Argument constructions and language processing:
Evidence from a priming experiment and pedagogical implications1
Francisco Ruiz de Mendoza
The notion of argument construction is widely accepted in Cognitive Linguistics circles as a highly explanatory
theoretical construct. It has recently been incorporated into the Lexical-Constructional Model (LCM; Ruiz de
Mendoza and Mairal 2007, 2008), a theoretical approach to meaning construction that integrates argument
constructions into a broader model that incorporates meaning dimensions traditionally dealt with in the domain of
pragmatics and discourse analysis. The LCM has an argument level of description, in the form of constructional
templates, which are in general equivalent to the argument constructions postulated in Goldberg’s (1995, 2006)
Construction Grammar, although there are crucial differences in the descriptive metalanguage and the principles
used to explain their linguistic behavior. Since the LCM is geared to have psychological adequacy, like all cognitive
(cf. Lakoff 1990) and some functional approaches to language (e.g. Dik 1997), it is essential to find to what extent
the notion of argument construction is a real processing mechanism. This paper presents a priming experiment
measuring reaction times to sentences that had the same formal configurations but were based on different argument
constructions, and to sentences that were formally similar and embodied the same argument construction. Subjects
responded significantly faster to target sentences that shared the same construction as the previously viewed
sentences. Thus, constructions appear to be psychological entities that play an important role in linguistic
processing. If this is correct, then it has potentially important consequences for planning adequate L2 teaching
Keywords: lexical-constructional model; argument construction; constructional template; processing;
priming experiment; reaction time; teaching strategies
The notion of argument construction is central to a number of construction grammar accounts,
especially Goldberg’s Construction Grammar (Goldberg 1995, 2006; Michaelis 2003), which is
strongly grounded in Cognitive Linguistics (see Gonzálvez and Butler 2006, Dirven and Ruiz de
Mendoza 2008 for a comparison among the different approaches). It is also key to the core
grammar level of the Lexical Constructional Model (LCM), which combines insights from
cognitive and functional accounts of language (Ruiz de Mendoza and Mairal 2007, 2008; Mairal
and Ruiz de Mendoza, 2008; cf. Butler 2008 for a review). In all these approaches it is generally
postulated that we store in our minds higher-level argument configurations that can be
instantiated by lower-level configurations (e.g. verbal predicates and their associated arguments)
as licensed by coercion processes. For example, compare the use of push, a predicate expressing
caused-motion, in Noel Gallagher of Oasis got pushed off the stage at the Virgin Festival by a
rogue fan, and laugh, a non-causal activity predicate, in Ari Fleischer was laughed off the stage
by reporters. A linguistic account in which the argument structure of verbal predicates is seen as
determining the overall argument structure configuration of a sentence can deal with the sentence
containing push. However, such an account cannot adequately explain why it is possible to use
the verb laugh with the same argument structure configuration as push. Laugh typically occurs in
the intransitive form (e.g. He didn’t say anything; he just laughed) or in pseudo-transitive
configurations with a prepositional complement (e.g. The teacher laughed at me). In
Construction Grammar it is assumed that ‘laugh’ is coerced into a causal transitive format (cf.
Michaelis 2003). The LCM adds a licensing factor for coercion to take place. In the example, the
predicate ‘laugh’ can be built into the caused-motion construction because it is possible to see
psychological or emotional impact in terms of physical impact, which is a form of metaphorical
thought that Ruiz de Mendoza and Mairal (2007) have labeled high-level metaphor, since it has a
more abstract nature than lexical or situational metaphors. Because of their abstract nature, high-
level metaphors usually underlie grammatical phenomena.
Since postulating the existence of constructional meaning is necessary to explain the
linguistic behavior of laugh (and many other uses of other verbal predicates, as will be seen
below), we need to know if argument constructions are more than a theoretical convenience and
can be regarded as cognitively adequate constructs, which means that they actually play a role in
language processing. There has been some research that is suggestive of the psychological
adequacy of argument constructions (cf. Goldberg 2006 for a review). The present paper aims to
contribute to the still scarce literature on the psychological adequacy of postulating
constructional meaning. We will do so by means of a priming experiment. The experiment shows
that prior presentation of sentences with a particular construction speeds the subsequent
recognition of a sentence containing the same construction. The experimental results by no
means involve that constructional meaning have an overriding effect on purely syntactic
processing, but they point to constructional meaning having some role. Further and more
complex experimentation will be needed to find out about the balance between lexical,
constructional and syntactic clues to processing. In the meantime, we believe it is enough to find
that constructions are at least plausible constructs from a psychological perspective. This being
so, we contend that constructions should be taken into account in second language teaching
As noted by Schönefeld (2005), the notion of construction has a long tradition in linguistics. It
may refer to either specific syntactic patterns (e.g. passives, existentials, participials) or to clause
types conveying a complete idea. In Cognitive Linguistics constructions are understood
differently as learned pairings of form and function, and are regarded as the basic units of
language. In this view of language, the theoretical machinery used to describe lexicon and
grammar is essentially the same. Many constructions are idiomatic in different degrees (cf.
Fillmore, Kay, and O’Connor 1988). When a construction is fully idiomatic, i.e. when its
elements resist lexical or grammatical alteration (e.g. spill the beans, kick it/kick the bucket) we
have a substantive idiom. When a construction is a combination of fixed and variable elements,
as is the case of What’s X Doing Y?, studied by Kay and Fillmore (1999), which conveys the
meaning that the situation described worries the speaker (cf. What’s that child doing in the
kitchen with the carving knife?), we have a formal idiom.
Other constructions, like the caused-motion configuration (which is described by
Goldberg, 1995 as X CAUSES Y TO MOVE Z), have a more abstract nature. These are called
argument constructions since they contain argument roles related by abstract predicates. These
constructions, like lexical items, have their own range of meanings (i.e. a construction can be
polysemous) and meaning relations. An example of constructional polysemy is provided by the
ditransitive construction (Goldberg 1995: 75). Compare Mary gave John a cake (where X causes
Y to receive Z), with Mary baked John a cake (where X intends to cause Y to receive Z), or
Mary denied John a cake (where X causes Y not to receive Z). Relations between constructions,
which in functional and formal models of language have been called alternations (cf. Levin,
1993), depend on other factors, some of which have been discussed by Goldberg (1995) in terms
of “links” between constructions. One kind of link occurs when a construction is a more specific
instance of another construction. This is the case of the ditransitive, which is a specification of
the more generic subject-object construction in terms of the type of verb phrase (involving a
potential or actual transfer) and semantic roles that can occur in it (agent, theme, recipient). We
may also find subpart links, which relate independent constructions, such as the caused-motion
and the intransitive constructions, where one profiles part of the elements of the other: The
sergeant marched the recruits profiles the cause (the sergeant) and the theme (the recruits)
elements, whereas The recruits marched only profiles the theme (recruits). Instance links, in
turn, apply when one construction is a special case of another construction. For example, the
verb drive may be used in a resultative sense that has an idiomatic character (She drives me
crazy/mad/*bored/*sad). This kind of idiomatic construction is a special case of the resultative
construction. Finally, the resultative construction in She drank him unconscious can be
considered to be a metaphorical extension of a literal construction as in She hammered the metal
flat, on the basis of the experiential conflation between goals and destinations (reaching a
destination is a form of attaining a goal) that allow us to understand the former in terms of the
The LCM bases its descriptive apparatus on lexical and constructional templates, both of
which share the same descriptive metalanguage. Lexical templates consist of combinations of
lexical functions and semantic primes. The former capture the semantic and pragmatic
parameters that underlie predicate meaning; the latter are based on the Aktionsart (i.e. lexical
aspect) characterizations proposed by Van Valin and LaPolla (1997) and Van Valin (2005) as a
development of Vendler’s (1957) seminal work in this respect. Both lexical functions and
semantic primes have a number of variables (or predicate arguments) within their scope. In a
parallel way, constructional templates are made up of constructional functions and abstract
semantic primes. Lexical variables fuse into corresponding argument structure variables (or
roles) and their associated structure, thus giving rise to fully developed argument structure
configurations. Interestingly, both lexical and constructional templates share the same Aktionsart
characterizations, e.g. abstract predicates like CAUSE, BECOME, INGR (for ‘ingressive’),
SEMEL (for ‘semelfactive’), which respectively signal cause-effect relationships, change,
suddenness in a change of state, and brief duration (cf. Van Valin and LaPolla 1997, Van Valin
2005). The list of abstract predicates is only partially overlapping with those used on a rather ad
hoc basis in Goldberg’s (1995) Construction Grammar. For example, INGR and SEMEL are not
used by Goldberg, and predicates like MOVE, HAVE, DIRECT ACTION, which are used by
Goldberg, are not based on Aktionsart distinctions, but have a role parallel to that of lexical
functions in a lexical description. The advantage of constructional descriptions in the LCM is
twofold: (i) using the same metalanguage for lexical and constructional templates endows this
aspect of the model with descriptive regularity and responds to the fact that constructional
descriptions are based on abstractions over a large number of verbal predicate descriptions (for
example, the caused-motion construction is derived from verbal predicates such as push and
throw that express caused-motion); (ii) all descriptive elements are assigned their proper role, i.e.
there is no mixing up of descriptive categories: Aktionsart elements (e.g. CAUSE, BECOME)
belong to the logical structure characterization of the construction and other elements are
regarded as constructional functions (e.g. LOC in the caused-motion constructional template [do´
(x, y)] CAUSE [BECOME *NOT be-LOC´ (y, z)], which is read as ‘an entity x does something
to another entity y such that y is caused to become in a different location z’). However, despite
these differences with Construction Grammar, the LCM shares with Construction Grammar the
view that argument structure does not only derive from lexical items but results from the
principled interaction between lexical and constructional structure.
Until fairly recently, the adequacy of postulating constructions has been mainly defended
on the basis of linguistic evidence (e.g. Bybee 2003; Croft 2001; Fillmore et al. 1988; Goldberg
1995, 2006; Jackendoff 2002; Kay and Fillmore 1999; Langacker 2005; Michaelis 2003;
Verhagen 2005) and some complementary evidence coming from the field of child language
(e.g. Diessel and Tomasello 2001; Goldberg 2006; Tomasello 2003). The theoretical import of a
constructional account of language from a linguistic perspective has been shown by Goldberg in
a number of studies. To illustrate her views, we will adapt some of the observations given in
Goldberg and Bencini (2005) about the power of construction-based explanations of linguistic
structure. Consider, for example, the verb sneeze in the following sentences:
a. Mary sneezed all day long.
b. The song is about a meatball that is sneezed off of the top of a plate of spaghetti.
c. She sneezed herself silly.
d. Did you hear the tale of a woman who took a pinch of snuff and sneezed herself to
e. The elephant sneezed a sneeze that nearly unseated his rider.
f. She hacked and coughed and sneezed her way to the doctor’s office.
g. Who sneezed on my laptop?
h. Grandma was trying to feed the baby but got sneezed at first.
i. Naturopathy is nothing to be sneezed at!
In order to account for all these uses of sneeze, a verb-based (or projectionist) account would
have to postulate multiple argument structures, each associated with a specific meaning.
However, it would be rather odd to postulate senses such as ‘to cause something to move by
sneezing at it’–as in example (1b)– ‘to cause a resultant state by sneezing’ –as in examples (1c)
and (1d)– ‘to cause a specific type of sneeze to happen’ –example (1e)– ‘to remove all obstacles
to reach a place by sneezing’ –example (1f)– ‘to make a place dirty by sneezing’ –example (1g)–
and ‘to ignore something without giving it much thought’. In a constructionist account we only
have one generic meaning for sneeze as the action of expelling air through the nose violently and
involuntarily. This meaning is then adapted to constructional meaning requirements, thus
acquiring a causal sense –examples (1b)- (1f)– and/or focusing attention on the result of sneezing
–examples (1b)-(1g)– or even seeing the act of sneezing as directed to an object, whether in a
literal (1h) or a non-literal way (1i). Each of these constructions may take in many other verbal
predicates that share the same sensitiveness to the constructional requirements as sneeze.
Consider now some of the many possible instantiations of the caused-motion construction:
a. He blew the leaf off the tree stump.
b. She drove me into the arms of another woman.
c. It nearly scared me out my seat.
d. The audience laughed the actor off the stage.
e. He stared me out of the room.
f. They talked me into the business.
g. She loved me back into existence.
The examples in (2) make use of a caused-motion configuration for different purposes. In (2a)
the predicate ‘blow’, whose canonical use is intransitive, is transitivized thus acquiring a causal
sense that licenses its use in the caused-motion construction. In this construction, the object is
what Ruiz de Mendoza and Mairal (2007) have called an effectual object, i.e. an object that
directly receives and is affected by the physical impact of the action. The adaptation of the verbal
predicate and its associated argument structure that takes place in (2a) is very similar to the one
in (2b), with the difference that ‘drive’ has acquired a specialized sense as a (literal) caused-
motion predicate (e.g. drive a nail into the wall), which here, in a default interpretation, is used
figuratively. Sentence (2c) has a transitive verb that would not normally signal physical
causation, which points to a metaphor-based licensing of its participation in the caused-motion
construction, in which there is physical causation. As a result of the metaphor, the object is
treated as if it were an effectual object. Examples (2d) and (2e) take us a step further. In both we
have an object-goal, grammatically marked by the preposition at, seen as an effectual object. In
(2d), and (2e), just as in (2c), there is literal caused motion. The overall interpretive rationale
behind these three examples is used in (2f), with the only difference that here motion is
figurative. Finally, (2g) goes even further since ‘love’ does not designate an action but a state (cf.
van Valin and LaPolla 1997) with an object within its scope. In Hallidayan grammar (cf.
Halliday and Matthiessen 2004) the predicate ‘love’ has two associated roles: a sensor and a
phenomenon. In (2g) the protagonist, the sensor, is seen as if she were the effector of an action
and the speaker, the phenomenon, is seen as it were an effectual object.
Our discussion of the linguistic regularities in examples in (1) and (2) points to the need to
take constructional meaning seriously if we want to have a better understanding of the intricacies
of meaning construction. Note that Bencini and Goldberg (2000) do not deny the importance of
verbal semantics in figuring out the meaning of a given construct. In fact, it is readily
acknowledged that on some occasions verbal meaning is much more powerful than
constructional semantics (e.g. “envy” in The prince envied him his fortune, where the ‘successful
transfer’ core meaning of the ditransitive construction is overridden by the meaning of the verbal
predicate). Rather, the fundamental point Bencini and Goldberg make is that constructions are
overall better determinants of sentence meaning than verb meaning. This claim is partially
compatible with our own findings (see Section 4) and also with the spirit of the LCM. In our
view, the constructional account complements verb-based approaches that regard syntax either as
the result of instantiating the argument structure of verbal predicates or as an autonomous system
that is constrained by the syntactically-relevant thematic structure of predicates. If this is so, it is
a natural question to determine the psychological adequacy of the concept of construction and its
apparent role in linguistic processing at sentence level. We turn to this question in the following
3. Syntax-based or construction-based processing?
The idea that syntactic structure is of primary importance in sentence comprehension is hardly
questionable. Powerful evidence in favor of this claim can be found in pause tests and eye
movement analysis while processing structurally ambiguous sentences (e.g. Frazier and Rayner
1982). A good example is provided by work on so-called “garden-path sentences”, i.e. sentences
whose structural ambiguity mislead readers into selecting the wrong interpretation, such as The
old man the boats (Rumelhart 1977) or The prime number few (Milne 1982). For example, in a
reading-aloud protocol, Rumelhart (1977) reports that experimental subjects would make a pause
between man and the boats while reading aloud the sentence The old man the boats. The pause
was interpreted as an indication of structural reanalysis and subsequent reinterpretation where
man is a verb and the adjective old is the head of a noun phrase. However, there is also evidence
not only that syntactic parsing is not always the primary factor in sentence comprehension, but
also that syntactic structure may be completely overridden by, for example, lexical and world
knowledge (e.g. MacDonald 1993; MacDonald et al. 1994, St. John and McClelland 1990) or
factors related to exemplar-based frequency, i.e. the piecemeal learning of thousands of
constructions and the frequency-biased abstraction of regularities within them (Ellis 2002).
The observation that world knowledge may override syntax-based processing is not new at
all. For example, as far back as 1966, Slobin found that non-reversible passives like The flowers
are being watered by the girl are understood and remembered more easily than reversible
passives like The boy was hit by the girl. The lexical sequence flowers-water-girl allows us to
come up with an interpretation without making use of syntax (we do not expect flowers to water
girls), which is not the case with the sequence boy-hit-girl (boys may hit girls or girls may hit
boys). Evidently, lexically-cued common world knowledge plays a role in interpretation to such
an extent that it may override syntax (cf. Slobin 1966).
But the question of sentence comprehension goes beyond merely attesting to the
importance of lexical knowledge and frequency factors. Since there is linguistic support in favor
of the existence of argument-based constructions, it is legitimate to inquire into their
psychological status and their role, if any, in sentence comprehension. The psychological nature
of idiomatic constructions, whether substantive or formal, has been investigated in much detail
by Gibbs (1994), who argues that they are often partially non-compositional and grounded in
metaphor. Whether we accept Gibb’s thesis or not, the psychological reality of idioms (i.e. their
status as accessible and interpretable linguistic units) is as self-evident as the psychological
reality of simple and compound content-carrying lexical items and should pose no problem in
terms of what cognitive linguists often refer to as the cognitive commitment (cf. Lakoff 1990).
This commitment aims to make linguistic generalizations either compatible with or subservient
to empirical evidence coming from research in any branch of the cognitive sciences (including
experimental psychology, neurology, and artificial intelligence). However, the psychological
reality of non-idiomatic (or argument) constructions, like those discussed in the previous section,
is a matter of controversy.
In recent years the validity of argument constructions has been explored from the point of
view of psycholinguistic research (cf. Hare and Goldberg 1999; Bencini and Goldberg 2000;
Goldberg and Bencini 2005). Let us take an example. Bencini and Goldberg (2000) report
experimental evidence of two sorting experiments designed to find out whether argument
constructions play a role in determining sentence meaning. Participants were required to sort four
sets of sentences in terms of their overall meaning. Each set presented one of the verbs in four
different constructions: transitive (e.g. Pat threw the hammer), ditransitive (e.g. Chris threw
Linda the pencil), caused-motion (e.g. Pat threw the key onto the roof), and resultative (e.g. Lyn
threw the box apart). The stimuli used were designed so as to minimize content overlap that
could distort the results of the experiment. Thus, the experiment avoided the use of identical
lexical items or near synonyms in the stimuli other than the verbal predicate tested. For example,
sentences like Pat shot the duck and Pat shot the duck dead were discarded since they would
likely be grouped together on the basis of shared content, despite the fact that they make use of
different constructions (transitive versus resultative). Sentences like Pat shot the elephant and
Patricia stabbed a pachyderm, which make use of the same construction, were also excluded
since they would probably be grouped together not on constructional grounds but simply because
of the use of close synonyms. In the experiment, 7 out of 17 participants sorted entirely by
construction, no participant sorted entirely by verb, and the other 10 produced mixed sorts.
Further testing, which consisted of computing a deviation score from entirely verb-based and
construction-based scores, showed that the mixed-performance participants did not sort by
relying exclusively either on the verb or on the construction. The results suggest that verb-
centered (i.e. projectionist) views of meaning interpretation may not be sustained since
constructions play a significant role whatever the processing style of the individual.
Further evidence in favor of the psychological reality of argument constructions comes
from priming experiments. Priming has been argued to be a promising way of tapping into
linguistic processes (Braningan et al. 2005). Interestingly enough, as reported by Hare and
Goldberg (1999) and Goldberg and Bencini (2005), there are priming experiments that tend to
support syntactic processing as the overriding factor in sentence comprehension. These
experiments, in principle, would undermine the theoretical status of constructions as relevant
factors in guiding interpretation. For example, when shown a picture that could be described
either with a ditransitive (e.g. John gave the dog a biscuit) or a prepositional dative sentence (e.g.
John gave a biscuit to the dog), subjects were more likely to describe it with a ditransitive if they
had been primed with a preceding unrelated ditransitive (Bock and Loebell 1990; see also Bock,
Loebell, and Morey 1992). However, as Hare and Goldberg (1999) note, the syntactic primes
used in the experiments sometimes shared significant semantic structure with the target
sentences. For example, the prepositional locative The widow drove an old Mercedes to the
church was found to prime the prepositional dative The widow gave an old Mercedes to the
church. At first sight, priming may be attributed to syntactic similarity despite the different
thematic structure. But the two sentences share enough lexical content to interfere in the results
and, as observed by Goldberg and Bencini (2005), the dative and locative expressions are
instances of the caused-motion construction. These two facts are enough to provide a reliable
alternative interpretation of Bock and Loebell’s (1990) results. However, in order to test whether
semantic factors play a role in cases of priming like this, Hare and Goldberg (1999) replicated
Bock and Loebell’s (1990) experiment, but introducing a third prime, a ‘provide with’ structure
(e.g. The officers provided the soldiers with guns), which is syntactically similar to the
prepositional dative (NP V NP PP) and parallels the ditransitive in terms of the order of semantic
roles (<agent, recipient, theme>). Results showed that ‘provide with’ expressions prime
ditransitives as much as ditransitives themselves, which again argues in favor of the important
weight of constructional meaning in sentence comprehension.
4. A grammaticality judgment priming experiment
In the research context that we have described, priming effects were obtained for ditransitive
constructions. Other experiments, such as the one based on sorting tasks, considered transitive,
ditransitive, caused-motion, and resultative configurations. We wonder whether priming effects
would reinforce the findings obtained on the basis of sorting tasks. We further wonder whether
other constructions, such as the benefactive, the instrument subject, the X one’s way, and the
reciprocal constructions would also respond positively to priming tests. Note that the
constructions explored in the sorting task were very closely related from a formal and even
semantic perspective. This observation is fairly evident in the case of the transitive-ditransitive
connection, but it also holds for the resultative and caused-motion constructions, which can be
considered extensions of the basic transitive pattern; cf. The blacksmith hammered the metal
(transitive) > The blacksmith hammered the metal flat (resultative)/into the shape of a bird
(figurative use of caused-motion to express result). The resultative and caused-motion
constructions require an effectual object of the action, which explains the need for them to be
constructed on the basis of a prototypical transitive pattern. In contrast, our own choice of
additional constructions enhances formal and semantic diversity rather than relatedness: the
benefactive focuses on the argument that receives the benefit of the action; the instrument subject
construction gives prominence to the instrument of the action by promoting it to the subject
position; the way-construction (e.g. Ellen pushed her way through obstacles ‘created a way
through obstacles by pushing’) conflates action and manner of action into the verbal predicate
which has a fixed object (one’s way) that is treated (through metaphor-based licensing) as if it
were an effectual object rather than the result of the action; and the reciprocal construction is
semantically transitive but grammatically intransitive (e.g. in Raymond and Carla met, Raymond
meets Carla and Carla meets Raymond). We also decided to explore a different priming task
based on grammaticality judgments. The reason for this is that grammaticality judgments would
allow us to design a more complex experiment in terms of the broader range of constructions for
which pictorial representation would hardly be feasible.
We achieved our experimental design goal by comparing sentences that contain particular
constructional templates, such as Henry and Myra agreed (which contains a reciprocal
construction) with sentences such as Kristie and Dan studied which have the same syntactic
structure, but no reciprocal construction. This was inspired by a study by Costa and Sebastián-
Gallés (1998), in which the syllabic structure of primes had an effect on the reaction times of
targets with the same syllabic structure. In the grammaticality judgment task we employed,
subjects saw strings of words on the computer screen and had to decide whether the string made
a good English sentence or a bad one. Reaction times served as the dependent variable. The
following subsections give the details of the experiment and discuss the results.
Test subjects consisted of 36 linguistically naïve native English-speaking college students
enrolled in undergraduate foreign language classes: 12 females and 23 males, with an average
age of 22.8 (range: 18-26).
4.2. Stimulus items
A total of 12 test sentences were used: two sentences for each of six different constructional
templates (see Appendix). In the grammaticality judgment task, these 12 sentences were targets
that were immediately preceded by three priming sentences. The primes either had the same
constructional template as the target or only shared syntactic similarity. The former were our
research objective and the latter acted as a control condition on the former. For example, the
reciprocal test sentence Henry and Myra agreed was preceded by one of the two following sets
Same Constructional Template
Only Syntactically Similar
Raymond and Karla met.
Beth and Wally sneezed.
Jon and Barney visited.
Kristie and Dan studied.
Klein and Garcia collaborated.
Lee and Holly graduated.
All the sentences in (3) share a ‘NP objectless VP’ formal configuration, but the set on the left
fulfills the requirements for ascription to the reciprocal construction (cf. Raymond and Karla met
each other; Jon and Barney visited each other; Klein and Garcia collaborated with each other)
while the set on the right does not (*Beth and Wally sneezed (at) each other; *Kristie and Dan
studied each other; *Lee and Holly graduated each other). We used this arrangement for all our
test-stimuli sentences, which resulted in 12 blocks of four items: three primes followed by a test
target. All of the sentences in these blocks were licit English sentences. An additional 40 licit
sentences (e.g. They competed for her attention) and 68 illicit sentences (e.g. Josey knelt the
ocean awake) that were syntactically similar to the other primes were also included as distractors
so that subjects were not aware of our real test targets. The constructional heterogeneity of the
control primes and the inclusion of distractors were intended to reduce the possibility of the
control primes creating misleading expectations that would slow down the interpretation of the
control targets. This design resulted in a total of 156 sentences.
Two different test sets were designed. In the first set, one of each of the 12 target sentences
with one of the six constructional templates was preceded by three priming sentences with the
same constructional template, and the other six target sentences by three priming sentences that
were only syntactically similar (see (3)). The second test set differed only in that the type of
primes that preceded each target was reversed. This design allowed us to compare reaction times
to the same sentence that had been preceded by different kinds of primes.
Before starting the experiment the subjects were randomly assigned to view one of the two test
sets. They were then seated at a computer and given these instructions:
In this study you will see sentences on the screen. Your job is to determine if the sentence
is correct or not. If it is correct press the key marked GOOD. If it is not correct press the
key marked BAD. Now, put your left index finger on the BAD key and your right index
finger on the GOOD key. The screen also has notes marked GOOD and BAD to help
you. Let's do some practice sentences so you know what to expect. Please try to answer
as quickly as possible.
At this point the subjects responded to four practice items and were then given a chance to ask
questions. If they understood the task they went on to the experiment that lasted about 10
minutes. If not, they were given more instructions and repeated the practice items before
continuing on to the test items.
The order in which the sentences were presented was quasi-randomized so that the 108
distractor sentences (68 incorrect and 40 correct) were randomly dispersed among the test
blocks, and the test blocks were randomized in regards to each other. In this way, each subject
received a different order of the 156 items. Of course, within the test blocks the three primes all
appeared before the target sentences and their order was fixed. The sentences were presented in
large black font on a white background in the center of the screen in standard American English
orthography. The sentences appeared on the screen for 2200ms with a lag of 1933ms of blank
screen between each one. Pilot studies showed this pace was slow enough that people felt they
had enough time to read and respond, to the sentences, but quick enough that they had to remain
focused on the task. The DMDX2 presentation software was used to carry out the experiment.
4.4. Results and discussion
Technical difficulties resulted in the loss of data from one subject. Of the 420 responses to the
targets (35 subjects times 12 test items) three responses were not registered because they were
made while the next item was being presented. In addition, 41 responses were errors in which
subjects mistakenly pressed the BAD key for a test item. Errors were marked in the output file
generated by the presentation software. Together errors comprise ten percent of responses to test
items. Mean reaction times for the remaining items were used as the dependent variable in a
paired t-test. The independent variable was whether the test sentence was preceded by priming
sentences with the same constructional template or sentences that were only syntactically similar.
No significant difference was found between the two groups (t (11) = 1.9, p = 0.084).
However, inspection of the mean reaction time for each of the 12 test sentences revealed that in
all cases but two, priming the target sentences with sentences with the same constructional
template yielded quicker response times. What is more, the two exceptions were the two
resultative test sentences (Garrett cut their hair short; Protesters burned the office black). The
goal of the experiment was to test whether constructional templates in general play a part in
linguistic processing. The fact that the resultative sentences do not yield the same priming effect
needs to be pointed out, and merits further attention. At this stage we can only speculate. We
suspect that the problem is that test-stimuli sentences like Brittany saw the statute broken, Sarah
perceived the relationship unhealthy, and Josh thought the treatment fair are more than
syntactically similar to resultatives; they are non-central cases of resultative, that is, they have
some resultative ingredient at some stage in processing. For example, Brittany saw the statute
broken metonymically suggests that someone did something to the statute and as a result it was
broken. The same reasoning schema may be applied to the rest of the examples. If this
speculation is true, resultatives will most likely resist an experimental design that attempts to
discriminate between constructional and syntactic priming. At the same time, this problem
should not distract us from the broader goal of examining constructional templates
experimentally. For this reason, we felt justified in removing the resultative sentences and
performing another t-test on the remaining ten items. This analysis demonstrates a clear tendency
toward facilitative priming (t (9) = 2.92, p = 0.017). On average, target sentences that were
primed by sentences with the same constructional template were responded to 180.44ms faster
than in the syntactically similar condition (1536.21ms versus 1716.65 ms respectively).