Modl 12149

Second Language Learner
Knowledge of Verb–Argument
Constructions: Effects of
Language Transfer and Typology
UTE RÖMER NICK C. ELLIS
Georgia State University University of Michigan
Department of Applied Linguistics and ESL Department of Psychology
34 Peachtree Street, Suite 1200 530 Church Street
Atlanta, GA 30303 Ann Arbor, MI 48109
Email: uroemer@gsu.edu Email: ncellis@umich.edu
MATTHEW BROOK O’DONNELL

The University of Pennsylvania
Annenberg School for Communication
3620 Walnut Street
Philadelphia, PA 19104
Email: mbod@asc.upenn.edu
This article examines second language (L2) learner knowledge of English verb–argument constructions
(VACs), for example, the ‘V against n’ construction. It investigates to what extent constructions underpin
L2 learners’ linguistic competence, how VAC mental representations in native speakers and learners
differ, and whether there are observable effects of the learners’ first language. Native speakers of English
and advanced learners of 3 different first language backgrounds (Czech, German, Spanish) were asked to
generate the first verb that came to mind to fill the gap in 20 sparse VAC frames like “she ____ against
the….” The comparison of learner and native speaker verb responses highlights crosslinguistic transfer
effects as well as effects of language typology that impact verb semantics (cf. Talmy, 1985). Our findings
suggest that learners whose L1 is, like English, satellite-framed (here Czech and German) produce more
target-like verbs than learners whose L1 is verb-framed (here Spanish).
Keywords: usage-based language acquisition; Construction Grammar; advanced learners; crosslinguistic
influence; satellite- vs. verb-framed languages
THIS ARTICLE PRESENTS SELECTED FIND- 1991, 2004; Stubbs, 2001). Written texts and
ings from a large research project at the interface spoken utterances are not just random sequences
of Corpus Linguistics, Construction Grammar, of individual words that can be solely explained
and language acquisition. Recent work in corpus on the basis of grammatical rules, but are made
linguistics has provided ample evidence for the up to a large extent of fixed or semi-fixed
highly patterned nature of language (e.g., Hun- elements that convey meanings. Cognitive linguis-
ston & Francis, 2000; Römer, 2005, 2009; Sinclair, tic theories of construction grammar posit that
language comprises many thousands of construc-
tions: form–meaning mappings, conventional-
The Modern Language Journal, 98, 4, (2014) ized in the speech community, and entrenched
DOI: 10.1111/modl.12149 as language knowledge in the learner’s mind
0026-7902/14/952–975 $1.50/0 (Bybee, 2010; Goldberg, 1995; Robinson &
© 2014 The Modern Language Journal
Ellis, 2008; Trousdale & Hoffmann, 2013).
Ute Römer et al. 953
Construction Grammar suggests a fixed corre- order to determine regularities in their acquisi-
spondence between a linguistic form and its tion and use (Ellis & Ferreira–Junior, 2009;
meaning and argues that combinations of Goldberg, 2006; Goldberg, Casenhiser, & Se-
words (‘constructions’) carry meaning as a whole thuraman, 2004; Ibbotson, 2013). These studies
(Goldberg, 2003, 2006). Psycholinguistic research conclude that there is a strong tendency for one
demonstrates language processing to be sensitive single verb to occur with particularly high
to usage frequency across many language pro- frequency in comparison to other verbs and
cesses and representations: phonology and pho- that the overall distribution of verbs in construc-
notactics, reading, spelling, lexis, morphosyntax, tions follows Zipf’s (1935) law, which states that
formulaic language, language comprehension, the frequency of words decreases as a power
grammaticality, sentence production, and syntax function of their ranks in the frequency table. The
(Ellis, 2002). That language users are sensitive to studies show how the frequencies of verbs
the input frequencies of constructions entails that influence acquisition, and how Zipfian distribu-
they must have registered their occurrence in tional properties of language usage help make
processing, and these frequency effects are thus language learnable, for both first and second
compelling evidence for usage-based models of language learners. The findings are revealing but
language acquisition (Bybee, 2006, 2010; Ellis, have yet to be backed up by evidence from more
2002; MacWhinney, 2001; Tomasello, 2003). constructions and larger datasets. Also needed is
Second language (L2) and first language (L1) experimental data on what speakers of English
learners alike share the goal of understanding know about the verbs that occur in particular
and producing language. Since they achieve this VACs. Evidence on speaker knowledge of VACs
based upon their experience of language usage, will help us determine whether constructions are
there are many commonalities between L1 and psychologically real and how strongly they are
L2 acquisition that can be understood from entrenched in the speaker’s mind.
corpus analyses of speaker input and from We have taken a large sample of 50 construc-
cognitive and psycholinguistic analyses of con- tions, identified and discussed in COBUILD
struction acquisition following associative and Grammar Patterns 1: Verbs (Francis, Hunston, &
cognitive principles of learning and categoriza- Manning, 1996), as a starting point for a
tion. Usage-based approaches, Cognitive Linguis- systematic analysis of VACs in the 100-million
tics, and Corpus Linguistics are thus increasingly word British National Corpus (BNC). In Römer,
influential in second language acquisition (SLA) O’Donnell, & Ellis (2015), we describe the steps
research (Collins & Ellis, 2009; Ellis, 1998, 2003; involved in mining the BNC for VACs and suggest
Ellis & Cadierno, 2009; Robinson & Ellis, 2008). a new approach to making verb construction
However, because L2 learners have previously analyses scalable. We have also carried out
devoted considerable resources to the estimation psycholinguistic experiments to capture native
of the characteristics of their native tongue in speaker and nonnative speaker associations of
which they have become fluent, their computa- verbs and the selected constructions. In Ellis,
tions and inductions are often affected by O’Donnell, & Römer (2014a), we use generative
transfer, with L1-tuned expectations and selective free association tasks to test the psychological
attention (Ellis, 2006; Ellis & Sagarra, 2011) reality of VACs in terms of their form–function
blinding the acquisition system to aspects of the representation, type–token distribution, verb–
L2 sample. Learned attentional biases from construction contingency, and semantic struc-
various L1s may influence the ultimate language ture. In one experiment, 285 native English
attainment of L2 learners from various L1 back- speakers generated the first word that came to
grounds (Ellis & Sagarra, 2011). SLA is thus mind to fill the verb slot in 20 sparse VAC frames
different from first language acquisition in that it such as ‘she _____ across the….’ In another
involves processes of construction and reconstruc- experiment, 40 native English speakers generated
tion. We explore these issues in this article. as many verbs fitting each VAC frame as they
In a collaborative project among psycho-, could think of in a minute. Through our large-
corpus-, and computational linguists, we study scale corpus analyses (based on the BNC), we
speaker knowledge and use of English verb– demonstrated the reliability and validity of VACs
argument constructions (henceforth VACs), such in language usage. We found that verb construc-
as the ‘V against n’ construction (e.g., he leaned tions are (a) Zipfian in their type–token distribu-
against the door frame) or the ‘V n n’ construction tions, with one verb type accounting for the lion’s
(e.g., they sent her a letter). Small sets of VACs have share of all VAC tokens, (b) selective in their
been analyzed in native and learner corpora in verb form occupancy, and (c) coherent in their
954 The Modern Language Journal 98 (2014)
semantics (for details, see Ellis, O’Donnell, & central aim of our present article is to uncover and
Römer, 2013; Römer et al., 2015). Through our discuss these differences and transfer effects in
psycholinguistic experiments, we demonstrated order to better understand which realizations of
the reliability and validity of VACs in language which VACs are not, or not yet, well entrenched in
users’ minds. We observed that adult native the minds of learners, and which ones are. In the
speakers of English represented similar bindings discussion of differences between native speaker
of form and function as retrieved from usage data. and learner knowledge of VACs, we consider
The verbs produced by fluent language users are issues of language typology that affect the verb
determined by (a) their token frequencies in the system, particularly the semantics of verbs. A
respective VAC in usage, (b) how faithful verbs are useful typological distinction, introduced by
to particular VACs in usage, and (c) the centrality Talmy (1985, 1991, 2000), can be made between
of the verb meaning in the VAC’s semantic verb-framed and satellite-framed languages,
network in usage (for details, see Ellis et al., which differ in how they encode the path and
2014a). manner of motion within the verb phrase. We will
Following the empirical design and methodol- provide an overview of these concepts in the
ogy described in Ellis et al. (2014a) and Römer following section of this article, followed by a
et al. (2015), we have also used corpus- and summary of our research questions and hypo-
psycholinguistic evidence to measure second theses. We will then describe the design and
language learner knowledge of VACs. We were implementation of the psycholinguistic experi-
interested in finding out whether, and to what ments carried out for this study, and summarize
extent, constructions also underpin L2 learners’ the data retrieval and evaluation steps. The core
linguistic competence. We were also interested in section of the article is dedicated to the discussion
determining how similar or different the mental of results on speaker knowledge of 20 selected
representations of common VACs are between VACs. We will end with a summary of main
native speakers and learners of English and findings, implications for instruction, and further
whether there are observable effects of the directions for related research.
learners’ first language. We had English native
speakers and advanced English language learners LANGUAGE TYPOLOGY: VERB-FRAMED AND
of three different first language backgrounds SATELLITE-FRAMED LANGUAGES
(German, Czech, and Spanish) complete the
same type of free association task (details provid- Languages differ in the ways in which verb
ed in the Data and Method section) and phrases express motion events. According to
compared responses across those four groups. Talmy (2000),
We correlated the results from these association
tasks (for L1 and L2 speakers) with results from the world’s languages generally seem to divide into a
large-scale corpus analyses of the same VACs. We two-category typology on the basis of the characteris-
found that learners have strong constructional tic pattern in which the conceptual structure of the
macro-event is mapped onto syntactic structure. To
knowledge and that, similar to native speakers,
characterize it initially in broad strokes, the typology
the VAC processing of L1 German, Czech, and consists of whether the core schema is expressed by
Spanish advanced learners of English, too, the main verb or by the satellite. (p. 221)
showed effects of frequency, contingency, and
prototypicality. These findings are discussed in
Ellis, O’Donnell, & Römer (2014b). Our discus- The core schema here refers to the framing event,
sion highlights similarities in the patterns that that is, to the expression of the path of motion.
underlie both first and second language VAC Talmy goes on to say that “[l]anguages that
acquisition. Our findings reflect L2 knowledge of characteristically map the core schema into the
language that comes from usage and indicate that verb will be said to have a framing verb and to be
all groups of participants are sensitive to distri- verb-framed languages” and that “languages that
butions in the language they are exposed to, albeit characteristically map the core schema onto the
to varying extents. satellite will be said to have a framing satellite and
One thing that Ellis et al. (2014b) does not to be satellite-framed languages” (p. 222; empha-
discuss is in what ways native speaker mental sis in original). Included in the former group are
representations differ from those of advanced Romance and Semitic languages, Japanese, and
language learners. It also does not provide details Tamil. Languages in the latter group include
on potential crosslinguistic transfer (Jarvis, 2013; Germanic, Slavic, Finno–Ugric languages, and
Odlin, 2013) from German, Czech, or Spanish. A Chinese. This means that a Germanic language
such as English often uses a combination of verb man. This observation is incorporated in the
plus preposition or particle (go into, jump over) discussion of our survey results.
where a Romance language like Spanish uses a
single form (entrar, saltar). RESEARCH QUESTIONS AND HYPOTHESES
While verb-framed languages express the path
of motion in the main verb and are path- The research questions we are addressing in
incorporating (Talmy, 1985) or path-type languages this article are:
(Mani & Pustejovski, 2012), satellite-framed
languages are manner-incorporating or manner-type
languages in which manner is expressed in the RQ1. Following from the observation that con-
main verb (e.g., English run, stroll). According to structions underpin L2 learners’ linguistic
Slobin (2003, p. 162), “English speakers get competence (Ellis et al., 2014b), how similar
manner for free.” They commonly use manner or different are the existing mental repre-
sentations of common VACs between
verbs in the expression of motion events and have
advanced L2 learners and native speakers?
more lexical items available to do so than speakers
of satellite-framed languages like Spanish. The RQ2. Are there observable differences in the
Spanish motion verb saltar, for example, has a mental representations of common VACs
range of English translation equivalents including among L1 German, L1 Czech, and L1
jump (over, up), leap, climb, skip, spurt, and hop. Spanish learners? Is one learner group closer
Manner of motion is a “highly saturated” semantic to the native speaker group than the others?
space in satellite-framed languages (Slobin, 2003,
RQ3. If there are such differences, can they be
p. 163). In verb-framed languages, manner of
explained on the basis of transfer from the
motion is less commonly expressed. It is “an learners’ first languages and/or on the basis
adjunct—an optional addition to a clause that is of language typology effects?
already complete” (Slobin, 2003, p. 162), such as a
participial form (e.g., Spanish entró corriendo,
“enter running”). We therefore assume manner All groups of speakers in our study are asked to
of motion to be a less entrenched, less salient produce verbs in response to VAC frames the
concept in the minds of speakers whose L1 is verb- majority of which encode a path of motion, with
framed. The concept is less easily codable and the path expressed by a satellite (a particle or
requires additional effort to express. Cifuentes– preposition). Against the background of the
Férez and Gentner (2006) provide empirical language-typological issues discussed in the
evidence in support of this assumption by showing previous paragraphs, our research hypotheses
that, in a word mapping task, Spanish speakers (H) are:
were more likely to infer a path interpretation of a
novel motion verb than a manner interpretation.
English speakers showed the opposite behavior H1. The mental VAC representations of
and favored manner over path interpretations German, Czech, and Spanish advanced
(see Brown & Gullberg, 2011; Cadierno, 2008, learners of English will differ in diverse ways
2013; and Slobin, 2003, 2006, for reviews of from native speakers’ mental VAC
additional studies that demonstrate similar effects representations, showing that learners are
biased by their L1s.
of language typology on linguistic production).
Whereas Slavic languages are generally consid- H2. Learners whose L1 is satellite-framed (and
ered satellite-framed (Slobin, 2003, 2006), hence typologically similar to English) will
Gehrke (2008) cautions that Czech is “neither produce more target-like verbs (verbs that
straightforwardly verb-framed nor straightfor- correlate more closely with those produced
wardly satellite-framed” (p. 203). While motion by L1 English speakers) than speakers whose
and manner in Czech are included in the verb (as L1 is verb-framed.
is typically the case for a satellite-framed lan- H3. Speakers of satellite-framed languages (here
guage), paths of motion may be mapped onto the German and Czech, even though the latter
verb and/or a directional preposition. To give is not a clear-cut case) will produce more
one example, Czech offers three ways of express- verbs that express specific manners of
ing jump over: skočit přes (‘jump over’), přeskočit přes motion in the verb generation tasks (in line
with native speakers).
(‘overjump over’), and přeskočit (‘overjump’).
Czech hence appears to be a less prototypical H4. Conversely, speakers of a verb-framed lan-
satellite-framed language than English or Ger- guage (here Spanish) will produce specific
manner of motion verbs less frequently and associated with a particular category (Battig &
instead respond with more general motion Montague, 1969; Rosch & Mervis, 1975). The
verbs such as GO, COME, or MOVE. actual instructions that participants received are
H5. The verb responses of all learner groups given in Figure 1. After the survey instructions, the
will show effects of collocational transfer participants saw the 20 sentence frames displayed
(Yamashita & Jiang, 2010) from the in Table 1, shown once with either she or he as
learners’ first languages. subject and once with it as subject. These 40
prompts were presented in random order and
participants filled the gaps in each frame. For
We will refer back to these hypotheses in our each VAC, we recorded the verbs produced and
results discussion and conclusion. the participants’ response times. The entire
survey took between 5 and 15 minutes to
DATA AND METHOD complete.2
The participants were predominantly university
The data collected for this study come from a students recruited through emails sent by mem-
series of psycholinguistic experiments adminis- bers or associates of the research team, either to
tered online using the Qualtrics survey system.1 the students directly or (in the case of the learners
Native English speakers and German, Czech, and who participated) to one of their instructors. The
Spanish advanced learners of English (described English native speakers were mostly students
in more detail later) completed the same genera- enrolled at a large Midwestern research universi-
tive free association task that presented them with ty. The L1 German, L1 Czech, and L1 Spanish
VAC frames such as ‘she _____ off the…’ or ‘it learners were students enrolled at research
_____ over the…’ and asked them to type the first universities in Germany, the Czech Republic,
word that came to mind to fill the blank. Free and Spain, respectively. The learners in all three
association tasks like this are standard in psychol- groups had been in instructed EFL settings in
ogy for determining which items are most closely Germany, the Czech Republic, or Spain for at least
FIGURE 1
Instructions Given to Participants at the Beginning of the Online Survey
TABLE 1
Selected Verb–Argument Constructions (VACs) and Prompts Used in Experiments
Selected VACs Survey Prompts

V about n he _____ about the…; it _____ about the…
V across n she _____ across the…; it _____ across the…
V after n he _____ after the…; it _____ after the…
V against n she _____ against the…; it _____ against the…
V among n she _____ among the…; it _____ among the…
V around n he _____ around the…; it _____ around the…
V as n she _____ as the…; it _____ as the…
V at n2 he _____ at the…; it _____ at the…
V between n he _____ between the…; it _____ between the…
V for n she _____ for the…; it _____ for the…
V in n he _____ in the…; it _____ in the…
V into n she _____ into the…; it _____ into the…
V like n he _____ like the…; it _____ like the…
V of n he _____ of the…; it _____ of the…
V off n she _____ off the…; it _____ off the…
V over n she _____ over the…; it _____ over the…
V through n he _____ through the…; it _____ through the…
V towards n she _____ towards the…; it _____ towards the…
V under n he _____ under the…; it _____ under the…
V with n she _____ with the…; it _____ with the…
7 years. The mean number of years of formal learner responses with lists based on English
English instruction was 10.04 years for German, native speaker responses: L1 German vs. English,
11.37 for Czech, and 12.68 for Spanish learners. L1 Czech vs. English, and L1 Spanish vs. English.
According to their instructors, the proficiency We used a simple regression general linear model
levels of the German and Czech learners corre- (GLM) framework to build models for each of
sponded to level C1 in the Common European these language pairings for each of the 19 VACs.
Framework of Reference for Languages (CEFR), We used R (R Development Core Team, 2012) to
described as the “Effective Operational Proficien- perform statistical analyses of the data. Verb
cy” level (Council of Europe, 2001). Our Spanish frequency in the English sample was taken as a
contacts reported that the majority of students proxy for native speaker VAC knowledge (i.e., this
who participated in the survey were advanced is what a native speaker says confronted with the
learners at CEFR level C1 while some of them (an VAC frame) and was used as the single predictor
estimated 10%) were at level B2 (“Vantage” level). of verb frequency in the L1 German, Czech, and
Our contacts confirmed that they did not share Spanish samples. Frequencies were log trans-
the survey link with learners at lower levels of formed due to the Zipfian nature of the distribu-
proficiency. tion (i.e., a long-tailed distribution, see Tables 3 to
The following numbers of participants volun- 5) to bring them into linear space for the
teered to complete the VAC survey: 285 native comparison of the two L1 distributions. Nonoc-
English speakers, 276 L1 German learners of currence (i.e., zero frequency) of a verb in one L1
English, 185 L1 Czech learners of English, and background sample that occurs in the other (e.g.,
131 L1 Spanish learners of English. To ensure none of the 131 native speakers used ARGUE in the
comparability across datasets, we based our ‘V against n’ frame but 5 of the German L1
analyses on only 131 responses from each of the speakers did) becomes an issue because the
four participant groups, including all of the L1 logarithm of zero is not defined. It is, however,
Spanish responses and 131 randomly selected important to include instances such as ARGUE and
responses each from the native speaker, L1 SIT (5 occurrences in native speaker responses
German, and L1 Czech groups. For each group, and none in L1 German) in the ‘V against n’
the lists of responses were lemmatized by verb type comparison of L1 German responses to native
(e.g., runs, ran, was running etc. ! RUN) and speaker responses as they contribute to the overall
ordered by verb token frequencies.3 We then shape of the distribution which we take as a proxy
carried out comparisons of lists based on the for speaker knowledge of the VAC. Therefore,
when computing the correlations and plotting RESULTS: COMPARING NATIVE AND
these items, we use the log value 0.1. We NONNATIVE SPEAKER VAC KNOWLEDGE
examined the correspondence between the verb
frequency distributions of paired languages using Our discussion of results from the verb list
a simple regression framework where verb fre- comparisons begins with an overview of correla-
quency in learner responses (i.e., L1 German, tions between the native speaker and learner
Czech, or Spanish) within a specific VAC is taken responses for the VACs listed in Table 1. For each
as the dependent variable and frequency in native VAC and each comparison (L1 German vs.
speaker responses as the independent (predictor) English, L1 Czech vs. English, L1 Spanish vs.
variable.4 We used the standardized residuals; that English), we also report which verbs have particu-
is, the amount of divergence between the larly high (positive and negative) standardized
predicted and actual values. While residuals are residuals and are unusually frequent or infrequent
commonly used in regression analysis as a way of in the learner responses compared to the native
identifying outliers in the data that may be overly speaker responses. This overview of results for all
influencing the model or leading to a poor fit, we VACs is followed by a detailed analysis of three
use them here on an item-based level as an VACs that have been selected to provide us with a
indicator of the over- or underuse of a verb by a more in-depth picture of potential L1 transfer
nonnative speaker compared to the native norm. effects and effects of language typology on the
The basic insight behind this method is that a learners’ survey responses. The selected VACs are
statistical model can be built based on empirical ‘V against n,’ ‘V in n,’ and ‘V over n.’
data gathered from native speakers and be
thought of as a model of “what would a native Overview of VACs
speaker do?” This model can then be used to
predict the responses for nonnative speakers. The Table 2 shows the overall correlations between
predicted and the actual values can be compared learner and native speaker responses to the survey
both across the whole distribution (i.e., correla- prompts listed in Table 1. Figure 2 provides a
tion) and on an item-based level (i.e., residuals) to visual representation of these correlations, with
look at under- and overuse of specific items and to data points represented by prepositions. The
identify potential areas of L1 interference and possible range of values is 0 to 1. The closer the
influence. value is to 1, the stronger the correlation between
TABLE 2
Correlations Between Learner and Native Speaker Responses (n ¼ 131 per Group)
VAC L1 German L1 Czech L1 Spanish

V about n 0.81 0.78 0.75
V across n 0.84 0.73 0.78
V after n 0.77 0.69 0.62
V against n 0.62 0.54 0.55
V among n 0.63 0.30 0.47
V around n 0.82 0.76 0.75
V as n 0.62 0.40 0.40
V between n 0.63 0.68 0.57
V for n 0.72 0.78 0.72
V in n 0.79 0.69 0.35
V into n 0.86 0.89 0.70
V like n 0.72 0.68 0.70
V of n 0.76 0.73 0.71
V off n 0.83 0.69 0.56
V over n 0.72 0.76 0.48
V through n 0.81 0.67 0.62
V towards n 0.90 0.80 0.81
V under n 0.71 0.75 0.70
V with n 0.73 0.60 0.58
Average: 0.75 Average: 0.68 Average: 0.62
FIGURE 2
Visual Representation of Correlations Between Learner and Native Speaker Responses
the responses. Correlation figures express how groups (i.e., a correlation of 1), all verb labels
much the sets of verbs produced by a group of would be neatly placed along the diagonal
learners (both in terms of types and tokens) through the middle of the graph. This is not
overlap with the sets of verbs produced by the the case in any of our three comparisons. Instead,
group of native speakers in response to the same verbs are scattered to the left and right of the
VAC prompt. Figure 3 provides three graphs that diagonal in all three graphs. Verbs that appear to
illustrate this comparison of the verb responses the left of (or above) the diagonal are markedly
given by German/Czech/Spanish learners and more frequent in the learner than the native
native speakers for one of the selected VACs: ‘V in speaker responses; verbs that appear to the right
n.’ The x-axis shows the logarithmic frequency of of (or below) the diagonal are markedly less
the verb type in the native speakers’ responses; the frequent in the learner than the native speaker
y-axis shows the logarithmic frequency of the verb responses. In the right hand scatterplot panel in
type in the L2 learners’ responses. If there were Figure 3 (L1 Spanish vs. native speakers), most
perfect overlap in verb responses between two verbs are much farther away from the diagonal
FIGURE 3
Correlations of Verb Responses Between Three Groups of Learner Responses (L1 German, left panel;
L1 Czech, middle panel; L1 Spanish, right panel) and Native Speaker Responses (L1 English) for ‘V in n’
than in the middle (L1 Czech) and left panels (L1 Czech) to 0.9 (‘V towards n,’ L1 German). As
German). L1 Spanish learners respond with verbs Figure 2 indicates, L1 German vs. English
to this VAC that are quite different from those correlations are much more homogeneous across
produced by native speakers. For example, these VACs (0.62 to 0.9) than L1 Spanish vs. English and
learners produce BE, LIVE, and STAY comparatively (even more so) L1 Czech vs. English correlations
more often and GO, LOOK, and SIT comparatively (0.35 to 0.81 and 0.3 to 0.89 respectively).5 For
less often than native speakers. This lack of L1 German, we also observe a higher average
overlap is reflected in the rather low correlation correlation of 0.75 than for L1 Czech (0.68) and
figure of 0.35 (compared to values of 0.69 and L1 Spanish (0.62). None of the L1 German vs.
0.79 for Czech and German). English correlations falls below 0.6, whereas three
Across the 57 datasets captured in Table 2 and of the L1 Czech correlations (for ‘V against n,’ ‘V
Figure 2 (19 VACs times three learner groups), among n,’ and ‘V as n’) and eight of the L1 Spanish
correlations range from 0.3 (‘V among n,’ L1 correlations do (for ‘V against n,’ ‘V among n,’ ‘V as
n,’ ‘V between n,’ ‘V in n,’ ‘V off n,’ ‘V over n,’ and ‘V Appendix A indicates a number of verb
with n’). This means that, overall, the German preferences that are shared across the three
learner responses most closely and the Spanish learner groups. Verbs that are produced signifi-
learner responses least closely match the native cantly more frequently by learners in all groups
speaker responses, with the Czech learner re- than by native speakers are: COME for ‘V across n’
sponses falling somewhere between these two and ‘V towards n’; LOOK for ‘V after n,’ ‘V as n,’ and
groups (see our earlier comments on Czech’s ‘V into n’; BE for ‘V against n’ and ‘V among n’; MOVE
status as a less clear-cut exemplar of a satellite- for ‘V around n’; STAY, BE, and LIVE for ‘V in n’; and
framed language). It appears that, at least with STAY for ‘V with n.’ All of these verbs have high
respect to a large number of VACs, Spanish frequencies in general English language use and
learners’ form–meaning mappings are less in line appear to be highly entrenched in the learners’
with native speaker peers than those of German or minds. Appendix A also indicates that a large
Czech learners, confirming our Hypothesis 1, number of verbs with high positive standardized
which predicted that our different learner groups residuals are not shared by learners of different L1
would differ from the native speaker group in backgrounds. To give a few examples: THINK is
diverse ways. This is particularly true for the VACs overused by German learners in response to the
‘V against n,’ ‘V among n,’ ‘V as n,’ ‘V between n,’ ‘V ‘V about n’ frame but not by Czech and Spanish
in n,’ ‘V off n,’ ‘V over n,’ and ‘V with n,’ all of which learners who favor SPEAK instead. For ‘V among n,’
have below average correlations (see “L1 Spanish” German learners show strong associations with
column in Table 2). The corresponding correla- COME and STAND, Czech learners with BELONG and
tions for L1 German and L1 Czech tend to be STAND, and Spanish learners with APPEAR and GO.
higher, often considerably so (with the exception For ‘V through n,’ the verbs with the highest
of ‘V among n’ and ‘V as n’ in the L1 Czech positive residuals are WALK and CLIMB for German
dataset). We will investigate a selection of these learners, GET and SEE for Czech learners, and GO
VACs and related learner verb preferences in and PASS for Spanish learners.
more detail in the following sections. This The verb GO is a particularly interesting case. GO
confirms our Hypothesis 2, predicting that appears in the L1 Spanish positive standardized
Spanish learners find it harder than German residuals lists for 7 out of 19 VACs. Responding to
and Czech learners to produce verbs that corre- a VAC frame with a form of this verb appears to be
late closely with those produced by native English a productive strategy for the Spanish survey
speakers. participants. Other general motion verbs that
Our analysis of standardized residuals aimed are overused by L1 Spanish learners are COME and
at highlighting verbs that are either particularly MOVE. The Spanish learners in our study seem to
common in the learner responses (high positive favor these general verbs (especially GO) over
standardized residuals) or particularly rare in more specific manner of motion verbs that appear
or absent from the learner responses (high in the negative standardized residuals lists (e.g.,
negative standardized residuals), always in JUMP, RUN, CRAWL, SLIP). The German and Czech
comparison with the native speaker responses learners also overuse general motion verbs with
to the same VAC frames. We consider residuals individual VACs but not as often as the Spanish
that fall outside of þ2 or 2 standard deviations learners. The verb GO appears in two of the
unusual and include the corresponding verbs German and four of the Czech positive standard-
and absolute response token frequencies in ized residuals lists. This is evidence in support of
the table in Appendix A. To facilitate data our Hypotheses 3 and 4, predicting that Spanish
interpretation, we use gray shading for verbs learners will indicate manner of motion less often
with negative standardized residuals below 2. than native English speakers and German and
We are interested in verb selection patterns Czech learners will. Our finding is in line with
that emerge across VACs and across L1s. In line Cadierno’s (2010) observation that beginning to
with Hypotheses 3 and 4, we expect that, in intermediate L1 Spanish learners of Danish (like
their verb responses to VAC frames that serve English, a satellite-framed language) did not
to express directed motion events, Spanish produce specific manner of motion verbs in a
learners of English will indicate manner of production task but instead overgeneralized and
motion less often than native English speakers used Gå (GO, WALK) in all walking-related contexts
and German and Czech learners do. Instead, whereas Russian and German learners of Danish
we expect Spanish learners to overuse general employed more manner of motion verbs. Another
motion verbs that do not express a specific interesting pattern is the occurrence of BE in the
manner. positive residuals lists for a number of VACs. This
applies to all three learner groups, although to a rely less on specific, lower frequency verbs and
much larger extent to Spanish than German and more on general, high-frequency verbs (see also
Czech learners. BE is among the significantly Römer et al., 2015). We will comment more on
overused verbs in Spanish learner responses to 12 particular differences between learner and native
VACs including ‘V around n,’ ‘V between n,’ ‘V speaker VAC responses in the following sections.
towards n,’ and ‘V under n’ (compared to only six
VACs in the German and five VACs in the Czech Zooming in on ‘V Against N’
datasets). This provides additional support for
our Hypothesis 2. Spanish learners find it more ‘V against n’ is a VAC with particularly low
difficult to retrieve specific target-like lexical verbs correlation values. Correlations are below average
when confronted with bare VAC frames of the for all three L1s (0.62 for German, 0.54 for Czech,
‘s/he _____ preposition’ kind than German and and 0.55 for Spanish). We therefore expect to
Czech learners do. Instead, they often opt for find considerable variation in verb choices
forms of the semantically bleached verb BE. between native speaker and learner responses
Even more evidence in support of (especially (in terms of verb types, verb token numbers, or
Spanish) learners’ avoidance of specific motion both). Table 3 shows lemmatized lists of the 20
verbs in the free association task can be found most frequent verbs produced by the four groups
in the negative standardized residuals included of survey participants in response to the prompts
in Appendix A (shaded gray). While there is ‘she _____ against the . . .’ and ‘it _____ against
considerable variation across L1 groups (and the . . . .’ The native speaker responses in the left
across VACs) with respect to underused verbs, a hand column (shaded gray) serve as a reference
common feature of many of the verbs with high point for comparisons with the German, Czech,
negative standardized residuals is that they are less and Spanish learner responses. Verbs are itali-
frequent in general English language use than cized in a learner list if they also appear in the
most of the overused verbs discussed in the native speaker list.
previous paragraphs. Examples include REVOLVE Of the 20 verbs most often produced by L1
and CIRCLE (‘V around n’), SLIP and FALL (‘V between German survey participants, 10 are shared with
n’), REACH (‘V for n’), BUMP (‘V into n’), SWIM (‘V like the native speaker list. While this may indicate
n’), HOP ‘V over n’), and CRAWL (‘V under n’). considerable overlap, the actual token numbers
Compared to the native English speakers, the and rank positions of these verbs are rather
advanced learners who participated in our study different. The verbs BE, FIGHT, and HIT, for
TABLE 3
‘V Against n,’ Top 20 Verbs in Native Speaker and Learner Responses
Rank Native Speakers German Learners Czech Learners Spanish Learners

1 LEAN 23 BE 29 FIGHT 27 FIGHT 40
2 PUSH 13 LEAN 14 BE 25 BE 31
3 BE 13 FIGHT 12 SPEAK 12 STAND 5
4 FALL 12 RUN 12 LEAN 11 PLAY 5
5 RUN 10 HIT 8 STAND 9 GO 5
6 GO 10 FALL 7 VOTE 6 LEAN 4
7 FIGHT 6 ARGUE 5 PUSH 5 ARGUE 3
8 RAG 5 VOTE 5 RUN 5 FALL 3
9 SIT 4 REBEL 4 GO 5 SPEAK 2
10 PROTEST 3 GO 3 PROTEST 4 PUSH 2
11 WORK 3 WALK 3 ARGUE 3 REACT 2
12 HIT 3 DEMONSTRATE 3 RISE 2 RUN 2
13 RISE 2 KICK 2 COME 2 CRASH 2
14 RAIL 2 CRASH 2 MOVE 2 STAY 2
15 REST 2 PROTEST 1 OBJECT 1 SAVE 1
16 CROUCH 1 RISE 1 TURN 1 CHANGE 1
17 STRUGGLE 1 WORK 1 SAY 1 CAN 1
18 RAM 1 SPEAK 1 ROLL 1 CLAIM 1
19 BUMP 1 SHOUT 1 DECIDE 1 DISCUSS 1
20 FLY 1 STUMBLE 1 HIT 1 PLACE 1
Note. Italicized verbs indicate overlap between a learner list and the native speaker list.
instance, are more frequent in the German constitutes additional evidence in support of our
learner than the native speaker responses, where- Hypothesis 5.
as native speakers more often produce forms of
LEAN, FALL, and GO in response to against frames.
Zooming in on ‘V in N’
The two lists suggest that native speakers and
German learners have different semantic associ- The correlation values for ‘V in n’ vary
ations with this VAC. Native speakers associate considerably across learner groups. Correlations
verbs that express (forced) physical contact or are high for German (0.79), slightly above average
collision with ‘V against n,’ especially the top- for Czech (0.69), and extremely low for Spanish
ranked LEAN (23 instances) and PUSH (13 instances; (0.35). We hence expect strong overlap in terms
not in the German list), but also the less frequent of verb preferences between native speaker and
SIT, REST, RAM, BUMP, and FLY that do not appear in German and Czech learner responses. We expect
the learner list. German learners show weaker the verb choices of Spanish learners to be rather
associations with these verbs and instead produce different from those of native speakers and from
verbs that express a (mostly verbal) reaction or those of their German and Czech peers. Table 4
argument, including FIGHT, ARGUE, VOTE, REBEL, shows lemmatized lists of the 20 most frequent
DEMONSTRATE, SPEAK, and SHOUT. A possible expla- verbs produced by the four groups of survey
nation for this semantic preference is crosslin- participants in response to the prompts ‘he _____
guistic transfer from German where the verbs in the . . .’ and ‘it _____ in the . . . .’ The native
KäMPFEN (‘FIGHT’), PROTESTIEREN (‘PROTEST’), VOTIE- speaker responses in the left-hand column
REN (‘VOTE’), and STIMMEN (‘VOTE’) are among the (shaded gray) serve as a reference point for
most significant left-hand collocates of gegen,6 the comparisons with the German, Czech, and
translation equivalent of against. Verbs that Spanish learner responses. Verbs are italicized
express the meaning of PUSH (‘DRüCKEN’) or FALL in a learner list if they also appear in the native
(‘FALLEN’) tend to be used without or with a speaker list.
different preposition (DRüCKEN auf, put pressure on; Indeed, for this VAC, we observe much more
FALLEN in/auf/von, FALL in/on/off). Learners’ verb overlap between native speaker and German (13
responses appear to be influenced by collocation- verbs) and native speaker and Czech (14 verbs)
al preferences in their L1, providing evidence in top-20 lists than between native speaker and
support of our Hypothesis 5. Spanish lists (7 verbs), further confirming Hy-
Czech learners show similar patterns of overlap pothesis 1. This higher degree of overlap for
and semantic preference as German learners. German and Czech than for Spanish responses
They share 9 verbs (out of 20) with the native also became apparent in the graphs provided in
speakers and have a strong preference for verbs Figure 3. The shared verbs do, however, occupy
that express a (verbal) reaction against some- different ranks across lists and/or have quite
thing, including FIGHT (rank 1), SPEAK (rank 3), different token frequencies. Although shared
VOTE (rank 6), PROTEST (rank 10), and ARGUE (rank among the top 20, verbs that express static
11). Proti, the Czech translation equivalent of meanings (including BE, LIVE, STAY, and STAND)
against, strongly collocates with verbs that express are more often produced by German and Czech
negative attitudes and evokes a sense of “reacting learners than by native speakers. Several of the
against” an opponent or enemy. Compared to the motion verbs produced by native speakers (GO,
group of native speakers, fewer learners in the WALK, COME) have the same or similar frequencies
Czech group associate verbs such as LEAN, PUSH, or in the German and (though to a lesser extent)
SIT with this VAC frame. These verbs are also Czech lists. Other motion verbs produced by
infrequent in the L1 Spanish verb list. Between native speakers (SLIDE, BLOW, DRAW, JUMP, SWIM) are
zero and four Spanish learners produce verbs of absent from or less common in the German and
physical contact or collision when they are Czech learner responses. Again we observe that
presented with a ‘V against n’ frame. The two learners produce verbs that have high frequencies
top responses from this group are forms of the in usage and have stronger associations with verbs
verbs FIGHT and BE which together account for 71 that are common in general language use.
or 54.2% of all participant responses. Again, L1 The Spanish learner responses are very differ-
transfer may explain the strong association ent from both the native speaker and the
between FIGHT and against. In a large corpus of German/Czech learner responses, providing
Spanish, the Corpus del Español,7 LUCHAR (FIGHT) further evidence in support of our Hypothesis
was found to be by far the most frequent collocate 2. The scatterplot in Figure 3 already provided an
immediately to the left of contra (against). This illustration of this lack of overlap between Spanish
TABLE 4
‘V in n,’ Top 20 Verbs in Native Speaker and Learner Responses

1 BE 19 BE 27 BE 33 BE 53
2 SIT 15 SIT 11 LIVE 10 LIVE 9
3 JUMP 10 LIVE 8 STAND 7 STAY 8
4 WALK 8 GO 8 SIT 7 PLAY 4
5 GO 7 WALK 8 WAIT 6 SLEEP 3
6 LOOK 6 HIDE 5 WORK 5 HIDE 3
7 FALL 6 STAND 5 COME 5 COME 3
8 COME 4 LOOK 4 SLEEP 4 STAND 3
9 LIVE 4 SLEEP 3 STAY 4 PUT 3
10 SING 3 COME 3 PARTICIPATE 3 WORK 2
11 RUN 3 FALL 3 FALL 3 TRAVEL 2
12 STAND 3 STAY 3 LIE 3 ENTER 2
13 SWIM 3 WORK 2 LOOK 3 ARRIVE 2
14 HIDE 3 PARTICIPATE 2 GO 3 GET 2
15 SLEEP 3 STUDY 2 WALK 3 FILL 2
16 SLIDE 2 WAIT 2 HIDE 2 REMAIN 2
17 DRAW 2 BITE 2 SWIM 2 GO 2
18 LIE 2 SEARCH 2 PUT 2 EAT 2
19 READ 2 RUN 2 JUMP 2 INVOLVE 1
20 BLOW 2 JUMP 2 RELAX 1 STUDY 1
Note. Italicized verbs indicate overlap between a learner list and the native speaker list.
learner and native speaker verb responses. Over section. Apparently, Spanish learners find it
40% of Spanish survey participants (53 of 131) harder to produce target-like verbs than their
respond to the ‘V in n’ prompt with forms of the Czech and German peers, again providing
most frequent, semantically bleached verb BE. supportive evidence for Hypothesis 2. As in the
They share their preference for LIVE and STAY with case of ‘V in n,’ this is likely related to L1-specific
the German and Czech groups but largely avoid differences with respect to expressing path and
motion verbs. WALK, FALL, and JUMP are absent from manner of motion. Table 5 shows lemmatized lists
the Spanish list while COME and GO are rare. The of the 20 most frequent verbs produced by the
strong differences between native speaker and four groups of survey participants in response to
Spanish learner responses likely are a result of the the prompts ‘he _____ in the . . .’ and ‘it _____ in
typological differences between English and the . . . .’ The native speaker responses in the left-
Spanish that we discussed earlier. ‘V in n’ is one hand column (shaded gray) serve as a reference
of many VACs in our set in which a path of motion point for comparisons with the German, Czech,
is expressed by a satellite (here the preposition and Spanish learner responses. Verbs are itali-
in). The verb-framed language Spanish tends to cized in a learner list if they also appear in the
encode this path in the verb and the manner of native speaker list.
motion in an adjunct, so walk in is realized as entrar The higher correlation figures observed for
caminando (enter walking). Hence, it is not German and Czech learners are supported by a
surprising that our Spanish learners do not (or fairly high number of verbs that are shared across
very rarely) produce verbs such as WALK, GO, FALL, these two groups and the native speaker top 20
or JUMP in response to the ‘V in n’ prompt. lists (11 for each group). JUMP, FALL, BE, RUN, GO,
and LOOK are among the highest-ranking verbs in
all three lists. Like native speakers, Czech and
Zooming in on ‘V Over n’ German learners associate over with verbs of
directed motion. Compared to native speakers,
For ‘V over n’ we see a split in terms of Czech learners show a preference for FALL and RUN
correlation values between German (0.72) and (16 and 12 compared to 9 responses); German
Czech learners (0.76) on the one hand and learners more often respond with GO, BE, and
Spanish learners (0.48) on the other—similar to COME. This learner group also produces motion
the ‘V in n’ construction discussed in the previous verbs (WALK, SWIM) that do not occur in the native
TABLE 5
‘V Over n,’ Top 20 Verbs in Native Speaker and Learner Responses

1 JUMP 29 JUMP 22 JUMP 23 GET 13
2 FALL 10 FALL 13 FALL 17 COME 13
3 RUN 9 GO 13 RUN 12 BE 13
4 GO 9 BE 10 GO 7 TAKE 12
5 CLIMB 7 WALK 10 LOOK 6 GO 12
6 BE 6 COME 9 COME 5 LOOK 10
7 FLY 5 LOOK 9 BE 5 RUN 5
8 LOOK 5 RUN 7 CLIMB 5 FLY 4
9 HOP 4 BEND 4 WALK 4 CROSS 4
10 ROLL 4 ROLL 3 GET 3 FALL 4
11 DRIVE 3 FLY 2 ROLL 3 TURN 3
12 COME 3 SWIM 2 BEND 3 BEND 3
13 READ 3 LIE 2 CROSS 3 WALK 3
14 LEAP 3 SIT 2 SLEEP 2 LEAN 3
15 STEP 2 SLEEP 1 CALL 2 JUMP 2
16 CROSS 2 SHIN 1 SAIL 2 WORK 1
17 SLEEP 1 LEAVE 1 TURN 2 TRAVEL 1
18 CYCLE 1 JULPED 1 TAKE 2 OVER 1
19 AIR 1 LIVE 1 TRIP 2 MIND 1
20 PUSH 1 WRITE 1 THINK 2 DANCE 1
Note. Italicized verbs indicate overlap between a learner list and the native speaker list. The form julped (number
18 in the L1 German list) presumably resulted from a learner’s attempt to type the form jumped.
speaker response list, and for which the preposi- this is the most frequent verb in the native
tion across may be a better, more idiomatic fit speaker, German, and Czech learner lists (with 29,
(swim across instead swim over). The reason for this 22, and 23 instances, respectively). As mentioned
may be L1 transfer. German learners may find it earlier, jump over is not realized by a verb plus
difficult to distinguish between over and across preposition in Spanish (but it is in German and
because both share the same translation equiva- can be in Czech).8 Instead, Spanish uses the verb
lent: über. Neither learner group includes any of SALTAR, which encodes the path of motion. This
the more specific motion verbs in their responses may be why JUMP is so infrequent in the Spanish
that native speakers associate with this VAC (e.g., learner responses. The same applies to climb over
HOP, DRIVE, LEAP, STEP, PUSH, or CYCLE). These are which can be translated as SALTAR (escalando) or
verbs that are strongly associated with the TREPAR. This further supports our Hypotheses 2
construction in language use but of lower overall and 4 and confirms our assumption that the low
frequency and less accessible to German/Czech correlation of Spanish learner and native speaker
learners. responses to the ‘V over n’ frame may be due to
This is also true for the Spanish learners who issues of language typology that are related to
participated in the survey. None of the more different strategies of expressing motion.
specific motion verbs (HOP, DRIVE, etc., plus CLIMB)
appear in their list of top 20 verb responses. The CONCLUSION AND OUTLOOK
verb at the top of the Spanish frequency list is GET,
which does not appear at all in the native speaker This article set out to examine L2 language
or German responses and is infrequent in the learners’ knowledge of verb–argument construc-
Czech responses. This verb is followed by COME, BE, tions (VACs). In psycholinguistic experiments, we
and TAKE—all of which are much less common in gathered evidence on L1 German, L1 Czech, and
the native speaker list (TAKE does not occur at all). L1 Spanish advanced English learners’ mental
It appears that Spanish learners associate with over representations of 19 different VACs. A compari-
constructions not primarily the expression of a son of data from these experiments with data
directed motion but instead think of metaphori- collected from native English speakers perform-
cal uses such as get over and take over. Particularly ing the same task allowed us to determine how
striking in this context is the fact that only 2 similar or different learners’ verb–VAC associa-
Spanish participants responded with JUMP whereas tions are from those of native speakers. The
experiments enabled us to highlight verbs (or survey responses and avoided specific manner of
groups of verbs) that are more entrenched in motion verbs. Spanish learners also produced the
native speakers’ than in learners’ mental repre- highest numbers of non-target-like verbs in
sentations of particular VACs, and vice versa. response to VACs that encode a path of motion
With respect to our Research Question 1, we in the preposition (e.g., ‘V over n’ and ‘V against
found that, while there is some overlap between n’). Based on these observations, we believe that a
learners’ and native speakers’ mental representa- major factor that influences the level of target-like
tions of VACs (also see our discussion in Ellis form–meaning mapping of common English
et al., 2014b), there are also differences in the VACs is language typology or, more precisely,
associations of verbs and constructions. We the type of motion event conceptualization across
observed that all three learner groups rely more languages. Fewer manner of motion verbs are
on general, highly frequent verbs (e.g., BE, COME, produced by Spanish learners who in turn
DO) and produce lower numbers of specific, less struggle more with VACs that encode a path,
frequent verbs (e.g., SLIP, REACH, CRAWL) than because Spanish is a verb-framed language in
native speakers do. We also observed that, for which manner of motion verbs are less readily
certain VACs, learners’ semantic associations with available and in which the path of motion tends to
a VAC are different from native speakers. For be encoded in the verb. Given that patterns of
example, learners associate verbs that express a expressing motion are language type specific, the
reaction or argument (including FIGHT, ARGUE, and challenge for the language learner is to acquire
SPEAK) with ‘V against n’ while native speakers the respective patterns for each new language. As
associate verbs of physical contact or collision (e. our results indicate, this becomes harder when
g., LEAN, PUSH, and BUMP) with this VAC. the new/second language is typologically differ-
In response to Research Question 2, we found ent from the learner’s first language. Echoing
that, in their verb–VAC associations, our three observations previously made on learned atten-
groups of learners are not equally different from tion and SLA (Ellis & Sagarra, 2011), we can say
the native speaker group but that L1 German and that a learner’s L1 and the L1-tuned expectations
L1 Czech learners are closer to the native speaker that come with it bias her/his system and,
group than L1 Spanish learners. In an overview depending on how typologically similar or differ-
chapter of studies that provide empirical evidence ent the L1 and L2 are, make her/him more or less
for language typology effects on linguistic pro- open to internalizing structures in the L2. Further
duction, Cadierno (2008) asks “how do L2 addressing the crosslinguistic transfer issue (Hy-
learners with typologically different L1s and L2s pothesis 5), we also found evidence of verb–
acquire the characteristic meaning–form map- preposition combinations in the learner survey
pings of the L2? And how does the performance data that are likely the result of collocational
of this type of learner compare to learners whose transfer from the L1s of the learners. An example
L1 and L2 share the same typological patterns?” was German learners’ association of FIGHT, PRO-
(p. 258). We have addressed these questions with TEST, and VOTE with the ‘V against n’ constructions
reference to learners’ knowledge of English verb– —all verbs that strongly collocate with the
argument constructions. We found that, for most translation equivalent of against (gegen) in Ger-
of the 19 VACs we examined, the mappings of L1 man. Similar effects were observable in the Czech
German and L1 Czech learners (i.e., speakers of and Spanish learner survey results. Searches in
languages that share the same typological pattern corpora of the learners’ L1s helped us confirm
as English) are more target-like than those of L1 our assumptions.
Spanish learners (i.e., speakers of a language that We also considered language proficiency as a
is typologically different from English). While all potential factor that may have influenced our
three groups of learners have developed construc- results. Given that the majority of learners who
tional knowledge, the overlap with native speaker participated in our study were at the same
verb–VAC associations is generally greater for advanced level of proficiency (CEFR level C1),
German and Czech than for Spanish learners. we can disregard this as an influential factor. The
This brings us back to Research Question 3, small number of level B2 learners among the L1
which asked whether differences across L2 Spanish group (around 10 of the 131 partic-
learner groups could be explained on the basis ipants) is unlikely to have had a major effect on
of L1 transfer and/or language typology effects. the overall results. The Spanish learners who
We think the answer to this question is yes. participated in our study also reported longer
Spanish learners, more than German and Czech times of having had English instruction at school
learners, favored general motion verbs in their than German and Czech learners (an average of
12.68 years, compared to 10.04 years for German restructuring of form–meaning associations takes
and 11.37 for Czech). time, and it needs a lot of exposure to natural
We have been able to confirm all of the research language use—ideally in the form of “extensive
hypotheses formulated earlier: The mental VAC interaction in a variety of contexts with members
representations of German, Czech, and Spanish of the target language community” (Jarvis &
advanced learners of English differ in diverse ways Pavlenko, 2008, p. 152). We find some of the
from those of native speakers, indicating that recent research in applying Cognitive Linguistics
learners are biased by their L1s. In generative free to teaching English modals and prepositions
association tasks, learners whose L1 is satellite- particularly promising and inspiring (see e.g.,
framed (and hence typologically similar to En- Tyler, 2012; Tyler, Mueller, & Ho, 2011) and
glish) produce more verbs that correlate more believe that the teaching of VACs could benefit
closely with those produced by L1 English speakers from a similar approach.
than speakers whose L1 is verb-framed. Speakers of In our SLA research agenda, we need to include
a satellite-framed language produce more verbs related work on an even larger set of construc-
that express specific manners of motion in the tions, including speakers of additional L1 back-
verb generation tasks. Conversely, speakers of a grounds, and collecting larger and richer data
verb-framed language produce specific manner of sets. We have begun to gather responses to
motion verbs less frequently and instead respond additional VAC frames from native speakers and
with more general motion verbs such as GO, COME, German and Spanish learners. From the same
or MOVE. Lastly, the verb responses of all learner groups of learners, we have also begun to collect
groups show effects of collocational transfer from richer data in verbal production tasks that ask
the learners’ first languages. participants to generate as many verbs as they can
Our findings have implications for language think of in one minute (following the methodol-
teaching and for research in SLA. Second ogy suggested in Ellis et al., 2014a; see also
language instruction needs to acknowledge the Cadierno, 2010). It would also be interesting to
pervasiveness of constructions more than it collect data from learners at additional proficien-
currently does. With few exceptions, current cy levels. A concern here, however, would be that
EFL and ESL textbooks are still largely based on the type of task we used in our study may be too
models of language that suggest a strict separation difficult for beginning or intermediate learners
of lexis and grammar and fail to reflect the (L1 Spanish learners at CEFR level B1 who were
interconnectedness of the two (see e.g., Meunier given the survey as a test struggled with the gap-fill
& Gouverneur, 2007; Römer, 2005, 2007). We task and gave up after looking at the first few
suggest that materials focus more on typical prompts). Additional evidence on learner VAC
associations of lexical items and constructions knowledge could come from analyses of learner
and emphasize patterns in form–meaning rela- corpora which capture the output of learners of
tions. Constructions that are semantically related different L1s and at different proficiency levels.
(e.g., VACs expressing directed motion) could be We are currently mining subsets of written and
grouped and taught together. That way, as spoken corpora of advanced learner English, the
Littlemore (2011) points out, it may be possible International Corpus of Learner English (ICLE;
for learners to “use their existing knowledge of Granger et al., 2009) and the Louvain Interna-
constructions to infer the meanings of ones that tional Database of Spoken English Interlanguage
are new to them” (p. 171). At the same time, it (LINDSEI; Gilquin, De Cock, & Granger, 2010)
needs to be highlighted which meanings are most for VACs. Initial results of these learner corpus
typically construed by which construction and analyses are discussed in Römer, Roberseon,
what the most common lexical items are in each O’Donnell, & Ellis (2014). One thing that our
construction. It may also be necessary to make initial ICLE and LINDSEI explorations highlight
learners aware of differences between VACs in is that L1-specific subsets of these two learner
their L1 and the L2. Learners of L1s that are corpora provide robust token numbers for
typologically different from English (such as some VACs (e.g., ‘V in n’ and ‘V about n’) but
Spanish, covered in our study) may need addi- are too small to give us enough tokens of the
tional help with specific constructions for which majority of VACs in our sample to identify
their form–meaning mappings are less target-like. semantic patterns or even lead verbs. This calls
Our findings could help raise instructors’ and for larger corpora of learner production that are
materials writers’ awareness of learners’ most carefully differentiated and marked up with
entrenched verb–VAC associations and how they learner metadata (like ICLE and LINDSEI are).
differ from those of native speakers. Learners’ Longitudinal learner corpora that consist of
3
learner data at different proficiency levels and Lemmatization was carried out using the morphy
allow us to capture learners’ language develop- function in the WordNet dictionary implemented in the
ment would also be extremely valuable in this Natural Language Toolkit (NLTK). The function uses a
context. As convincingly pointed out by Byrnes series of suffix rules based on part-of-speech category (e.
g., V ¼ verb, N ¼ noun) and an exception (i.e., irregular
(2009) in a study of the emergent writing ability of
forms) lookup list (see http://wordnet.princeton.edu/
L2 German learners, the adoption of a “develop- man/morphy.7WN.html for details, last accessed 12
mental view has the potential of capturing the April 2014) and attempts to find the base form (lemma)
dynamic nature of language use, language in WordNet that matches the supplied form. If a match
development, and the language system” (p. 64). cannot be made the supplied form is returned. For
In this article, we have taken a snapshot of part example: running (V) > run; running (N) > running; ran
of the linguistic knowledge of three groups of (V) > run; ran (N) > ran. This matching strategy does
advanced English language learners. Our study not use frequency or probabilistic data to select base
has provided evidence for representations of a set forms from WordNet, yielding some unexpected results.
of verb–argument constructions in the minds of For instance, there is a verb lemma fell (‘cause to fall by
or as if by delivering a blow,’ e.g., ‘the woodcutter felled
these learners. It has highlighted which verbs
the tree’), which leads to these results: fell (V) > fell; felled
learners most strongly associate with these con- (V) > fell; falls (V) > fall; falling (V) > fall. In the verb
structions and how their associations differ from completion experiment fell was provided in response to
those of native speakers. We believe that this a number of frames such as ‘s/he/it ____ across the . . .,’
snapshot has helped us gain a better understand- and with among, against, between, for, of, over, through, and
ing of what speakers know about verbs in towards. Forms of fall (fall, falls, falling) were also
constructions and of the role that language provided with these frames. As a result, we end up
typology and language transfer play in this context. with two items in the initial frequency lists. We merged
instances of fall and fell and lay and lie into two lemmas
and also searched for other similar items in our lists that
may have homonyms in WordNet.
4
ACKNOWLEDGMENTS See Gries and Deshors (2014) and Gries and
Adelman (2014) for examples of using regression to
examine the relationships between NS and NNS usage
We would like to thank contacts at the following
of linguistic features. These articles argue for a more
universities who helped with survey participant recruit-
involved use of regression analysis and use multivariate
ment by distributing the survey link: University of
and multilevel approaches.
Cologne (Germany), University of Giessen (Germany), 5
The average numbers of verb types produced per
University of Hanover (Germany), University of Heidel-
VAC were 40.6 for L1 German, 34.0 for L1 Czech, and
berg (Germany), University of Oldenburg (Germany),
37.4 for L1 Spanish learners. Native speaker participants
University of Trier (Germany), Masaryk University
produced an average of 45.1 different verbs per VAC.
(Czech Republic), Charles University (Czech Republic), 6
Source: DWDS (Das digitale Wörterbuch der deutschen
University of Extremadura (Spain), University of
Sprache), a corpus-based dictionary of German. URL of
Granada (Spain), University of Jaen (Spain), University
the search: http://www.dwds.de/?qu¼gegen, last ac-
Jaume I of Castellon (Spain), University of Salamanca
cessed 12 April 2014. Selected collocate statistic: Mutual
(Spain), and University of Zaragoza (Spain). We are also
Information (MI).
grateful to Markéta Malá for her native speaker advice 7
http://www.corpusdelespanol.org/, last accessed
on Czech translation equivalents of “jump over,” to Petr
12 April 2014.
Sudický for his help with interpreting some of the L1 8
As mentioned in the section on language typology,
Czech learner responses, and to Paco Barrón Serrano
translation equivalents of jump over in Czech include
for his help with interpreting some of the L1 Spanish
skočit přes (‘jump over’), přeskočit přes (‘overjump over’),
learner responses.
and přeskočit (‘overjump’).
NOTES REFERENCES
1
http://www.qualtrics.com, last accessed 12 April Battig, W. R., & Montague, W. E. (1969). Category norms
2014. for verbal items in 56 categories: A replication and
2
We decided to exclude ‘V at n’ from further extension of the Connecticut category norms.
analyses because we were unable to get reliable results Journal of Experimental Psychology, 80, 1–46.
(in terms of search precision) in the corpus mining, Brown, A., & Gullberg, M. (2011). Bidirectional cross-
which constitutes a central component in the compara- linguistic influence in event conceptualization?
tive part of the knowledge study described in Ellis et al. Expressions of path among Japanese learners of
(2014a). English. Language and Cognition, 14, 79–94.
Bybee, J. L. (2006). From usage to grammar: The mind’s Ellis, N. C., O’Donnell, M. B., & Römer, U. (2013).
response to repetition. Language, 82, 711–733. Usage-based language: Investigating the latent
Bybee, J. L. (2010). Language, usage, and cognition. structures that underpin acquisition. Language
Cambridge: Cambridge University Press. Learning, 63 (Supplement 1), 25–51.
Byrnes, H. (2009). Emergent L2 German writing ability Ellis, N. C., O’Donnell, M. B., & Römer, U. (2014a). The
in a curricular context: A longitudinal study of processing of verb–argument constructions is
grammatical metaphor. Linguistics and Education, sensitive to form, function, frequency, contingen-
20, 50–66. cy, and prototypicality. Cognitive Linguistics, 25, 55–
Cadierno, T. (2008). Learning to talk about motion in a 98.
foreign language. In P. Robinson & N. C. Ellis Ellis, N. C., O’Donnell, M. B., & Römer, U. (2014b).
(Eds.), Handbook of cognitive linguistics and second Second language processing of verb–argument
language acquisition (pp. 239–275). London: constructions is sensitive to form, function,
Routledge. frequency, contingency, and prototypicality. Lin-
Cadierno, T. (2010). Motion in Danish as a second guistic Approaches to Bilingualism.
language: Does the learner’s L1 make a differ- Ellis, N. C., & Sagarra, N. (2011). Learned attention
ence? In Z.-H. Han & T. Cadierno (Eds.), in adult language acquisition: A replication
Linguistic relativity in second language acquisition: and generalization study and meta-analysis.
Thinking for speaking (pp. 1–33). Clevedon, UK: Studies in Second Language Acquisition, 33, 589–
Multilingual Matters. 624.
Cadierno, T. (2013). Thinking for speaking in second Francis, G., Hunston, S., & Manning, E. (Eds.). (1996).
language acquisition. In C. A. Chapelle (Ed.), The Collins COBUILD grammar patterns 1: Verbs. Lon-
encyclopedia of applied linguistics. Malden, MA: don: Harper Collins.
Wiley–Blackwell. Accessed 12 April 2014 at Gehrke, B. (2008). Ps in motion. On the semantics and
http://onlinelibrary.wiley.com/doi/10.1002/ syntax of P elements and motion events. Utrecht, the
9781405198431.wbeal1213/full Netherlands: LOT Publications. Accessed 12 April
Cifuentes–Férez, P., & Gentner, D. (2006). Naming 2014 at http://www.lotpublications.nl/publish/
motion events in Spanish and English. Cognitive issues/Gehrke/index.html
Linguistics, 17, 443–462. Gilquin, G., De Cock, S., Granger, S (Eds.). (2010).
Collins, L., & Ellis, N. C. (2009). Input and second LINDSEI: Louvain International Database of Spoken
language construction learning: Frequency, form, English Interlanguage. Louvain-la-Neuve, Belgium:
and function. Modern Language Journal, 93, 329– Presses Universitaires de Louvain.
335. Goldberg, A. E. (1995). Constructions: A construction
Council of Europe. (2001). Common European framework grammar approach to argument structure. Chicago, IL:
of reference for languages: Learning, teaching, assess- University of Chicago Press.
ment. Cambridge: Cambridge University Press. Goldberg, A. E. (2003). Constructions: A new theoreti-
Accessed 16 April 2014 at http://www.coe.int/t/ cal approach to language. Trends in Cognitive
dg4/linguistic/Source/Framework_EN.pdf Science, 7, 219–224.
Ellis, N. C. (1998). Emergentism, connectionism and Goldberg, A. E. (2006). Constructions at work: The nature of
language learning. Language Learning, 48, 631– generalization in language. Oxford: Oxford Univer-
664. sity Press.
Ellis, N. C. (2002). Frequency effects in language Goldberg, A. E., Casenhiser, D. M., & Sethuraman, N.
processing: A review with implications for theories (2004). Learning argument structure general-
of implicit and explicit language acquisition. izations. Cognitive Linguistics, 15, 289–316.
Studies in Second Language Acquisition, 24, 143–188. Granger, S., Dagneaux, E., Meunier, F., & Paquot, M.
Ellis, N. C. (2003). Constructions, chunking, and (Eds.). (2009). ICLE: International Corpus of Learner
connectionism: The emergence of second lan- English. Louvain-la-Neuve, Belgium: Presses Uni-
guage structure. In C. Doughty & M. H. Long versitaires de Louvain.
(Eds.), Handbook of second language acquisition (pp. Gries, S. T., & Adelman, A. S. (2014). Subject realization
33–68). Oxford, UK: Blackwell. in Japanese conversation by native and nonnative
Ellis, N. C. (2006). Selective attention and transfer speakers: Exemplifying a new paradigm for
phenomena in SLA: Contingency, cue competi- learner corpus research. In J. Romero–Trillo
tion, salience, interference, overshadowing, block- (Ed.), Yearbook of corpus linguistics and pragmatics.
ing, and perceptual learning. Applied Linguistics, New York/Berlin: Springer.
27, 164–194. Gries, S. T., & Deshors, S. (2014). Using regressions to
Ellis, N. C., & Cadierno, T. (2009). Constructing a explore deviations between corpus data and a
second language. Annual Review of Cognitive standard/target: Two suggestions. Corpora, 9, 109–
Linguistics, 7, Special section, 111–290. 136.
Ellis, N. C., & Ferreira–Junior, F. (2009). Constructions Hunston, S., & Francis, G. (2000). Pattern grammar:
and their acquisition: Islands and the distinc- A corpus driven approach to the lexical grammar
tiveness of their occupancy. Annual Review of of English. Philadelphia/Amsterdam: John
Cognitive Linguistics, 7, 111–139. Benjamins.
Ibbotson, P. (2013). The role of semantics, pre-emption grammar and discourse: In honour of Susan Hunston.
and skew in linguistic distributions: The case of the Philadelphia/Amsterdam: John Benjamins.
un-construction. Frontiers in Psychology, 4, 989. Römer, U., Roberson, A., O’Donnell, M. B., & Ellis, N. C.
Accessed 21 April 2014 at http://www.ncbi.nlm. (2014). Linking learner corpus and experimental
nih.gov/pmc/articles/PMC3872292/ data in studying second language learners’
Jarvis, S. (2013). Crosslinguistic influence and multilin- knowledge of verb–argument constructions.
gualism. In C. A. Chapelle (Ed.), The encyclopedia of ICAME Journal, 38, 115–135.
applied linguistics. Malden, MA: Wiley–Blackwell. Rosch, E., & Mervis, C. B. (1975). Cognitive representa-
Accessed 12 April 2014 at http://onlinelibrary. tions of semantic categories. Journal of Experimental
wiley.com/doi/10.1002/9781405198431. Psychology, 104, 192–233.
wbeal0291/full Sinclair, J. M. (1991). Corpus, concordance, collocation.
Jarvis, S., & Pavlenko, A. (2008). Crosslinguistic influence Oxford: Oxford University Press.
in language and cognition. London: Routledge. Sinclair, J. M. (2004). Trust the text. Language, corpus and
Littlemore, J. (2011). Applying cognitive linguistics to discourse. London: Routledge.
second language learning and teaching. Basingstoke, Slobin, D. I. (2003). Language and thought online:
UK: Palgrave Macmillan. Cognitive consequences of linguistic relativity. In
MacWhinney, B. (2001). Emergentist approaches to D. Gentner & S. Goldin–Meadow (Eds.), Language
language. In J. Bybee & P. Hopper (Eds.), in mind: Advances in the study of language and thought
Frequency and the emergence of linguistic structure (pp. 157–192). Cambridge, MA: MIT Press.
(pp. 449–470). Philadelphia/Amsterdam: John Slobin, D. I. (2006). What makes manner of motion
Benjamins. salient? Explorations in linguistic typology, dis-
Mani, I., & Pustejovski, J. (2012). Interpreting motion: course, and cognition. In M. Hickmann & S.
Grounded representations for spatial language. Oxford: Robert (Eds.), Space in languages: Linguistic systems
Oxford University Press. and cognitive categories (pp. 59–81). Philadelphia/
Meunier, F., & Gouverneur, C. (2007). The treatment of Amsterdam: John Benjamins.
phraseology in ELT textbooks. In E. Hidalgo, L. Stubbs, M. (2001). Words and phrases. Corpus studies of
Quereda, & J. Santana (Eds.), Corpora in the foreign lexical semantics. Oxford, UK: Blackwell.
language classroom (pp. 119–140). Amsterdam: Talmy, L. (1985). Lexicalization patterns: Semantic
Rodopi. structure in lexical form. In T. Shopen (Ed.),
Odlin, T. (2013). Crosslinguistic influence in second Language typology and syntactic description: Grammat-
language acquisition. In C. A. Chapelle (Ed.), The ical categories and the lexicon (pp. 57–149). Cam-
encyclopedia of applied linguistics. Malden, MA: bridge: Cambridge University Press.
Wiley–Blackwell. Accessed 12 April 2014 at Talmy, L. (1991). Path to realization: A typology of event
http://onlinelibrary.wiley.com/doi/10.1002/ conflation. In L. A. Sutton, C. Johnson, & R.
9781405198431.wbeal0292/full Shields (Eds.), Proceedings of the seventeenth annual
R Development Core Team. (2012). R: A language and meeting of the Berkeley linguistics society (pp. 480–
environment for statistical computing. Vienna, 520). Berkeley, CA: Berkeley Linguistics Society.
Austria: R Foundation for Statistical Computing. Talmy, L. (2000). Towards a cognitive semantics, vol. 2:
Accessed 12 April 2014 at http://www.R-project. Typology and process in concept structuring. Cam-
org/ bridge, MA: MIT Press.
Robinson, P., & Ellis, N. C. (Eds.). (2008). Handbook of Tomasello, M. (2003). Constructing a language: A usage-
cognitive linguistics and second language acquisition. based theory of language acquisition. Boston: Harvard
London: Routledge. University Press.
Römer, U. (2005). Progressives, patterns, pedagogy. A Trousdale, G., & Hoffmann, T. (Eds.). (2013). Oxford
corpus-driven approach to English progressive forms, handbook of construction grammar. Oxford: Oxford
functions, contexts and didactics. Philadelphia/Am- University Press.
sterdam: John Benjamins. Tyler, A. (2012). Cognitive linguistics and second language
Römer, U. (2007). Learner language and the norms in learning: Theoretical basis and experimental evidence.
native corpora and EFL teaching materials: A case New York: Routledge.
study of English conditionals. In S. Volk–Birke & J. Tyler, A., Mueller, C., & Ho, V. (2011). Applying
Lippert (Eds.), Anglistentag 2006 Halle. Proceedings cognitive linguistics to learning the semantics of
(pp. 355–363). Trier, Germany: Wissenschaft- English prepositions to, for, and at: An experimen-
licher Verlag Trier. tal investigation. VIGO International Journal of
Römer, U. (2009). The inseparability of lexis and Applied Linguistics, 8, 180–205.
grammar: Corpus linguistic perspectives. Annual Yamashita, J., & Jiang, N. (2010). L1 influence on the
Review of Cognitive Linguistics, 7, 140–162. acquisition of L2 collocations: Japanese ESL users
Römer, U., O’Donnell, M. B., & Ellis, N. C. (2015). Using and EFL learners acquiring English collocations.
COBUILD grammar patterns for a large-scale TESOL Quarterly, 44, 647–668.
analysis of verb–argument constructions: Explor- Zipf, G. K. (1935). The psycho-biology of language: An
ing corpus data and speaker knowledge. In N. introduction to dynamic philology. Cambridge, MA:
Groom, M. Charles, & S. John (Eds.), Corpora, The M.I.T. Press.
APPENDIX A
Comparison of Learner Against Native Speaker Responses (n ¼ 131)
L1 German L1 Czech L1 Spanish

Ute Römer et al.
Verb Standard # of # of NS Verb Standard # of # of NS Verb Standard # of # of NS

VAC lemma residual responses responses lemma residual responses responses lemma residual responses responses
V about n tell 3.45 4 0 speak 3.95 18 2 speak 3.37 10 2
talk 3.38 42 22 go 2.38 4 1 talk 2.91 44 22
think 2.04 30 28 worry 2.65 4 1
lie 2.02 1 3 tell 2.24 3 0
run 2.02 1 3
hear 3.40 1 6
Correlation: 0.81 Correlation: 0.78 Correlation: 0.75
V across n lie 4.28 6 0 lie 4.52 12 0 come 3.01 35 14
come 2.59 35 14 go 2.12 22 11 go 2.92 28 11
walk 2.40 8 3 come 2.10 26 14 pass 2.50 3 0
skip 2.43 0 3 swim 3.69 1 11 skip 2.03 0 3
jump 2.55 0 4
roll 3.31 0 6
V after n look 4.24 34 4 look 4.26 47 4 look 4.67 58 4
follow 3.58 0 9 take 3.03 7 0 arrive 2.39 5 0
be 2.03 11 4 come 2.29 12 4
follow 2.86 0 9 follow 2.51 0 9
V against n be 2.57 29 13 speak 2.94 12 0 fight 4.26 40 6
argue 2.25 5 0 fight 2.67 27 6 be 3.38 31 13
vote 2.15 5 1 stand 2.51 9 1 play 2.40 5 0
push 2.66 1 13 be 2.10 25 13 stand 2.32 5 1
vote 2.00 6 1
fall 2.21 0 12
V among n be 4.12 58 31 belong 4.32 30 0 be 5.41 74 31
come 3.13 11 3 be 3.57 35 31 appear 2.68 5 0
stand 2.98 8 2 stand 2.54 10 2 go 2.63 8 3
lie 2.22 9 5 rank 2.12 6 0 happen 2.30 4 0
971
(Continued)
APPENDIX A (Continued) 972

roll 2.07 3 0 stay 2.30 4 0
hide 2.11 1 6 sit 2.05 0 11
fall 2.47 1 8
V around n move 3.67 6 1 turn 2.94 13 4 be 2.58 28 14
stand 3.49 5 0 move 2.74 4 1 come 2.38 17 8
walk 2.14 10 4 go 2.37 36 24 go 2.28 35 24
revolve 2.22 0 3 wind 2.33 3 0 lie 2.26 3 0
come 2.31 16 8 move 2.11 3 1
roll 2.63 1 5 circle 2.52 0 5
roll 2.52 0 5
V as n look 3.19 10 4 look 4.14 27 4 work 3.82 16 3
do 2.95 4 1 work 3.66 18 3 look 3.40 14 4
smile 2.92 6 2 smile 2.42 7 2 seem 2.99 6 0
cry 2.74 7 3 speak 2.42 7 2 act 2.91 6 1
count 2.43 3 0 appear 2.09 4 0 be 2.54 11 8
play 2.29 3 1 behave 2.09 4 0 behave 2.29 4 0
go 2.06 2 11 go 2.03 1 11 go 2.09 1 11
sing 2.38 0 4
V between n stand 4.06 21 2 stand 3.82 22 2 be 5.01 66 26
lie 2.87 26 11 stay 2.34 5 1 place 2.45 4 0
be 2.63 35 26 lie 2.19 24 11 lie 2.33 12 11
stick 2.59 5 0 fit 2.00 4 1 slip 2.14 0 8
slip 2.34 0 8 slip 2.64 0 8 fall 2.83 1 18
fall 3.16 1 18
V for n come 2.83 4 0 vote 3.23 9 2 look 3.58 32 11
search 2.83 4 0 care 2.94 4 0 wait 2.68 14 6
apply 2.24 3 0 look 2.42 24 11 apply 2.32 3 0
vote 2.08 5 2 search 2.33 3 0 come 2.32 3 0
The Modern Language Journal 98 (2014)
(Continued)
APPENDIX A (Continued)

Ute Römer et al.
ask 2.04 12 7 hop 2.15 3 1 stand 2.17 3 1
reach 3.21 1 7 stand 2.15 3 1 ask 2.13 12 7
reach 3.31 0 7
V in n stay 2.50 3 0 wait 3.26 6 0 be 4.91 53 19
be 2.32 27 19 be 2.85 33 19 stay 3.02 8 0
live 2.13 8 4 work 2.79 5 1 live 2.53 9 4
swim 2.08 1 3 stay 2.50 4 0
sing 2.31 0 3 live 2.37 10 4
jump 2.77 2 10 stand 2.05 7 3
V into n go 3.10 26 9 come 3.09 11 3 go 2.80 33 9
crash 2.81 3 0 be 2.57 6 2 get 2.62 10 2
look 2.63 13 5 get 2.13 5 2 come 2.56 13 3
walk 2.57 17 7 look 2.11 12 5 put 2.50 5 0
bump 2.51 1 3 run 2.87 13 36 look 2.32 17 5
come 2.51 1 3 jump 2.70 1 9
V like n behave 3.18 5 0 look 3.59 60 21 look 4.17 56 21
be 2.63 26 14 do not 2.17 6 2 be 2.93 25 14
speak 2.17 3 0 run 2.57 2 17 behave 2.32 3 0
cry 2.02 3 1 eat 2.03 0 4
talk 2.02 5 2 swim 2.03 0 4
swim 2.23 1 4 run 4.14 1 17
sing 2.58 1 5
V of n jump 4.28 9 1 fall 2.99 10 2 think 3.46 28 22
fall 3.28 9 2 come 2.96 6 1 dream 3.38 5 1
think 2.35 28 22 get 2.35 4 0 go 3.01 4 0
go 2.18 3 0 talk 2.67 1 6 one 2.36 3 0
973
(Continued)
APPENDIX A (Continued) 974

say 2.01 0 3 put 2.36 3 0
smell 2.25 1 4 smell 2.15 1 4
V off n take 2.70 16 6 put 3.20 6 0 switch 3.13 10 0
turn 2.55 6 2 fall 2.80 47 32 turn 2.98 15 2
walk 2.34 3 1 take 2.76 17 6 take 2.74 24 6
fly 3.12 0 4 get 2.43 9 3 put 2.63 7 0
fly 2.06 0 4 get 2.54 14 3
be 2.41 1 8 jump 3.39 0 21
V over n walk 3.70 10 1 walk 2.46 4 1 get 3.42 13 1
bend 2.18 4 1 fall 2.01 17 10 take 3.31 12 1
come 2.15 9 3 leap 2.05 0 3 come 2.65 13 3
hop 2.04 0 4 read 2.05 0 3 be 2.20 13 6
climb 2.64 1 7 hop 2.49 0 4
fly 2.64 1 5
V through n walk 4.244 26 8 get 3.00 6 0 go 2.97 43 20
climb 2.82 3 1 see 3.00 6 0 pass 2.93 11 2
swim 2.82 3 1 go 2.72 41 20 get 2.52 5 0
look 2.40 14 9 look 2.05 17 9 see 2.52 5 0
move 2.43 0 3 saw 2.05 6 2 be 2.41 5 1
think 3.93 0 6 think 2.26 1 6 travel 2.06 4 1
run 3.77 1 17 think 2.24 0 6
fall 2.56 1 9
V towards n come 2.95 6 2 come 2.78 11 2 be 3.83 8 0
fall 4.25 1 6 turn 2.46 5 1 go 2.64 37 14
head 2.10 4 1 come 2.38 7 2
fall 2.91 0 6 fall 3.09 0 6
The Modern Language Journal 98 (2014)
(Continued)
APPENDIX A (Continued)

Ute Römer et al.
V under N lie 3.66 31 4 sleep 3.67 15 2 be 3.72 37 9
stand 3.17 8 1 work 2.48 4 0 stay 2.47 4 0
live 2.09 4 1 lie 2.17 11 4 live 2.34 4 1
roll 2.26 0 4 sleep 2.20 6 2
roll 2.05 0 4
crawl 2.16 1 5
V with N play 3.44 15 3 agree 2.68 6 1 be 3.13 23 6
be 2.65 16 6 sleep 2.49 14 5 stay 3.08 8 1
stay 2.51 4 1 stay 2.39 5 1 deal 2.73 6 0
eat 3.57 1 11 live 2.24 7 2 agree 2.35 5 1
run 2.04 1 6 come 2.09 14 8
dance 2.21 1 7 play 2.07 8 3
eat 2.90 0 11 run 2.03 0 6
walk 2.33 0 8
Note. To facilitate data interpretation, verbs with negative standardized residuals below 2 are shaded gray.
975

Modl 12149

Uploaded by

Copyright:

Available Formats

Modl 12149

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Modl 12149

Uploaded by

Copyright:

Available Formats

Second Language Learner

MATTHEW BROOK O’DONNELL

Selected VACs Survey Prompts

VAC L1 German L1 Czech L1 Spanish

Rank Native Speakers German Learners Czech Learners Spanish Learners

Rank Native Speakers German Learners Czech Learners Spanish Learners

Rank Native Speakers German Learners Czech Learners Spanish Learners

L1 German L1 Czech L1 Spanish

Verb Standard # of # of NS Verb Standard # of # of NS Verb Standard # of # of NS

L1 German L1 Czech L1 Spanish

L1 German L1 Czech L1 Spanish

L1 German L1 Czech L1 Spanish

L1 German L1 Czech L1 Spanish

You might also like