-
DOI: 10.1007/s10936-005-6139-3Journal of Psycholinguistic
Research, Vol. 34, No. 4, July 2005 ( 2005)
Syntactic Priming: A Corpus-based Approach
Stefan Th. Gries1
The present study is a corpus-based investigation of syntactic
priming, i.e. the tendency to reusesyntactic constructions. On the
basis of data from the ICE-GB corpus, I analyze two differentpairs
of syntactic patterns, the so-called dative alternation and
particle placement of transitivephrasal verbs. Although it has
sometimes been argued that only experimental data can contrib-ute
to studies of priming, the analysis shows that (a) the corpus-based
results for datives arevery similar to the experimental ones; (b)
priming is also obtained for the verb-particle con-struction, a
construction hitherto not explored in the priming literature and
(c), most impor-tantly, in line with much previous psycholinguistic
and corpus-linguistic work, priming effectsturn out to be strongly
verb-specic such that some verbs are much more resistant or
responsiveto priming than others. I conclude with a discussion of
how corpus data relate to experimentaldata and how the corpus-based
ndings can contribute to psycholinguistic model building.
KEY WORDS: collostructions; corpus data; verb subcategorization;
verb bias; structural/syntactic priming.
INTRODUCTION
As a variety of studies has shown, speakers tend to repeat
syntactic struc-tures they have just encountered (produced or
comprehended) before. Thistendency has been referred to as
structural priming, syntactic persistenceor syntactic priming; I
will use the latter term throughout the remainder
1 This paper is dedicated to Gunter Rohdenburg on the occasion
of his 65th birth-day. Max Planck Institute for Evolutionary
Anthropology, Leipzig, Germany. email:[email protected],
http://people.freenet.de/Stefan Th GriesI thank (in alphabetical
order) Doris Schonefeld (Ruhr University of Bochum), AnjaSteinlen
(University of Southern Denmark) and Stefanie Wulff (University of
Bremen)for comments on a previous draft of this paper. In addition,
I have beneted a lot fromdiscussion with Benedikt Szmrecsanyi
(University of Freiburg) and from the very detailedand useful
comments of one anonymous reviewer, which made it possible to
improve thepaper considerably. Naturally, all remaining
inadequacies are my own.
365
0090-6905/05/07000365/0 2005 Springer Science+Business Media,
Inc.
-
366 Stefan Th. Gries
of this paper.2 Levelt and Kelter (1982) and Weiner and Labov
(1983)belong to the earliest studies to which the identication of
this phenom-enon is commonly attributed. The former study found
that merchants inthe Netherlands tended to formulate their answers
to questions such thatthe syntactic structure of their answers was
identical to that of the ques-tions; for example, the Dutch
equivalents of the questions in (1a) and (1b)tended to trigger (2a)
and (2b) respectively as answers.
(1) a. At what time does your shop close?b. What time does your
shop close?
(2) a. At ve oclock.b. Five oclock.
The latter study found that the likelihood of a passive
utterance ata particular point of time in a sociolinguistic
interview is signicantlyincreased by the presence of another
passive utterance in the previousve sentences. Similar ndings are
presented in a corpus-based study onactives, lexical passives and
transformational passives by Estival (1985),who also found a
signicant priming effect of actives and passive acrossve clauses
which was robust enough to remain even after a varietyof
potentially interfering discourse-functional variables had been
factoredout.
The bulk of the studies on syntactic priming, however, consists
ofexperimental approaches. The classic study in this respect is the
picturedescription study of Bock (1986). Under the guise of a
memory task, sub-jects rst repeated prime sentences coming in one
out of two alternatingstructures: the transitivity alternation
(i.e., active vs. passive sentence form)or the so-called dative
alternation (i.e., the syntactic choice between the di-transitive,
or double-object, construction and the prepositional dative withto
and for). Then, the subjects described semantically unrelated
picturesallowing both syntactic alternatives from one of the two
alternating struc-tures. Bock found that subjects indeed preferred
to formulate a descriptionthe syntactic structure of which matched
that of the prime sentence.
Other studies (especially Bock & Loebell, 1990, but also,
e.g., Smith& Wheeldon 2001) ruled out other attempts to explain
syntactic priming.For instance, syntactic priming cannot be
attributed to metrical similaritybetween primes and target choices
by subjects: (3a) and (3b), instantiating
2 I adopt the denition of syntactic priming proposed by Branigan
et al. (1995: 490): Wedene syntactic priming as the proposal that
processing a particular syntactic structurewithin a sentence
affects the processing of the same (or a related) syntactic
structure withina subsequently presented sentence; cf. Szmrecsanyi
(2005) for discussion of whether persis-tence is the more
appropriate cover term (specially for corpus-based studies).
-
Syntactic Priming: A Corpus-based Approach 367
the dative alternation just referred to, are identical in terms
of their met-rical structure but differ in terms of their different
syntactic structures(adopted from Bock & Loebell, 1990:24),
which is why (3a) results in asignicant preference of prepositional
datives (relative to a baseline/controlcondition) whereas (3b) does
not.
(3) a. Susan [VP brought [NP a book] [PP to Stella]]b. Susan [VP
brought [NP a book] [S to study]]
Similarly, syntactic priming does not derive from the presence
ofclosed-class lexical items in particular slots of the sentences
or event-struc-tural or thematic utterance characteristics. As to
the former, Bock (1989),for example, provides experimental evidence
showing that the primingeffects she obtained cannot be explained by
reference to closed-class lex-ical items involved in the dative
alternation, viz. to and for. As to the lat-ter claim, consider the
examples in (4) and (5). While (4a) and (4b) aredifferent in terms
of their thematic structure, they are identical in terms oftheir
syntactic structure and, thus, both (4a) and (4b) prime
prepositionaldatives (relative to control conditions); the same
holds for (5a) and (5b),both of which prime passives.
(4) a. The wealthy widow [VP gave [NP an old Mercedes]
[NPrecipientto the church]]
b. The wealthy widow [VP drove [NP an old Mercedes]
[NPlocativeto the church]]
(5) a. The 747 was alerted [PP agentive by the airports control
tower]b. The 747 was landing [PP locative by the airports control
tower]
Subsequent experimental work has mainly focused on spoken
English,but has also been concerned with written English as well as
Dutch(cf. Hartsuiker & Kolk, 1998; Hartsuiker, et al., 1999;
Hartsuiker &Westenberg, 2000) and German (cf. Scheepers &
Corley, 2000) in bothspeaking and writing. The range of
experimental methodologies hasalso been broadened considerably and
now includes a wide variety ofofine experimental paradigms such as
sentence completion tasks (cf., e.g.,Pickering & Branigan,
1998; Hartsuiker & Westenberg, 2000 etc.), sentencerecall tasks
(Potter & Lombardi, 1998), and picture descriptions in dia-logs
(cf. Branigan et al., 2000). In addition, Smith & Wheeldon
(2001) andCorley & Scheepers (2002) did online studies where
priming effects werealso measured in terms of production latencies.
While most studies haveinvestigated the dative alternation and the
activepassive alternation inEnglish (as in (4) and (5)), more
recent work has also looked at the equiv-alent constructions in
Dutch as well as Dutch locative PP alternations, the
-
368 Stefan Th. Gries
order of auxiliary verb and past participle in Dutch as well as
dative-accu-sative verb alternation in German (cf. Scheepers &
Corley, 2000), and theorder of syntactic functions in Japanese (cf.
Yamashita et al. [2002]).
The currently most pressing issues concerning syntactic priming
(manyof which will also be addressed in the present approach) are
the following:
(i) the duration of syntactic priming: on the one hand, Levelt
andKelter (1982) and Branigan et al. (1999) report that priming
(inspoken and written production respectively) is fairly
short-lived.On the other hand, other studies report priming effects
acrosslonger time interval or more intervening material (cf. Bock
&Grifn, 2000; Pickering et al., 2000; Chang et al., 2000).
(ii) the directionality of syntactic priming: Branigan et al.
(1995)discuss a variety of different studies which, taken all
together,support the assumption that syntactic priming can operate
fromproduction to production (cf. Bock, 1986; Bock & Loebell,
1990),from comprehension to comprehension (cf. Branigan et al.,1995
for an overview) and from comprehension to production(cf. Branigan
et al., 2000; Bock, 2002).
(iii) the grammatical characteristics of the priming verb:
Pickering andBranigan (1998) found that (a) syntactic priming is
stronger ifthe priming verb lemma and the target verb lemma are
identical(compared to different lemmas in prime and target) and
that (b)morphological differences between the priming verb and the
tar-get verb (in terms of tense, aspect and number) do not result
instrongly varying priming strengths.
(iv) the degree to which syntactic priming is asymmetric and
con-struction-(pair)-specic: From a between-alternations
perspective,Bock (1986: Exp. 1) found stronger priming for the two
syntacticframes involved in the dative alternation than for those
involvedin the activepassive alternation in English; a similar
prominenceof datives over transitives was found for English by
Potter andLombardi (1998: Exp. 3) and for Dutch by Hartsuiker and
Kolk(1998). In addition, from a within-alternation perspective,
furtherasymmetries were sometimes obtained: Bock (1986: Exp. 1)
foundthere was stronger priming for ditransitives than for
prepositionaldatives while Potter and Lombardi (1998) report the
opposite(and Pickering et al., 2002: 587 mention evidence for
symmet-ric/balanced priming).
(v) the degree to which syntactic priming is
language-specic:Hartsuiker et al., (2002) demonstrate syntactic
priming fromcomprehending Spanish to producing English, Salamoura
(2002:
-
Syntactic Priming: A Corpus-based Approach 369
Exp. 2) demonstrates priming from Greek (L1) structures
toEnglish (L2) structures, and Gries and Wulff (in press) show
thatGerman learners of English as a foreign language exhibit
primingin an English sentence completion task.
The present study is concerned with the issues raised in
(i)(iv). How-ever, it is different from most others in two
respects. First, its main pointis that it goes beyond previous
works by paying closer attention to therole individual verbs play
for priming. Recent studies demonstrated thatdifferent verbs
exhibit differentially strong associations to particular syn-tactic
patterns or, put differently, constructions. Although the
experimentalpriming studies mentioned above did control for the
frequencies of primeconstructions and for item-specic effects (in
terms of F1 and F2 statis-tics) and, thus, allowed for a clear
conrmation of syntactic priming, thereappear to be no studies at
all which investigated to what degree, if any, thestrength of
priming effects is conditioned by the prime and target verbs.The
main issue of this study is, therefore, the question of whether
par-ticular verbs are more responsive or resistant to priming as
target verbssuch that, across many different prime verbs, they
either have a tendencyfor a particular construction that overrides
the prime structure or not.More generally speaking, the present
study takes into account the degreeto which syntactic priming may
be verb-specic.
Second, contrary to most previous works, the present study is
basedon naturalistic corpus data rather than psycholinguistic
experimentation.Given that the rst studies reporting syntactic
repetition were based onnaturalistic data, it may appear somewhat
surprising that so little corpus-based work on priming has been
conducted, especially since larger cor-pora and the software
necessary for their analysis is so widely available.This absence
can probably partly be attributed to the fact that, althoughBocks
(1990) rst example for what she later refers to as syntactic
prim-ing is drawn from naturalistic conversation, priming
researchers such asBranigan and Pickering have argued against
corpus-based approaches topriming by stating that
there are several nonsyntactic factors which could lead to
repeti-tion. [ . . . ] Corpora have proved useful as a means of
hypothesisgeneration, but unequivocal demonstrations of syntactic
prim-ing effects can only come from controlled experiments
(Braniganet al., 1995: 492; cf. also Pickering & Branigan,
1999: 136).
In the general discussion and conclusion, I will discuss this
matter indetail. The overall plan of the present paper is as
follows: The next sectioninvestigates the dative alternation
already introduced above. I will rst be
-
370 Stefan Th. Gries
Table I. Ditransitives and all Prepositional Datives: Medium
ConstructionData (ICE-GB) V NP PPfor/to V NP/S NP/S Row totals
spoken 926 1254 2180written 854 759 1613Column totals 1780 2013
3793
concerned with general priming effects and their determinants
irrespectiveof the verbs guring in the constructions, and then I
will provide a morene-grained analysis of some verbs behavior. I
will then turn to the alter-nation known as particle placement. As
before, I will rst discuss verb-independent results before I turn
to verb-specic details. The last sectionwill conclude.
THE DATIVE ALTERNATION
General Investigation
In order to investigate syntactic priming corpus-linguistically,
I rst iden-tied all ditransitive constructions and all
prepositional datives with to and forin the British component of
the International Corpus of English (ICE-GB).The ICE-GB is a
POS-tagged and fully parsed corpus of spoken and writtenBritish
English of the 1990s; all annotation has been checked manually
byseveral linguists (cf.
http://www.ucl.ac.uk/english-usage/ice-gb/index.htm fordetails).
The distribution obtained is represented in Table I.
Out of these 3793 cases, 790 had to be discarded for the priming
ana-lysis because they were the rst or last construction either in
one of the500 corpus les or in a subtext of a corpus le, leaving
3003 prime-targetpairs (i.e. subsequent constructions of either
type) for the analysis. Eachof these was then coded for a variety
of variables:
Medium: the medium in which prime and target occurred: spokenvs.
written (automatically retrieved from the corpus les).
CPrime and CTarget: the constructions of the rst and the
secondof the two constructions constituting a prime-target pair:
ditransi-tive vs. prepositional dative (automatically retrieved
from the anno-tated parse trees within the corpus les).
CID: the fact whether the constructions in prime and target
areidentical: yes or no (this coding task was performed
semi-automat-ically with a spreadsheet software applied to the
output of the con-cordancing software).
-
Syntactic Priming: A Corpus-based Approach 371
Distance: the distance in parsing units between the occurrence
ofprime and target within each subtext of each le as determinedfrom
the annotation of the corpus: 0, 1, 2, 3, 4, 5, 6, 7,8, 9, 10, 15,
20, 25 and >25 (a parsing unit is the basicstructural unit of
each corpus le; in the majority of cases it corre-sponds to a
clause or sentence).
VFormPrime and VFormTarget as well as VLemmaPrime
andVLemmaTarget: the exact verb form and the verb lemma ofeach
prime and target (the verb forms in both constructions
wereretrieved automatically from the corpus les, the lemmatization
wasdone manually by myself).
VFormID and VLemmaID: whether both constructions involved
thesame verb form and verb lemma: yes or no (this coding task
wasperformed semi-automatically with a spreadsheet software).3
SpeakerID: whether in the spoken data both constructions
wereproduced by the same speaker or not: yes or no (this codingtask
was performed semi-automatically with a spreadsheet softwareapplied
to the output of the concordancing software).
To provide one example to illustrate this coding process,
consider thefollowing brief extract from the ICE-GB (S1A-007 72:1
to 73:1).
(6) a. Speaker B: You gave it to herb. Speaker A: That just sh.
. . shows you the policy of keeping
things
Applying the above coding scheme results in the data set
representedin Table II.
On the basis of the analogous classication of all 3003
prime-targetpairs, it is now possible to compare the frequencies of
all sorts of cong-urations of these variables and their impact on
the switch rate of CPrimeand CTarget, i.e. whether CPrime and
CTarget are identical (cf. Sankoff& Laberge, 1978 for the rst
approach of this kind).4 Consider Table III
3 The variable VFormID is of course nested into VLemmaID since,
if the lemmas of CPrimeand CTarget are already different, the forms
cannot be identical anymore.
4 Benedikt Szmrecsanyi (p.c.) pointed out to me that collapsing
switch rates of differentspeakers or of different corpus les this
way may be dangerous: It is possible that the con-ation of, for
example, two corpus les in which no priming takes place may result
in asummary table in which priming shows up as a statistical
artifact. He therefore recommendsusing scatterplots of the kind
used by Sankoff and Laberge (1978), in which for eachspeaker or le
the relative frequency of a construction on the x-axis is plotted
against theratio of switches to one construction on the y-axis; to
my mind, this is comparable to by-item statistics as used in
ANOVAs. It follows that only if most dots are located below the
-
372 Stefan Th. Gries
Table II. Application of the Coding Scheme to Priming from (6a)
to (6b)
Variable Value/level
Medium: spokenCPrime: prepositional dativeCTarget:
ditransitiveCID: noDistance: 1SpeakerID: noVFormPrime:
gaveVLemmaPrime: giveVFormTarget: showsVLemmaTarget showVFormID
noVLemmaID: no
Table III. CPrime CTarget: Observed vs. Expected Frequencies
(2(1) = 202.4, p < .001)
CTarget: CTarget: RowData (ICE-GB) V NP PPfor/to V NP/S NP/S
totals
CPrime: V NP PPfor/to 830 (647.1) 549 (731.9) 1379CPrime: V NP/S
NP/S 556 (762.1) 1068 (861.9) 1624Column totals 1386 1617 3003
for the most general result, namely the interaction of CPrime
and CTargetacross all other variables; the expected frequencies are
provided in paren-theses and are not computed on the basis of row
and column totals buton the basis of row totals and the overall
frequencies of the two construc-tions as listed in Table I.5
main diagonal, switches (from one construction to the other) are
rarer than the null hypoth-esis of the absence of priming would
predict. In order to show that the summary tablesused in the
present data set do not suffer from such an articial ination, I
also providecorresponding scatterplots; for that of the datives,
cf. Fig. (2)
5 The question may arise why the expected frequencies are not
computed the usual way.The reason for this is the following. If one
uses the column totals from Table III for thecomputation of the
expected frequencies, one treats these as given, as an independent
var-iable so to speak, while in the present design the column
totals are of course part of thedependent variable, namely the
frequency of one (target) construction as a response tosome (prime)
construction. The more appropriate logic underlying the present way
of com-putation is this: after each prime construction, the speaker
has two constructional choices,and the probabilities of each of the
two constructional choices are the frequencies with
-
Syntactic Priming: A Corpus-based Approach 373
As is obvious, there is a strong syntactic priming effect such
thatspeakers/writers prefer to use the primed syntactic structure.
It is alsoinstructive, however, to look at the strength of the
priming effect and tocompare it with that of previous experimental
results. In the present data,the ratios of the primed structure vs.
the non-primed structure are 1.5 and1.9 for prepositional datives
and ditransitives respectively. By comparison,in her classic study,
Bock (1986: 364) reports percentages instead of rawfrequencies
where the corresponding ratios of the percentages are 1.5 and2.1
for prepositional datives and ditransitives respectively; the
differencesbetween her ratios and mine are obviously negligible.
This also indicatesthat ditransitives prime more strongly than
prepositional datives.
While this is a rst promising result, it is more interesting to
look athow the above variables interact. To this end, the variables
Medium, CPrime,VFormID, VLemmaID and SpeakerID and Distance (as a
covariate) wereentered into a General Linear Model (GLM) analysis
with CTarget asthe dependent variable.6 While the overall
correlation between the aboveindependent variables and CTarget is
only moderate (adjusted R2 = .17,F(18, 2, 984) = 35.6, p <
.001), there are some effects worth mentioning.For reasons of
space, I will mainly discuss the signicant and marginallysignicant
results only; in the interest of readability, I only give
p-valueshere and provide all F -values and effect sizes in Table
A.I in Appendix A.
The priming effect already reported above in Table III is of
coursealso reected in the GLM analysis and CPrime is the strongest
predic-tor of the constructional choice in the target slot (p <
.0001). Whilethe effect of CPrime is independent of Medium (i.e.,
priming was obtainedequally strong in speaking and in writing as
one might have expected onthe basis of previous experimental work),
it does enter into noteworthyinteractions with the other variables,
namely VFormID, VLemmaID andSpeakerID. The signicant interactions
VFormID CPrime (p = .0354)as well as VLemmaID CPrime (p < .0001)
support Pickering and Bran-igans (1998: Exp. 1) results in that
they indicate that, when the verbform and/or the verb lemma are
identical across prime and target, thenpriming is considerably
stronger than if prime and target are different.In addition, there
is a very small and only marginally signicant effect(p = .0563)
such that if the speaker of the prime is the same as that of
which these constructions occur in the corpus (rather than .5 vs
.5). Thus, what is neededare the overall frequencies of the two
constructions in the corpus, which corresponds to thecolumn totals
of Table I.
6 In terms of interpretation, the results are identical to an
analogous analysis with Medium,CPrime, VFormID, VLemmaID and
SpeakerID and Distance as independent variables andCID as dependent
variable.
-
374 Stefan Th. Gries
the target, priming is slightly stronger. In other words,
production-to-production priming is stronger than
comprehension-to-production priming.Finally, let us turn to
Distance. Obviously, the distance between primeand target is
irrelevant to the constructional choice as operationalized
byCTarget. While this is not surprising, the analogous analysis
with CID asdependent variable also results in no signicant effect
(F(1, 2, 984) = .411,p = .521). Does this mean that Distance does
not have any inuence onthe strength of the priming effect (as
measured by the percentage of caseswhere CPrime equals CTarget)?
And if so, would this not invalidate thecorpus-based analysis
completely (since no effect of Distance would implythe relatively
unlikely situation that priming is equally likely across all
dis-tances)? A second, closer look at the data, however, shows that
this is notso because one has to bear in mind that Distance was
entered into a lin-ear model, but that the relation between the
distance between prime andtarget on the one hand and the strength
of the priming effect on the otherhand need not be linear. In fact,
there is evidence that this relation is loga-rithmic (cf. Gries (in
press) for empirical evidence on the basis of Pickeringand
Branigans [1998] conditional probabilities measure). Thus, while
thelinear relation between Distance and the strength of the priming
effect isnegligible, the logarithmic one is not (adjusted R2 = .77;
R1,13 = 44.08;P < .001).7 priming is in fact long-lasting (again
in accordance with recentndings by Bock and Grifn, 2000) and after
a decrease from Distance:0 toDistance:1, there was no consistent
decline in the magnitude of priming,although there were unstable
changes at particular lags (i.e., parse units inthe present study)
(Bock & Grifn, 2000: 187).
In sum, not only has the corpus-based analysis of syntactic
prim-ing revealed signicant priming effects for ditransitives and
prepositionaldatives, the results are also strikingly similar to
those of previous experi-mental studies in terms of strength of
effects, the inuence of morpholog-ical characteristics of the
verbs, construction-specicity, directionality anddistance effects
(i.e. the time course of priming). The following section willnow
provide a more detailed picture of how individual verbs gure in
thepriming effects.
Verb-specic Investigation
The above investigation has shown that the corpus-derived
results arequite similar in nature to those obtained
experimentally. However, as has
7 For this analysis, the Distance values of 0 and >25 were
recoded as 1E-06 and 30 respec-tively; other values for Distance:0
and Distance:>25 yielded identical results in terms ofexplained
variance; the resulting equation is (% of CID = 1) = .625
.0175ln(Distance).
-
Syntactic Priming: A Corpus-based Approach 375
Table IV. Exemplication of the Null Hypothesis H0
CTarget: CTarget: RowData (ICE-GB) V NP PPfor/to V NP/S NP/S
totals
CPrime: V NP PPfor/to a b a+bCPrime: V NP/S NP/S c d c+dColumn
totals a+c b+d a+b+c+d
already been argued above, the results discussed in the previous
section (andthe results of any other study on syntactic priming I
am aware of) did nottake into consideration the degree to which
priming effects might be sensitiveto particular verbs individual
preferences. This is all the more striking since(i) at least Potter
and Lombardi (1998: 278) mention, but do not investigate,that
individual verbs may affect preferences for particular syntactic
patternsand (ii) probabilistic (i.e. frequency-based) properties of
words, word sensesand words in particular constructions have proven
to be relevant to a varietyof linguistic and psycholinguistic
issues and models. Thus, what is necessaryis a rst exploratory
study of this issue. Such an exploratory study of thisissue using
experiments would be quite an enormous enterprise: Since it
isunclear which verbs to start with in the rst place, one would
have to usesuch a large number of different stimuli (and llers and
subjects, etc.) thatthis seems a daunting task. A more attractive
alternative is a corpus-basedapproach (cf. Branigan et al.s
statement on how of corpus approaches canbe useful for hypothesis
generation, which was quoted above in the intro-duction), where
part of the analysis can be (semi-)automated. In order tolook at
this in more detail, consider Table IV for an abstraction from
thestudy of the dative alternation.
The null hypothesis (H0) that has apparently been assumed in
exper-imental studies on syntactic priming is that (the strengths
of) the primingeffects are independent of the verb(s). More
technically, in the experimentalparadigms referred to above it was
argued that, on the whole, observed aand observed d (henceforth
aobs and dobs) should be higher than expecteda and d (henceforth
aexp and dexp) respectively across all verbs (the sameargument was
put forward in the preceding section on the corpus-basedapproach),
and the implicit assumption seems the be that this is also truefor
each individual verb. But rather than take H0 for granted, let us
look atwhether this hypothesis is actually borne out by the data.
To that end, let uslook at one stimulus set of one particularly
interesting study, namely Picker-ing and Branigan (1998). Their
experimental items (Pickering and Branigan,1998: Exp. 1) involve
the ten verbs listed in alphabetical order in (7), which
-
376 Stefan Th. Gries
Table V. CPrime CTarget: Observed vs. Expected Frequencies for
give
CTarget: CTarget: RowData (ICE-GB) V NP PPfor/to V NP/S NP/S
totals
CPrime: V NP PPfor/to 85 (126.2) 184 (142.8) 269CPrime: V NP/S
NP/S 51 (183.5) 340 (207.5) 391Column totals 136 524 660
also occur differently frequently as part of the priming or the
target struc-ture in the stimulus set; for example, offer is used
as a priming verb in bothditransitive and prepositional dative
priming contexts, but not as a targetverb.
(7) give, hand, lend, loan, offer, post, sell, send, show,
throw
If we look at the results for the rst of these verbs in the
present dataset, we obtain the results in Table V, where the
expected values of eachrow areas abovenot computed on the basis of
the column totals, buton the basis of the overall frequencies of
the two constructions in Table I.
Table V indicates that there are 660 occurrences of prime-target
pairswith the verb lemma give in the target position (where it
should be subject tosyntactic priming). In 269 of these 660 cases,
the prime had a prepositionaldative structure, in the remaining 391
cases it had a ditransitive structure.Although we have seen a
strong priming effect for the dative alternationacross all verbs
(cf. Table III), the results for give do not reect this over-all
tendency. As is obvious, the results are not exactly as predicted
by H0:While dobs is larger than dexp (indicating syntactic priming
of the ditransi-tive construction), no such effect is found for the
prepositional dativebycontrast, the ditransitive is preferred even
if the prime is a prepositionaldative: bobs is higher than bexp.
Interestingly, an analogous examination ofthe second verb listed in
(7), hand, results in a completely different distribu-tion.
Consider Table VI for the results concerning this verb. As is
clear, inthis case, aobs is higher than aexp (reecting a priming
effect for the prep-ositional dative), but no such effect is
obtained for the ditransitiverathercobs is higher than cexp.
It is only the third verb listed in (7) which appears to behave
inaccordance with H0, namely lend. Consider Table VII, where, at
last, thepriming effect is balanced: aobs and dobs are higher than
aexp and dexprespectively.
Interestingly, these are not isolated patterns. Out of the 10
verbs listedin (7), seven occur in both constructions in the corpus
(loan and throw
-
Syntactic Priming: A Corpus-based Approach 377
Table VI. CPrime CTarget: Observed vs. Expected Frequencies for
hand
CTarget: CTarget: RowData (ICE-GB) V NP PPfor/to V NP/S NP/S
totals
CPrime: V NP PPfor/to 7 (3.8) 1 (4.2) 8CPrime: V NP/S NP/S 9
(5.6) 3 (6.4) 12Column totals 16 4 20
Table VII. CPrime CTarget: Observed vs. Expected Frequencies for
lend
CTarget: CTarget: RowData (ICE-GB) V NP PPfor/to V NP/S NP/S
totals
CPrime: V NP PPfor/to 10 (5.6) 2 (6.4) 12CPrime: V NP/S NP/S 2
(7.5) 14 (8.5) 16Column totals 12 16 28
only occur in the ditransitive, post occurs only in the
prepositional dative).If we look at these seven verbs, we nd
that
show and offer pattern like give, which prefers the ditransitive
con-struction, (cf. Tables (A.2) and (A.3) in Appendix A for
details);
sell patterns like hand, a verb preferring the prepositional
dative(cf. Table (A.4) in Appendix A for the exact gures); and
send patterns roughly like lend (and, thus, as predicted by
H0)(roughly because, for send, dobs is only about the same as
dexp).
While it is important to note that this specic nding does not
inval-idate the general priming effect, some verbs appear more
likely to resistpriming. It seems as if they preferred to occur in
one construction and thequestion arises as to how to motivate this
discrepancy. One possible expla-nation for these ndings is based on
recent general research on subcatego-rization preferences of verbs
(and verb senses).
Most previous approaches to subcategorization preferences just
quan-tify the attraction of some word W to some construction C in
terms of theraw frequency of W in C (examples include Connine et
al., 1984, Hunston& Francis, 2000, Lapata et al., 2001; and
Hare et al., 2003, to namebut a few). In a series of publications,
Stefanowitsch and Gries devel-oped a family of techniques for
quantifying the strengths of associationbetween words and
particular (slots of) constructions, the so-called
col-lostructional analysis (cf. for details Stefanowitsch &
Gries, 2003; Gries& Stefanowitsch, 2004a, b). These techniques
make it possible to identify
-
378 Stefan Th. Gries
Table VIII. The Distribution of give in the Ditransitive and the
Prepositional Dative withto in the ICE-GB
(pFisherexact=1.84E120)
Data (ICE-GB) give other verbs Row totals
ditransitive 461 574 1035construction (213) (822)prepositional
dative 146 1773 1919construction (with to) (394) (1525)Column
totals 607 2347 2954
what they refer to as signicant collexemes, i.e., the verbs that
are moststrongly attracted by the V slot of the, say, ditransitive
construction, thepassive construction, the verb-particle
construction, etc.; elds of appli-cation of collostructional
analysis include, but are not limited to, theidentication and (more
precise) measurement of subcategorization pref-erences, the
investigation of semantic properties of constructions and
theirimplications for acquisition, etc. In contrast to the previous
raw-frequencyapproaches just mentioned, collostructional analysis
also takes into con-sideration the overall frequencies of W and C
in the corpus to deter-mine whether the distribution of W in the
relevant slot of C deviates fromthe one already expected by chance
alone, a precaution that many of theabove studies have failed to
take.8 The method most relevant to our pur-poses is an extension of
the investigation of distinctive collocates calleddistinctive
collexeme analysis.9 It requires to rst identify how a verb
isdistributed across two alternative constructions, as is
represented in TableVIII; gures in parentheses are again expected
frequencies.
For every such table (one for each verb), Gries and
Stefanowitschcompute a Fisher exact test to determine to which
construction the verbis more strongly attracted; in the above
example, it is immediately obviousthat give is much more strongly
attracted to the ditransitive than to theprepositional dative with
to (cf. the ratio of observed to expected frequen-cies in the upper
left cell of Table VIII), and corresponding ndings can
8 To obtain an R script that computes all methods constituting
collostructional analysis,contact the present author. Cf. Gries,
Hampe and Schonefeld (in press a, b) for experimen-tal evidence
that demonstrates the superiority of collostruction strength over
raw frequencyon the basis of comparing the results of a corpus
analysis to results of a sentence comple-tion experiment and a
self-paced reading-time experiment.
9 Distinctive collocates are collocates that differentiate
between two node words (cf. Churchet al. 1991; Gries, 2001, 2003b
for details).
-
Syntactic Priming: A Corpus-based Approach 379
be obtained for all verbs occurring at least once in at least
one of the twoconstructions under investigation.
For the present purposes, we will restrict our attention to the
verbs in(7), and it turns out that they indeed exhibit similar
kinds of preferences:Of the verbs used by Pickering and Branigan
(1998), give, show and offerare signicant collexemes of the
ditransitive (all ps < .1E 09) whereassell and hand are
signicant collexemes of the prepositional datives (allps < .01)
lend and send do not exhibit a signicant preference foreither
construction (p >.13); the computations are based on all 3973
di-transitives and prepositional datives mentioned in Table
I.10
Noting the strong distinctive collostruction strengths of
Pickering andBranigans (1998: Exp. 1) stimulus verbs, we can now
explain the ndingsof Table VVII by proposing an alternative
hypothesis. This alternative tothe above null hypothesis, the
collostruction-based hypothesis H1, is thata verb strongly
associated with a particular construction resists primingand rather
sticks to its associated construction. Again more technically,for
collexemes of the ditransitive, bobs and dobs should be higher
thanbexp and dexp, and for collexemes of the prepositional dative,
aobs and cobsshould be higher than aexp and cexp. Finally, the
verbs without a strongassociation to a construction, i.e. where no
verb-construction associationwould be expected to block the priming
effect, should exhibit the distribu-tion predicted by H0 as
observed for all verbs together in Table III. As isobvious, the
results obtained from the corpus data are the ones predictedby the
collostruction-based hypothesis H1.
More specically, there is a majority of (types of) verbs in the
cor-pus which have no strong association to one of the two
constructions andwhich are thus fully responsive to priming
effects;11 since for many ofthese the distribution postulated in H0
is found, the picture that emerges isthe overall priming effect of
previous studies and of Table III in the pres-ent study. On the
other hand, a minority of (types of) verbs is stronglyassociated
with a particular construction and these are therefore
moreresistant to priming, which is why their patterns or priming
success ratesdiffer from those of the others. Put differently, the
priming rates of these
10 Differences between the present results and those of Gries
(in press) or Gries and Stef-anowitsch (2004a) are due to the fact
that these earlier studies included only prepositionaldatives with
to and/or instances with nominal objects.
11 Only 86 out of the 316 dative verb types in the ICE-GB
(27.2%) have a signicant asso-ciation to one of the two
constructions. A yet more extreme distribution is observed forthe
transitive phrasal verbs to be discussed in the following section:
Only 40 out of the 700transitive phrasal verb types in the ICE-GB
(5.7%) have a signicant association to one ofthe two
constructions.
-
380 Stefan Th. Gries
latter verbs are inuenced by their preferences to occur in
particular con-structions.12
Although no study on priming has so far looked at these issues
indetail, this nding should not even strike one as particularly
surprising.Previous work within psycholinguistics has so
convincingly demonstratedthat verbs subcategorizing preferences
play an important role in the easeand speed of lexical access,
ambiguity resolution etc. (cf. Garnsey et al.1997; Hare et al.,
2003; Stallings et al., 1998 as well as the references citedthere)
that such results should actually have been anticipated long ago.
Itwould be interesting to test how the above results relate to
other data onverb frame preferences; cf., e.g., Connine et al.
(1984), where the verbslisted here were unfortunately not
included.
To summarize, on the one hand, the results from the
corpus-basedinvestigation of priming have documented a clear
priming effect for boththe ditransitive and the prepositional
dative, and many characteristics ofthe corpus-based priming effects
clearly resemble those of previous experi-ments. However, this
section has also provided evidence that offers a muchmore detailed
perspective on this global effect by showing that primingeffects
are verb-specic: Individual verbs associations to particular
con-structions as measured by distinctive collostruction strength
result in someverbs allowing priming in one particular direction
more readily than oth-ers. This tendency is masked by the overall
priming effects, but takinginto account individual verbs behavior
can provide a more precise pictureof the processing of the verbs
and the structures in which they are used.The following section
illustrates the potential of this way of analysis for adifferent
constructional alternation that has so far received little
attentionin the literature on priming, namely verb-particle
constructions.
PARTICLE PLACEMENT
General Investigation
If syntactic priming can indeed be attributed to the processing
of aparticular structure, then it should be manifested in a variety
of differ-
12 One important point must be claried here. To some readers,
this approach may seemsomewhat circular; they might object to my
line of reasoning by saying, Wait a minute!You start out by using
the corpus data to compute priming effects. And then you use
thevery same corpus data to compute collostruction strengths. No
wonder you get such a highcorrelation of ndings you measure the
same thing under two different labels! In thegeneral discussion
below, I will provide evidence why the issue of circularity is
unproblem-atic here.
-
Syntactic Priming: A Corpus-based Approach 381
Table IX. Verb-particle Constructions: Medium Construction
Data (ICE-GB) V Prt NP V NP Prt Row totals
spoken 698 963 1661written 553 229 782Column totals 1251 1192
2443
ent, nearly synonymous constituent structures. As I have
mentioned above,however, most, if not all, work on syntactic
priming in English has beenrestricted to the activepassive
alternation and the dative alternation dis-cussed above. This paper
builds on the work by Gries (in press) andextends the corpus-based
approach to priming to particle placement, thealternation of
transitive phrasal verbs exemplied in (8).
(8) a. John [VP picked up [ NPdirect object the book]b. John [VP
picked [ NPdirect object the book] up]
Just as above, I rst extracted all examples of these two
constructionsfrom the ICE-GB corpus; since the particles found in
verb-particle con-structions are tagged as adverbial particles of
phrasal verbs, this was doneby retrieving all examples parsed as
[VP V AdvPrt [NP/S ]] or [VP V [NP/S ]AdvPrt]. As a result, I
obtained the data set summarized in Table IX; cf.Fig. (3) for the
scatterplot representing the switch rates per corpus le.
Out of these 2443 cases, 646 had to be discarded for the priming
anal-ysis again because they were the rst or last such construction
either in acorpus le or in a subtext of a corpus le, leaving 1797
prime-target pairsfor the analysis. Each of these was then coded
for the same features as thetwo dative construcions with the
addition of analogously coded featuresfor PartPrime, PartTargetand
PartID as well as PhrasVPrime, PhrasV-Target and PhrasVID (i.e.
what is the particle in prime and target and isit identical in
both, and the same for the phrasal verb). As above, con-sider rst
the most general result concerning the frequencies of CPrimeand
CTarget across all variables in Table X.
Table X. CPrime CTarget: Observed vs. Expected Frequencies (2(1)
= 183.62, p < .001)
CTarget: CTarget: RowData (ICE-GB) V Part NP/S V NP/S Part
totals
CPrime: V Part NP/S 548 (444) 319 (423) 867CPrime: V NP/S Part
300 (476.2) 630 (453.8) 930Column totals 848 949 1797
-
382 Stefan Th. Gries
The result is again unambiguous: there is a strong tendency for
theprime construction to be repeated; the construction with the
VP-nalparticle is the one with the stronger priming effect. While
this providesadditional support to the general idea of syntactic
priming (with a newconstruction, though), it is again important to
look at the more detailedresults of a subsequent GLM analysis; cf.
Table (A.5) in the Appendix Afor the exact results of this
multifactorial analysis. Again the above effectof CPrime is the
strongest of all (p .5)for reasons discussed abovebut the
correlation between the percentages of successful priming and
Dis-tance can best be described by a quadratic equation (adjusted
R2 = .39,F(1, 12) = 3.85, p = .051), i.e. for the range of Distance
values included,priming decreases as the distance increases.
The majority of previous studies over the last 15 or so years
was con-cerned with active/passive and datives. As in Hartsuiker et
al. (1999), how-ever, the present results indicate that priming
effects can also be obtainedfor cases where the alternants consist
of the same phrases in differentorders. The general ndings
concerning particle placement are somewhatsimilar to those of the
dative alternation. There is a general priming effectso that
constructions are likely to be repeated at the next
opportunity.
-
Syntactic Priming: A Corpus-based Approach 383
In addition, both the dative alternation and particle placement
exhibit asimilar strength of the priming effect, and in both cases
one construc-tion primes more strongly than the other. Finally,
there are slight effectsof the verb form and the verb lemma which,
although insignicant, areat least in the same direction as the
analogous (experimental and cor-pus-based) effects for the datives;
the same holds for the directionality ofpriming. While the results
are in need of additional evidence, they provideprima facie
evidence of structural priming for a construction hardly relatedto
syntactic priming in previous work. But let us now see whether
parti-cle placement is subject to the same kind of verb-specicity
effects as thedative alternation.
Verb-Specic Investigation
By analogy to the discussion of the dative alternation, we now
turn tothe issue of whether the overall priming effect observed for
particle place-ment needs to be qualied with respect to individual
verbs preferences.Since there is no previous experimental study the
stimulus sets of whichcan be examined in this connection, I have
chosen six verbs for analysisfrom the data discussed in Gries and
Stefanowitsch (2004a); the methodol-ogy is the same as outlined
above for the datives: for each phrasal verb theconstruction which
it prefers has been determined by means of a Fisherexact test on a
distribution such as that exemplied by Table VIII. Two ofthe verbs
chosen are signicant collexemes of the construction [VP V
PartNP/S], two are signicant collexemes of the [VP V NP/S Part]
constructionand two have no association to either construction; cf.
(9).
(9) a. [VP V Part NP/S]: take up, nd outb. [VP V NP/S Part]: put
in, take outc. no association: pick up, put down
For reasons of space, I will not discuss the results for
particle place-ment at the same level of detail as before. As it
turns out, the ndings arequite similar to those for the dative
alternation with respect to the verbspecicity of priming. For
example, take up is a verb with a strong col-lostructional
attraction to the construction where the verb and the par-ticle are
adjacent, and as Table XI illustrates, priming is
correspondinglyrestricted to this construction; cf. Table (A.6) for
analogous results for ndout.
A similar point can be made for verbs associated with the
particle-nal construction. A case in point is put in in Table XII,
where primingis restricted to that construction; cf. also Table
(A.7) in Appendix A forthe data on take out. Note in this
connection that put in and take out
-
384 Stefan Th. Gries
Table XI. CPrime CTarget: Observed vs. Expected Frequencies for
take up
Data (ICE-GB) Target: V Part NP/S Target: V NP/S Part Row
totals
Prime: V Part NP/S 7 (5.6) 4 (5.4) 11Prime: V NP/S Part 9 (6.1)
3 (5.9) 12Column totals 16 7 23
Table XII. CPrime CTarget: Observed vs. Expected Frequencies for
put in
Data (ICE-GB) Target: V Part NP/S Target: V NP/S Part Row
totals
Prime: V Part NP/S 7 (8.2) 9 (7.8) 16Prime: V NP/S Part 7 (14.3)
21 (13.7) 28Column totals 14 30 44
are particularly relevant in this context becauseunlike some
other pre-viously discussed verbsthey are distinctive for one
construction withoutnearly exclusively occurring in that
construction: put in occurs in [VP VPart NP/S] and [VP V NP/S Part
] 21 and 33 times respectively while takeout occurs in [VP V Part
NP/S] and [VP V NP/S Part ] 15 and 26 timesrespectively. The fact
that these verbs priming effects still exhibit the verbspecicity
effect illustrates that their priming behavior and
collostructionstrength are not automatic reexes of their raw
frequencies.
Finally, consider a verb which has absolutely no preference for
oneverb-particle construction, namely pick up. As is evident from
Table XIII,priming occurs for both constructions, and the same
holds for anotherunbiased verb, namely put down (cf. Table (A.8) in
Appendix A).
To summarize, we have again obtained a clear priming effect
forboth constructions, but also more detailed evidence on the
verb-specic-ity effect: Some verbs association to a verb-particle
construction appearto allow for, or resist, the priming effect much
more strongly than others.Given that the two verb-particle
constructions are associated with seman-tically different groups of
verbs (cf. Gries 2003a, Gries & Stefanowitsch
Table XIII. CPrime CTarget: Observed vs. Expected Frequencies
for pick up
Data (ICE-GB) Target: V Part NP/S Target: V NP/S Part Row
totals
Prime: V Part NP/S 15 (13.8) 12 (13.2) 27Prime: V NP/S Part 16
(21.5) 26 (20.5) 42Column totals 31 38 69
-
Syntactic Priming: A Corpus-based Approach 385
2004a for details), this even invites inferences as to how
semantic proper-ties of verbs correlate with the strength of
priming effects.
DISCUSSION AND CONCLUSIONS
Interim Summary
Let me rst summarize what I believe to be the two main points
ofthis paper: First, the analysis of syntactic priming on the basis
of cor-pora has made it possible to identify priming effects for
two construc-tional alternatives, the dative alternation and
particle placement. What ismore, the results for the dative
alternation, for which many experimen-tal studies have provided
results against which the present results can becompared, are very
similar in many respects to the experimental results.Lastly, the
results of the corpus-based approach to priming make it possi-ble
to address many of the issues that are currently debated (i.e.,
duration,directionality, the construction-specicity and relative
independence of themedium of syntactic priming).
Second, the main result of this study, however, is that the
resultsfor both constructions have indicated that the degree to
which individualverbs are sensitive to priming differs strongly
across verbs. To my knowl-edge, this is the very rst study which
has identied this important effectwhich may be hidden in many
previous studies, and future work on theseinteresting ndings and
its exact implications are necessary. Ideally, thisnding would lead
to either re-analyses of existing studies or new experi-mental
approaches where collostruction strength is properly controlled
for;one such example will be discussed below.
The Relation between Experimental Data and Corpus Data
The previous sections have illustrated how the repetition of
syntacticstructures can be investigated from a corpus-based
perspective. However,as I mentioned above briey, Branigan and
Pickering have argued againstcorpus-based approaches to syntactic
priming, claiming that the primingeffects obtained from
naturalistic data may have to be attributed to non-syntactic
reasons. The nonsyntactic factors they mention include tempo-rary
switches to more formal registers at certain points in the
interview[ . . . or . . . ] the well-known facilitatory effects of
repeating particularwords (Pickering & Branigan 1999: 136).
Also, they briey refer to dis-course-motivated syntactic
repetitions. In the light of these potential pointsof critique, it
is necessary to address how corpus data compare to experi-mental
approaches in the analysis of priming.
-
386 Stefan Th. Gries
Let me rst state something quite explicitly: It is true that, in
gen-eral, experimental studies are in a better position to single
out particularaspects of priming more easily than corpus-based
studies, and the possibil-ity to hold experimental conditions
constant across a variety of trials and(combinations of) conditions
should not be underestimated. However, theexploratory benets of
corpus data have been mentioned by Pickering andBranigan themselves
(cf. above), and from a different perspective, the con-trolled
nature of experimental conditions also has some drawbacks.
First, the priming data are usually collected in a very
narrowlydened and articial setting. While this is desirable from
the point of viewof delimiting error variance, it does not allow
generalizations of the role ofregister effects on syntactic
primingthe corpus data, by contrast, allowfor a multifactorial
analysis of syntactic repetition in natural settings. Inaddition,
in their discussion of previous experimental approaches to
prim-ing, Hartsuiker and Kolk (1998: 148) criticize much previous
work fornot taking into consideration the overall frequencies of
syntactic construc-tions, whichif not considered properlymay
introduce frequency effectsinto the priming results. In the present
approach, the corpus data allowfor a natural computation of
construction baseline frequencies. Second,not all experimental
studies managed to account for all potential explan-atory factors.
For example, Bock and Loebells (1990) ndings were inter-preted as
evidence for the irrelevance of thematic utterance
characteristicsand that function words were irrelevant to priming
until Hare and Gold-bergs (1999) and Bencini, et al.s, (2002)
replications showed that this wasnot necessarily the case. Of
course, this does not invalidate the experimen-tal approach as
such, but it points out that the number of factors to betaken into
consideration is so high that it is not always possible to holdthem
all constant. Thus, including such confounding factors into a
cor-pus-based evaluation may sometimes be a useful alternative.
Finally, byinvestigating syntactic priming from a corpus-based
perspective, one candetermine to what degree it plays a role for
grammatical variation, i.e.the phenomenon that in a given discourse
situation the speaker may havethe choice between two
truth-conditionally equivalent, nearly synonymousconstructions
(e.g. between the two dative constructions, active vs. pas-sive, or
the of-genitive vs. the s-genitive, etc.). Including the priming
effectsinto the research design may make it possible to increase
the accuracy ofpredicting the construction the speaker will choose
(subconsciously); cf.Gries (2003a, in press) on verb-particle
constructions and Rohdenburg andMondorf (2003) for a more general
perspective.
Apart from these general methodological arguments, some other
morespecic comments on Pickering and Branigans nonsyntactic factors
are duebecause not all of these lend themselves to an explanation
of the present
-
Syntactic Priming: A Corpus-based Approach 387
results (and Pickering and Branigan have not provided empirical
evidencefor their claims). For example, the fact that one of the
two constructionsmay be predominant in a particular register is
taken into account here since(a) the corpus data cover a wide
variety of registers and (b) the medium(speaking vs. writing) was
included into the analysis. Note also that neitheralternation
investigated is inherently related to a particular level of
formal-ity, and explaining the frequent cases of syntactic priming
by hundreds ofsudden register/formality changes does not seem very
plausible.
Similarly, the effects cannot be straightforwardly reduced to,
say, thegivenness or semantic characteristics of the direct objects
referent: First,both datives have information structure properties
(cf. Thompson, 1990),so why should only one result in priming in
corpora? Second, one mightsuggest that the slight priming
prominence of the verb-particle construc-tion with a VP-nal
particle is due to the fact that this construction isassociated
with a given referent of the direct object (cf. Gries, 2003a,
Sect.6.1.4: Once the referent of the direct object has been
introduced, the verb-particle constructions in the subsequent
discourse will place it before theparticle. However, Gries (2003a,
120121, 131) found priming effects forVPCs regardless of whether
the referent of the direct object NP in the sec-ond construction is
coreferential with that of the rst. Third, the kind
ofanimacy/argument effects that might in principle affect datives
(such thatanimacy affects constructional choices) cannot explain
the results on VPCswhere animacy plays no role (cf. Gries, 2003a,
8889]) and the particleis often aspectual or idiomatically used and
can, thus, not be attributedargument status.
An additional important point is that other non-syntactic
factors canalso not be held responsible for the present ndings. For
example, thosewho would like to attribute the present results to
lexical repetition effectswould have to explain why, in the case of
dative alternation, it is theditransitive construction that primes
more strongly although (i) it is theprepositional dative which
allows for the priming of the function wordsto and for and (ii) the
fact that lexical activation decays too fast makesit unlikely that
the long duration of priming effects observed here and inother
(experimental studies) is just a lexical memory effect.
In sum, much of the present ndings resembles those obtained
exper-imentally so strongly that they cannot be explained away as
easily as sug-gested. While I do not rule out discourse-motivated
factors of priming atall, it is hard to explain all the
similarities between the different kinds ofresults and still simply
uphold the claim that all this is epiphenomenal.Without doubt,
further experimental evidence is necessary, but it seems asif the
utility of corpus-based, explorative results should not be
underesti-mated prematurely.
-
388 Stefan Th. Gries
Anticipating Circularity Objections
In the rst section on the verb-specicity of the priming effects,
I raisedthe potential objection of the circularity of argument (cf.
above note 12).Since this is an important point, let me clarify why
this objection is incon-sequential here. A rst reason is that the
computations of the verbs prim-ing effects on the one hand and
their collostruction strengths on the otherhand do not use the
exact same set of data: The former are computed onlyfrom 80% of the
full set of constructions, namely those 3003 cases
whereprime-target pairs are found in the same subtexts of the same
corpus leswhereas the latter is computed from all 3793 cases.
Second, and more impor-tantly, the two measures need not correlate
highly. Consider a case wherea verb V occurs 46 times in a corpus,
34 times with construction C1 and12 times with construction C2; the
frequencies of C1 and C2 in the cor-pus are each 1000 times, the
corpus size is 138,664 (the number of verbs inthe ICE-GB). From
this it follows that V is signicantly associated with C1(pFisher
exact 0.0007). Assume further that 6 out of these 46
constructions(13%, i.e. even less then in the actual data) had to
be discarded because theywere the rst constructional occurrences in
their respective les. This leavesus with 40 cases, which we assume
to be distributed as represented in TableXIV; as usual expected
frequencies are parenthesized.
As Table XIV indicates, there is signicant priming of verb V
inboth directions/for both constructions: aobs > aexp and dobs
> dexp(pFisher exact = .0144). But since we have already seen
that V is stronglyassociated with C1, this also shows that
strength/direction of primingeffects and collostruction strength
are not automatically correlated even ifcomputed from the same
corpus: A verb can be strongly associated withone construction but
still prime the other one and, thus, the argument isnot
circular.
Apart from these two more theoretical arguments, there is also
addi-tional empirical evidence to support my contention that
priming effectsare verb-specic irrespective of whether
collostruction strength is computedfrom the same corpus as the
priming effects. For one thing, collostruc-tion strength turns out
to be very robust across (comparable) corpora.
Table XIV. CPrime CTarget: A hypothetical Data Set
Data (ICE-GB) CTarget: C1 CTarget: C2 Row totals
CPrime: C1 12 (8) 4 (8) 16CPrime: C2 0 (2) 4 (2) 4Column totals
12 8 20
-
Syntactic Priming: A Corpus-based Approach 389
Gries, et al. (in press a, b), for example, computed the
collostructionstrengths of verbs to the as-predicative (e.g. I
regard her as clever) in theICE-GB and then, for a later study, of
the BNC sampler, a subpart of theBritish National Corpus consisting
of 2m words of (spoken and written)English. As it turns out, when
the signicant collexemes of the as-predica-tive are sorted in each
corpus, then the signicant collexemes of the as-pred-icative in one
corpus are not only also a signicant collexeme in the othercorpus,
but they also tend to occupy similar ranks in the other corpus.
Forexample, 60% of the 30 most signicant collexemes in the ICE-GB
are alsoamong the 30 most signicant collexemes in the BNC sampler.
What ismore, the correlation between the ranks of all verbs
occurring in both cor-pora is highly signicant ( = .535; z = 6.97;p
< .001) as is that betweenthe ranks of all signicant collexemes
( = .591; z = 5.29;p < .001). Inother words, given this high
correspondence between different corpora, it isvery likely that
computing verb-specicity from another corpus would nothave changed
the picture markedly. Once other manually parsed corpora
areavailable, checking the present results will be easy.
Finally, there is also experimental evidence supporting the
verb-specic-ity effect argued for here. Gries and Wulff (in press)
replicated Pickering andBranigans (1998) experiments on syntactic
priming in English with nativespeakers of German to determine
whether syntactic priming is also obtainedwith advanced learners of
a foreign language (recall point (v) in the Intro-duction). In
addition to a general priming effect, they found that, just likein
the present study, the strength of the priming effect of the seven
dative-alternation verbs discussed above is strongly correlated
with a general biasof the subjects to use the experimental verbs in
particular constructions.In Figure 1, their results are summarized:
the y-axis portrays the bias ofindividual verbs to either the
ditransitive or the prepositional dative in thecorpus data,
basically as measured by collostruction strength. The
x-axisportrays the preference of individual verbs to be completed
using eitherthe ditransitive or the prepositional dative in the
sentence-completion task.Finally, the strong correlation (r2 = .8;
t (5) = 4.47;p = .007) is indicatedby the slope of the regression
line, which shows that one can predict theoutcome in the priming
experiment on the basis of the verbs preferences asmeasured on the
basis of the corpus data.
Even though these results were not obtained with native
speakers,they do point to the fact that experimentally primed
sentence completionis strongly sensitive to verb bias. Thus, I
submit, this issue is clearly inneed of further research of which
corpus linguistic methods may play anessential role in determining
collostruction strengths for verbs to be tested(cf., e.g., Gries
and Stefanowitsch, 2004a on verbs distinctive for activesand
passives).
-
390 Stefan Th. Gries
Fig. 1. The verbs constructional preferences in the corpus data
and in thepriming-experi-ment responses.
Syntactic Priming and its Verb-Specicity in a Psycholinguistic
Model
Many previous results have been explained within, say, the
psycholin-guistic model proposed by Pickering and Branigan (1998).
In this model,syntactic priming is accounted for in terms of
combinatorial nodes whichare activated when a verb is used in a
particular construction. When aspeaker produces a verb in a
particular construction, the lemma nodes ofall words produced as
well as their feature nodes (representing morpho-logical features
such as number, tense, etc.) and the corresponding combi-natorial
node are activated. Since the activation level of these nodes
andthe links relating them decays only gradually, the nodes and
links thatwere just used are more likely to be used again when the
next opportu-nity arises; syntactic priming is the result. Since
the combinatorial nodesare directly related to the lemma nodes,
priming should be stronger whenthe same verb is used in both prime
and target, but the verb form as suchshould not inuence the priming
effect.
Since the present ndings concerning the dative alternation are
sosimilar to previous experimental work, they can of course be
equally
-
Syntactic Priming: A Corpus-based Approach 391
well integrated into Pickering and Branigans model. Remember,
for exam-ple, that not only priming was found, the priming effect
was also stron-ger when the verb lemmas were identical. In
addition, there was alsoa tendency for identical verb forms to
result in a stronger primingeffect, and while that was not
hypothesized in the above model, similartendencies were obtained in
Pickering and Branigans (1998: Exp. 5) onsingular-vs.-plural form
differences. Also, since the model has been arguedto involve a
shared representation in comprehension and production, thefact that
SpeakerID had no strong effect in this study can be
explainednaturally. Finally, the fact that the verb-particle
constructions exhibit prim-ing supports the idea that order
information is encoded within combina-torial nodes.
Given this kind of psycholinguistic model, the second kind of
nd-ing of this study, the verb-specicity of priming, can be
integrated straight-forwardly. Recall that each verb lemma is
connected to the combinatorialnodes of the construction in which
the verb can be used. Since syntacticpriming of a construction C
involves the repeated activation of Cs com-binatorial node (so that
its resting level is exceeded), it follows naturallythat when the
link between a verb and C is stronger, priming of (only)
thatconstruction should be stronger. This is exactly what we nd:
the verbswhich are strongly associated with one construction
exhibit priming withthis construction much more strongly than with
the other construction.Thus, we only need to supplement Pickering
and Branigans model withthe notion that the links between verb
lemmas and combinatorial nodesthey postulated anyway can also be
differentially strong to reect theirdegree of attraction/repulsion
to a construction as measured by collostruc-tion strength. This
would allow for the model to accommodate the presentndings on
verb-specicity, but also allows for an economical represen-tation
of many of the ndings concerning verb subcategorization
prefer-ences, verb bias etc. mentioned above.13 Given the current
interest inthe issue of whether syntactic priming is best explained
as activation pat-terns or implicit learning (cf. Chang et al.,
2000, 2003), it is even conceiv-able that the network architectures
used to test these different conceptionscould be somehow enriched
with the collostructional information.
All in all, the present ndings demonstrate how usefulin spite
ofsome limitationscorpus-based approaches to priming phenomena can
beto support and extend ndings obtained with other methodologies,
pro-moting once more the ideal of converging evidence.
13 For a comprehensive illustration of how particle placement
can be accounted for in amodel of this sort, cf. Gries (2003a: ch.
8); for further discussion of this model and others,cf. Hare et
al., (2003: 296297).
-
392 Stefan Th. Gries
APPENDIX A
Table A.1. GLM Results for the Priming of Datives
Effect source F p Partial 2
CPrime 161.671
-
Syntactic Priming: A Corpus-based Approach 393
Table A.4. CPrime CTarget: Observed vs. Expected Frequencies for
sell
CTarget: CTarget: RowData (ICE-GB) V NP PPfor/to V NP/S NP/S
totals
CPrime: V NP PPf or/to 4 (1.9) 0 (2.1) 4CPrime: V NP/S NP/S 5
(2.8) 1 (3.2) 6Column totals 9 1 10
Table A.5. GLM Results for the Priming of Verb-particle
Constructions
Effect source F p Partial 2
CPrime 25.451
-
394 Stefan Th. Gries
Table A.7. CPrime CTarget: Observed vs. Expected Frequencies for
take out
CTarget: CTarget: RowData (ICE-GB) V Part NP/S V NP/S Part
totals
CPrime: V Part NP/S 7 (7.7) 8 (7.3) 15CPrime: V NP/S Part 5
(9.7) 14 (9.3) 19Column totals 12 22 34
Table A.8. CPrime CTarget: Observed vs. Expected Frequencies for
put down
CTarget: CTarget: RowData (ICE-GB) V Part NP/S V NP/S Part
totals
CPrime: V Part NP/S 8 (5.1) 2 (4.9) 10CPrime: V NP/S Part 5
(8.2) 11 (7.8) 16Column totals 13 13 26
-
Syntactic Priming: A Corpus-based Approach 395
Fig. 2. Relative frequencies of ditransitives and prepositional
datives plotted against theirswitch rates per corpus le.
-
396 Stefan Th. Gries
Fig. 3. Relative frequencies of V Part DO and V DO Part plotted
against their switch ratesper corpus le.
-
Syntactic Priming: A Corpus-based Approach 397
REFERENCES
Bencini, G., Bock, J. K., & Goldberg, A. E. (2002). How
abstract is grammar? Evidence fromstructural priming in language
production. Paper presented at the 15th CUNY Confer-ence on Human
Sentence Processing.
Bock, J. K. (1986). Syntactic persistence in language
production. Cognitive Psychology 18,355387.
Bock, J. K. (1989). Closed-class immanence in sentence
production. Cognition 31(2), 163186.Bock, J. K. (1990). Structure
in language: Creating form in talk. American Psychologist
45(11), 12211236.Bock, J. K. (2002). Persistent structural
priming from comprehension to production. Paper
presented at the 15th CUNY Conference on Human Sentence
Processing.Bock, J. K., & Loebell, H. (1990). Framing
sentences. Cognition 35(1), 139.Bock, J. K., Loebell, H., &
Morey, R. (1992). From conceptual roles to structural
relations:
Bridging the syntactic cleft. Psychological Review 99(1),
150171.Bock, J. K., & Grifn, Z. M. (2000). The persistence of
structural priming: Transient activa-
tion or implicit learning. Journal of Experimental Psychology:
General 129(2), 177192.Branigan, H. P., Pickering, M. J., &
Cleland, A. A. (1999). Syntactic priming in written pro-
duction: Evidence for rapid decay. Psychonomic Bulletin and
Review 6(4), 635640.Branigan, H. P., Pickering, M. J., &
Cleland, A. A. (2000). Syntactic co-ordination in dia-
logue. Cognition 75(2), B13B25.Branigan, H. P., Pickering, M.
J., Liversedge, S. P., Stewart, A. J., & Urbach, T. P.
(1995).
Syntactic priming: Investigating the mental representation of
language. Journal of Psy-cholinguistic Research 24(6), 489506.
Branigan, H. P., Pickering, M. J., Stewart, A. J., & McLean,
J. F. (2000). Syntactic priming inspoken production: Linguistic and
temporal interference. Memory and Cognition 28(8),12971302.
Chang, F., Dell, G. S., Bock, J. K., & Grifn, Z. M. (2000).
Structural priming as implicitlearning: A comparison of models of
sentence production. Journal of PsycholinguisticResearch 29(2),
217229.
Chang, F., Dell, G. S. & Bock, J. K. (2003). Comprehending
structural priming: A connec-tionist learning account. Paper
presented at the 16th CUNY Conference on Human Sen-tence
Processing.
Church, K. W., Gale, W., Hanks, P., & Hindle, D. (1991).
Using statistics in lexical analysis.In Zernik Uri, (ed.). Lexical
acquisition: Exploiting On-line Resources to Build up a Lex-icon.
Hillsdale, NJ: Lawrence Erlbaum, pp. 115164.
Connine, C., Ferreira, F., Jones, C., Clifton, C., &
Frazier, L. (1984). Verb frame preferences:Descriptive norms.
Journal of Psycholinguistic Research 13(4), 307319.
Corley, M. & Scheepers, C. (2002). Syntactic priming in
English sentence production: Cat-egorical and latency evidence from
an internet-based study. Psychonomic Bulletin andReview 9(1),
126131.
Estival, D. (1985). Syntactic priming of the passive in English.
Text 5(1), 722.Garnsey, S. M., Pearlmutter, N. J., Myers, E., &
Lotocky, M. A. (1997). The contributions
of verb bias and plausibility to the comprehension of
temporarily ambiguous sentences.Journal of Memory and Language
37(1), 5893.
Gries, St. Th. (2003a). Multifactorial Analysis in Corpus
Linguistics: The Case of ParticlePlacement. London/New York:
Continuum Press.
Gries, St. Th. (2003b). Testing the sub-test: A
collocational-overlap analysis of English -icand -ical adjectives.
International Journal of Corpus Linguistics 8(1), 3161.
-
398 Stefan Th. Gries
Gries, St. Th., (in press). New perspectives on old
alternations. In Cihlar, J. E., L. Franklin,A., Kaiser, D. W.,
& Kimbara, I. (eds.). Papers from the 39th Regional Meeting of
theChicago Linguistics Society: Vol. II. The Panels. Chicago, IL:
Chicago Linguistics Soci-ety.
Gries, St. Th., & Stefanowitsch, A. (2004a). Extending
collostructional analysis: A corpus-based perspective on
alternations. International Journal of Corpus Linguistics 9(1),
97129.
Gries, St. Th., & Stefanowitsch, A. (2004b). Co-varying
collexemes in the into-causative. InAchard, M. & Kemmer, S.,
(eds.). Language, Culture, and Mind (pp. 225236) Stanford,CA:
CSLI.
Gries, St. Th., Hampe, B., & Schonefeld, D. (in press a).
Converging evidence: Bringingtogether experimental and corpus data
on the association of verbs and constructions.Cognitive
Linguistics.
Gries, St. Th., Hampe, B., & Schonefeld, D. (in press b)
Converging evidence II: More onthe association of verbs and
constructions. In J. Newman & S. Rice (Eds.). Empiricaland
Experimental Methods in Cognitive/Functional Research. Stanford,
CA: CSLI.
Gries, St. Th., & Wulff, S. (in press). Do foreign language
learners also have constructions:Evidence from priming, sorting,
and corpora. Annual Review of Cognitive Linguistics.
Hare, M. L., & Goldberg, A. E. (1999) Structural priming:
Purely syntactic? In Hahn, M.,& Stones, S. C. (eds.).
Proceedings of the 21st Annual Meeting of the Cognitive
ScienceSociety (pp. 208211). Mahwah, NJ: Lawrence Erlbaum.
Hare, M. L., McRae, K., & Elman, J. L. (2003). Sense and
structure: Meaning as a determi-nant of verb subcategorization
preferences. Journal of Memory and Language 48(2), 281303.
Hartsuiker, R. J., & Kolk, Herman H. J. (1998). Syntactic
persistence in Dutch. Languageand Speech 41(2), 143184.
Hartsuiker, R. J., Kolk, Herman H. J., & Huiskamp, P.
(1999). Priming word order in sen-tence production. The Quarterly
Journal of Experimental Psychology 52A(1), 129147.
Hartsuiker, R. J., Pickering, M. J., & Veltkamp, E. (2002).
Is syntax shared between lan-guages? Cross-linguistic syntactic
priming Spanish/English bilinguals. Paper presented atAMLap
2002.
Hartsuiker, R. J., & Westenberg, C. (2000). Word order
priming in written and spoken sen-tence production. Cognition
75(2), B27B39.
Levelt, W. J. M., & Kelter, S. (1982). Surface form and
memory in question answering. Cog-nitive Psychology 14(1),
78106.
Pickering, M. J., & Branigan, H. P. (1998). The
representation of verbs: Evidence from syn-tactic priming in
language production. Journal of Memory and Language 39(4),
633651.
Pickering, M. J., & Branigan, H. P. (1999). Syntactic
priming in language production. Trendsin Cognitive Sciences 3(4),
136141.
Pickering, M. J., Branigan, H. P., Cleland, A. A., &
Stewart, A. J. (2000). Activation ofsyntactic information during
language production. Journal of Psycholinguistic Research29(2),
205216.
Pickering, M. J., Branigan, H. P., & McLean, J. F. (2002).
Constituent structure is formulatedin one stage. Journal of Memory
and Language 46(3), 586605.
Potter, M. C., & Lombardi, L. (1998). Syntactic priming in
immediate recall of sentences.Journal of Memory and Language 38(3),
265282.
Rohdenburg, G. & Mondorf, B. (eds.). Determinants of
Grammatical Variation in English.Berlin, New York: Mouton de
Gruyter.
Salamoura, A. (2002). Cross-linguistic structural priming and
bilingual models of production.Paper presented at AMLap 2002.
-
Syntactic Priming: A Corpus-based Approach 399
Sankoff, D. & Laberge, S. (1978). Statistical dependence
among successive occurrences of avariable in discourse. In Sankoff,
D. & Laberge S. (eds.). Linguistic Variation: Methodsand
Models. (pp. 119126). New York: Academic Press.
Scheepers, C. & Corley, M. (2000). Syntactic priming in
German sentence production. In:Gleitman, L. & Joshi A. K.
(eds.). Proceedings of the 22nd Annual Meeting of the Cog-nitive
Science Society (pp. 435440) Mahwah, NJ: Lawrence Erlbaum.
Smith, M. & Wheeldon, L. (2001). Syntactic priming in spoken
sentence production: Anonline study. Cognition 78(2), 123164.
Stallings, L. M., MacDonald, M. C., & OSeaghdha, P. G.
(1998). Phrasal ordering con-straints in sentence production:
Phrase length and verb disposition in Heavy-NP Shift.Journal of
Memory and Language 39(3), 392417.
Stefanowitsch, A. & Gries, St. Th. (2003). Collostructions:
Investigating the interactionbetween words and constructions.
International Journal of Corpus Linguistics 8(2),209243.
Szmrecsanyi, B. (2005). Persistence phenomena in the Grammar of
Spoken English. Doctoraldissertation, University of Freiburg.
Thompson, S. A. (1990). Information ow and Dative Shift in
English discourse. In: Ed-mondson, J. A., Feagin, C. &
Muhlhausler, P. (eds.). Development and Diversity: Lan-guage
Variation Across Time and Space. (pp. 239253). Dallas: SIL and
University ofArlington, TX.
Weiner, E. J. & Labov, W. (1983). Constraints on the
agentless passive. Journal of Linguistics19, 2958.
Yamashita, H., Chang, F. & Hirose, Y. (2002). Separating
functions and positions: Evidencefrom structural priming in
Japanese. Paper presented at the 15th CUNY Conference onHuman
Sentence Processing.