothesis for adult L2 learners. ...
    ... This prediction is in line with previous work that states that bilinguals have weaker prediction mechanisms ( Grüter et al., 2012;Grüter, Rohde, & Schafer, 2014;Lew-Williams & Fernald, 2010;Martin et al., 2013). Grüter et al. (2014), for example, proposed the so-called RAGE (Reduced Ability to Generate Expectations) hypothesis for adult L2 learners. This hypothesis holds that L2 adult speakers have a generally limited ability to make predictions in sentence processing, irrespective of whether semantic, lexical or morphosyntactic information cues the prediction. ...
  • ... From this finding, it can be argued that L2 learners appear to effectively employ semantic knowledge to resolve pronouns. The finding provides a counter evidence to RAGE hypothesis (Grüter, Rohde, & Schafer, 2014). The theory posits that L2 speakers have a reduced ability to generate expectations during language comprehension, meaning that semantic information poses a great challenge for L2 speakers who may not be able to effectively use it in predictive processing. ...
  • ... Several L2 studies, however, suggest that L2 learners are less efficient in using gender-marked determiners for generating predictions (Guillelmon & Grosjean, 2001; Grüter et al., 2012; Martin et al., 2013). In response to L2 learners' nonnativelike behavior in generating predictions, Grüter, Rohde, and Schafer (2014) proposed that learners have a reduced ability to generate expectations, which they call the RAGE hypothesis (see also Grüter et al., 2016). The authors claim that the reduced ability of anticipatory processing is attributed to a limited capacity of the L2 processor, which does not have enough resources for updating predictions after dealing with the immediately required processes. ...
    ... It is important to place the current findings squarely in the burgeoning field of L2 predictive processing. The issues in regard to limited resources, pointed out by the RAGE hypothesis (Grüter et al. 2014Grüter et al. , 2016), and to stored differences of frequency information (Kaan, 2014) guide us to further consider what would make a certain linguistic cue drain processing resources and introduce L1–L2 differences in incremental processing. There are certainly many possibilities, in light of the fact that Japanese passives present a number of structural differences compared with passives in English, the learners' L1. ...
    ... This predictive behavior has been shown to be an integral component of language processing that makes comprehension fast and efficient (Federmeier, 2007; Lau et al., 2006; Staub & Clifton, 2006). A critical issue in L2 processing research is whether the same processing strategies that have been identified in native speakers of a target language are also found in L2 learners' processing (Grüter, Rohde, & Schafer, 2014 Kaan, 2014). One of the areas that have been extensively studied in L2 processing research is gender agreement during lexical and syntactic processing. ...
  • ... Therefore, they may have been left with insufficient cognitive resources to effectively compute implicit causality and consequentiality biases to resolve pronominal reference. In addition, the present findings are consistent with the RAGE hypothesis (Grüter et al., 2014), which holds that L2 speakers have a reduced ability to generate expectations. The Bayesian analysis of pronoun resolution (Kehler et al., 2008; Kehler & Rohde, 2013) maintains that pronominal reference is, to a large extent, dependent on language comprehenders' prediction about upcoming referent in the top-down predictive stage, which is determined by their computation of local coherence relations based on available semantic and discourse information, that is, implicit causality and consequentiality in our case. ...
    ... Therefore, given the challenge of using semantic and discourse information, L2 speakers may prefer the easier subject/first-mention strategy, resulting in increased NP1 references. According to the RAGE account (Grüter et al., 2014), this NP1 bias could also be related to L2 speakers' reduced ability to make predictions. Subject/first-mention bias is one of the major biases in pronoun resolution (e.g., Crawley et al., 1990; Gernsbacher et al., 1989). ...
  • ... To date, the RAGE hypothesis has only been supported with offl ine behavioral data. Grüter et al. ( 2014 ) used a truth value judgment task to ascertain whether learners with Korean or Japanese as their native languages knew the meaning of English viewpoint aspect. In a second task, a story-continuation task (Rohde, Kehler, & Elman, 2006 ), the authors provided their learners with sentences such as (3) below and asked them to continue the story as they saw fi t. ...
    ... As a result of these weakened expectations, nonnative speakers are expected to spend more time integrating words into the representation of previously encountered text or may fail to display the effi ciency (and economy of time/effort) that accompanies anticipation. Two well-articulated hypotheses in the SLA literature—the shallow structure hypothesis (Clahsen & Felser, 2006a , 2006b) and especially the more recent RAGE hypothesis (Grüter et al., 2014 )—both maintain that nonnative language processing is different from native language processing. The SSH attributes this difference primarily to a lack of nativelike knowledge about hierarchical grammatical relations, and thus to an overreliance on semantic and pragmatic information during processing. ...
    ... In this respect, our study helps broaden the scope of long-distance dependency investigation by examining a dependency in which the second element is a pronoun rather than a gap. In our experimental setting, as in the one by Grü ter and colleagues (Grüter et al., 2014 ), we also rely on the discourse expectations of the speakers—the property we examine relates discourse appropriateness and grammatical representations. In this case, by presenting a displaced (topicalized) object, we set up the expectation that a clitic (which must precede a verb) should appear subsequently. ...


We investigate the timecourse of pragmatic interpretation of speech, using disfluency as a time-locked cue to a pragmatic change. Our current focus is on deception.
    Discourse-level factors, such as event structure and the form of referential expressions, play an important role in native speakers’ referential processing. This paper presents an experiment with Japanese- and Korean-speaking learners of English, investigating the extent to which discourse-level biases that have gradient effects in L1 speakers are also implicated in L2 speakers’ coreference... [Show full abstract]
    Conference Paper
    January 2015
      Native (L1) speakers take advantage of prenominal cues, such as gender-marked articles and classifiers, to identify an upcoming noun during online processing (e.g., Lew-Williams & Fernald, 2007; Huettig et al., 2010). The extent to which non-native (L2) speakers are able to do so remains a topic of ongoing investigation. Findings from learners of gender-marking languages have not been entirely... [Show full abstract]
      April 2018 · Cognition
        Comprehenders' perception of the world is mediated by the mental models they construct. During discourse processing, incoming information allows comprehenders to update their model of the events being described. At the same time, comprehenders use these models to generate expectations about who or what will be mentioned next. The temporal dynamics of this interdependence between language... [Show full abstract]
        Discover more