How do people produce ungrammatical utterances?

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright

Author's personal copy Journal of Memory and Language 67 (2012) 355–370

Contents lists available at SciVerse ScienceDirect

Journal of Memory and Language journal homepage: www.elsevier.com/locate/jml

How do people produce ungrammatical utterances? Iva Ivanova a,b,⇑, Martin J. Pickering c, Janet F. McLean c, Albert Costa b,d, Holly P. Branigan c a

Departament de Psicologia Bàsica, Universitat de Barcelona, Spain Departament de Tecnologies de la Informació i les Comunicacions, Universitat Pompeu Fabra, Spain c Department of Psychology, University of Edinburgh, UK d Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain b

a r t i c l e

i n f o

Article history: Received 24 December 2011 revision received 17 June 2012 Available online 9 August 2012 Keywords: Ungrammatical sentences Structural priming Argument structure

a b s t r a c t We investigate whether people might come to produce utterances that they regard as ungrammatical by examining the production of ungrammatical verb-construction combinations (e.g., The dancer donates the soldier the apple) after exposure to both grammatical and ungrammatical sentences. We contrast two accounts of how such production might take place: an abstract structural persistence account, according to which it is caused by increased activation of an abstract structural rule; and a lexically-driven persistence account, according to which it requires previous exposure to the same (ungrammatical) verb-construction combination. In four structural priming experiments, we found that sentences with ungrammatical verb-construction combinations were produced only after exposure to similar ungrammatical exemplars containing the same verb, but not after such sentences with a different verb, or grammatical sentences with the same construction. These results indicate that people can produce sentences with ungrammatical verb-construction combinations after brief exposure to related sentences, and provide support for the lexically-driven persistence account of such production. Ó 2012 Elsevier Inc. All rights reserved.

Introduction The language we produce every day is far less orderly, error-free or grammatical than we might think. Although perhaps only the Rev. W.A. Spooner would have ridden well-boiled icicles, sewn people to sheets, or inquired after the dizzy bean, adult native speakers of a language could sometimes claim in spontaneous speech that a millionaire donated the charity a new building, even though they would normally consider such a sentence as ungrammatical. Although there are many reasons why speakers might produce such ungrammatical utterances, one possibility is that they can reflect persistence of syntactic structure from previous linguistic exposure. In this paper, we study whether, and what kinds of (brief) exposure can trigger ⇑ Corresponding author. Address: 9500 Gilman Drive, 0948, Department of Psychiatry, University of California, San Diego, La Jolla, CA 920930948, USA. Fax: +1 858 246 1284. E-mail address: [email protected] (I. Ivanova). 0749-596X/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jml.2012.06.003

the production of ungrammatical utterances, and we contrast two accounts of how such production might take place. Some evidence that speakers may persist in producing ungrammatical utterances after comprehending such utterances comes from anecdotal reports of native speakers using their language ungrammatically as a result of hearing it spoken ungrammatically by non-native speakers. In comprehension, a number of studies have demonstrated processing facilitation or increased acceptability of ungrammatical sentences after brief exposure to similar exemplars. For example, Kaschak and Glenberg (2004) observed that reading times for a construction that is ungrammatical in standard English (the needs-construction, as in The meal needs cooked) decreased with consecutive presentations. These results generalized across modalities (spoken to written language) and to different verbs (Kaschak & Glenberg, 2004) and sentential contexts (Kaschak, 2006). Furthermore, Luka and Barsalou (2005) showed that grammaticality ratings for moderately

Author's personal copy 356

I. Ivanova et al. / Journal of Memory and Language 67 (2012) 355–370

ungrammatical sentences (e.g., Armanda carried Fernando the package or Rachel needs to get a tattoo as colourful as Bob has) were higher for those participants who had read them previously than for those who saw them for the first time. This effect was induced by as little as a single presentation and was also obtained for sentences which shared only structure and no content words with those presented during initial exposure. In all, these findings suggest that ungrammatical sentences are processed, and yield persistent effects in the linguistic system, even after people are only briefly exposed to them. They also imply that such persistence occurs independently of lexical content and generalizes across sentence contexts.1 In this paper, we are concerned with sentences in which the ungrammaticality arises from the particular combination of the verb and the construction (as in The dancer donates the soldier the apple). A process by which people can come to produce such sentences after exposure (that is, can generalize from comprehension to production) is structural priming. Structural priming (in production) refers to the observation that speakers tend to repeat a syntactic structure they have experienced, in the presence of alternatives. For example, Bock (1986) showed that participants tended to describe target pictures using the same structure (e.g., a prepositional object dative) that they had used in repeating a previously encountered prime sentence (such as The rock star sold some cocaine to an undercover agent). Structural priming is very widespread, occurring with different constructions (Branigan, Pickering, McLean, & Stewart, 2006; Cleland & Pickering, 2003; Ferreira, 2003; Griffin & Weinstein-Tull, 2003; Hartsuiker & Westenberg, 2000), in different languages (e.g., Bock, 1986; Cai, Pickering, Yan, & Branigan, 2011; Hartsuiker & Kolk, 1998; Scheepers, 2003), in corpora as well as in experiments (Gries, 2005; Szmrecsanyi, 2005), and from anomalous as well as well-formed prime sentences (Ivanova, Pickering, Branigan, McLean, & Costa, 2012). It does not depend on lexical repetition (suggesting that the effect has an abstract nature) but is enhanced by such repetition (the so called lexical boost effect: Cleland & Pickering, 2003; Hartsuiker, Bernolet, Schoonbaert, Speybroeck, & Vanderelst, 2008; Pickering & Branigan, 1998). In addition, structural priming occurs in language comprehension, for example in relation to the resolution of ambiguity (e.g., Branigan, Pickering, & McLean, 2005). Importantly, priming occurs between comprehension and production. For example, Levelt and Kelter (1982) found that participants’ answers tended to use the same structure as the questions, and Potter and Lombardi (1998) demonstrated that the structure of recalled sentences was influenced both by comprehended and produced primes (see also Bock, Dell, Chang, & Onishi, 2007; Branigan, Pickering, & Cleland, 2000; Cleland & Pickering,

1 Additionally, Snyder (2000) observed that people found some ungrammatical sentences (e.g., Who does Mary think that likes John?) more acceptable after they had repeatedly judged their acceptability, a phenomenon known as syntactic satiation. However, acceptability did not increase for some other sentence types. In addition, Snyder’s study has been criticized for its unreplicability and for introducing biases in the experimental paradigm (Sprouse, 2009).

2003; Hartsuiker, Pickering, & Veltkamp, 2004; Hartsuiker et al., 2008). Two mechanisms have been proposed to explain structural priming, and are consistent with the production of ungrammatical verb-construction combinations as a result of structural persistence from previously comprehended sentences. According to Chang, Dell, and Bock (2006), when speakers process a given message with a given structure, the mappings between the message and the structure are strengthened and the linguistic system becomes more prone to expressing similar messages with the same structure. Alternatively, according to Pickering and Branigan (1998), structural priming is due to the increased activation of nodes representing constructions that are linked to respective lemma nodes, so that encountering a verb as part of a particular construction leads to activation of both the verb node and the construction node. In this model, the production of sentences with ungrammatical verb-construction combinations can be caused by the (temporary) establishment of new construction nodes (for constructions that have not been experienced before), or the establishment of new links between existing nodes (when a speaker knows a verb and a construction, but does not currently use that verb with that construction). In all, various empirical and theoretical sources suggest that, in principle, adults can come to produce sentences with ungrammatical verb-construction combinations on the basis of brief prior exposure. But such a tendency might be detrimental for communication if speakers were influenced by any ungrammatical utterance they encounter, as in the speech of one or two people with an imperfect grasp of the language. So our first goal in the present study was to determine whether people can come to produce ungrammatical sentences within a single experimental session, which we assume is roughly analogous to a conversation. Importantly, we focus on the production of ungrammatical verb-construction combinations – that is, on argument structure which is ungrammatical with respect to the verb in the sentence (e.g., The dancer donates the soldier the apple). Argument structure (specifications of the number and type of syntactic constituents obligatorily or optionally occurring with a verb in a sentence) is interesting because it relates to the interface between lexical properties and syntactic structure. The role of lexically-specific syntactic information versus. abstract syntactic structure in language production is currently debated (see Konopka & Bock, 2009). By studying whether people come to produce sentences involving ungrammatical verb-construction combinations (that is, argument structure that is ungrammatical in relation to the verb), we may shed light on the ways in which lexical restrictions and abstract syntactic structure might be implicated in language production. That is, we will see whether or not the influence of previously experienced abstract syntactic structure is sufficiently strong to override lexical constraints. We envisage two accounts of how structural persistence from prior exposure can lead to the production of sentences with ungrammatical verb-construction combinations. On one account, which we term here the abstract structural persistence account, speakers come to produce

Author's personal copy I. Ivanova et al. / Journal of Memory and Language 67 (2012) 355–370

such sentences because of the increased availability of an abstract syntactic rule, as a result of prior processing of (grammatical) utterances using that rule. The increased availability of the rule (or strengthening of the procedures responsible for accessing it) would result in the rule being used with the verb selected for production. For example, speakers might produce a sentence such as The dancer donates the soldier the apple after having recently processed the sentence The waitress gives the monk the book, and thereby accessed the double object rule (V ? V NP NP). There would be a conflict with the verb’s argument structure (DONATE [NP,PP]), but it would be overridden as a result of the increased availability of the double object rule.2 Thus, on this account, abstract structural influences would be dominant over lexical constraints of individual verbs during production. This account predicts that, after exposure to fully grammatical sentences, speakers may produce sentences with ungrammatical verb-construction combinations. On another account (henceforth lexically-driven persistence account), speakers come to produce sentences with ungrammatical verb-construction combinations because they form associations between individual verbs and a construction on an item-by-item basis. In other words, producing such sentences would not be induced by experiencing the double object construction in itself, but only after experiencing donate together with the double object construction (DONATE [NP,NP]). Thus, speakers might produce a sentence such as The dancer donates the soldier the apple after exposure to a structurally similar sentence which also contains the verb donate, such as The waitress donates the monk the book. This account predicts that lexical constraints of individual verbs are robust and cannot be overridden by abstract structural activation. On this account, the production of ungrammatical verb-construction combinations would happen only after exposure to similar ungrammatical exemplars, but not after well-formed ones. Our study therefore had two main goals. Firstly, we aimed to establish whether people can rapidly come to produce sentences with ungrammatical verb-construction combinations on the basis of exposure. Secondly, if such production occurs, we aimed to distinguish between an abstract structural persistence account and a lexically-driven persistence account of its underlying mechanisms. To address these goals, we used a structural priming paradigm. Specifically, we exposed people to well-formed and ungrammatical prime sentences, and we examined whether such sentences primed the production of ungrammatical sentences such as The dancer donates the soldier the apple.

2 It could be argued that double object sentences with non-alternating verbs are not completely ungrammatical, since they (extremely rarely) occur in practice (see Method section of Experiment 1). Thus, it is possible that the double object argument structure forms part of the lexical entries for non-alternating verbs, but is established only very weakly. In that case, persistence might result from strengthening a link between a verb and a construction (thus increasing its chances of subsequent production) instead of establishing such a link anew. It also might be that both of these possibilities hold true for different speakers.

357

Overview of the experiments We carried out four experiments, in which participants read ditransitive prime sentences such as (1–3) and decided whether they were accurate descriptions of subsequently presented pictures; and then described an unrelated target picture depicting a transfer event that involved three entities (such as a dancer donating an apple to a soldier), using the verb written under the picture. We examined whether the structure of participants’ target descriptions was affected by the structure of the prime sentences. 1a. 1b. 2a. 2b. 3a. 3b.

The waitress gives the monk the book. (Double Object) The waitress gives the book to the monk. (Prepositional Object) The waitress returns the monk the book. (Double Object) The waitress returns the book to the monk. (Prepositional Object) The waitress donates the monk the book. (Double Object) The waitress donates the book to the monk. (Prepositional Object)

We therefore compared participants’ structural choices following primes involving non-alternating ditransitive verbs that occur almost exclusively in the prepositional object construction (e.g., deliver, donate, return, recommend) and primes involving alternating ditransitive verbs (e.g., give, show, lend, send), which are equally acceptable in both the prepositional object and the double object constructions (although individual verbs have biases towards one or another construction: see Gries, 2005). We refer to double object sentences with non-alternating verbs as ungrammatical, given that linguists and English-language textbooks assume that such verbs do not permit the dative alternation (e.g., Levin, 1993); though see Footnote 2. Most importantly, we examined whether participants could be primed to produce the ungrammatical double object construction with non-alternating verbs (4). 4.

The dancer donates the soldier the apple

Our predictions were as follows. Firstly, if the production of sentences involving ungrammatical verb-construction combinations does not arise from prior linguistic exposure, we should not observe priming of (4) from any of the prime sentences (1–3). If it does, there are two possibilities. The abstract structural persistence account predicts that ungrammatical double object sentences with non-alternating verbs (4) should be primed by wellformed sentences with the same construction (1). This is because, in this account, the production of sentences with ungrammatical verb-construction combinations is driven by an abstract structural rule independent of lexical items. This account also predicts that sentences such as (4) should

Author's personal copy 358

I. Ivanova et al. / Journal of Memory and Language 67 (2012) 355–370

be primed by ungrammatical primes involving different verbs such as (2) (assuming that people assign the double object structure to 2a when they encounter it, a prediction that we test in Experiment 2), because the structural rule is abstract and hence can be generalized across verbs. Conversely, the lexically-driven persistence account predicts that ungrammatical sentences (4) should not be primed by well-formed primes (1) or ungrammatical primes with different verbs (2), but only by ungrammatical primes involving the same verb as the target verb (3). This is because the production of ungrammatical sentences in this account is driven by an association between verb and construction on an item-by-item basis, such that a verb is associated with a structural rule only after the verb and rule have been experienced in combination; the association thus formed does not generalize to other verbs (but note that, if such lexically-driven persistence results in long-term learning, generalization across verbs might eventually occur).

Fig. 1a. An example of a target picture with an alternating verb.

Experiment 1: primes with alternating verbs, targets with non-alternating verbs In this experiment, we tested whether ditransitive double object sentences with alternating verbs (e.g., The waitress gives the monk the book) primed the production of ditransitive double object sentences with non-alternating verbs (e.g., The dancer donates the soldier the apple). In addition, we included a ‘‘dummy’’ condition with double object primes only, to increase exposure to the double object construction and thus encourage its production. In the abstract structural persistence account, sentences with alternating verbs should prime targets with nonalternating verbs, since the persistence of the double object construction for non-alternating verbs happens as a result of the increased availability of an abstract structural rule. Hence, exposure to the double object construction should trigger the production of this construction with non-alternating verb targets. Conversely, in the lexicallydriven persistence account, no priming should occur. This is because the persistence occurs on an item-by-item basis, only after the verb and construction have previously been experienced in combination. Finally, if the production of sentences with ungrammatical verb-construction combinations is not triggered at all by previously comprehended sentences, we should not observe any priming effect either. Method Participants Twenty-eight participants from the University of Edinburgh community were paid to take part. All were native speakers of English and had normal or corrected-to-normal vision. Materials There were 32 experimental items. Each item consisted of a prime sentence, a picture to be verified against the prime sentence (henceforth match picture), and a target picture. The 64 target pictures depicted ditransitive events,

Fig. 1b. The associated target picture with a non-alternating verb.

each displaying persons or objects with the roles of agent, theme, and beneficiary (Fig. 1). The position of agent and beneficiary was counterbalanced, such that, on half of the target pictures, the agent appeared on the left side and the beneficiary on the right side, and on the other half of the target pictures these positions were reversed. Below each picture was a present-tense verb in capital letters, which participants were instructed to use in their descriptions. There were 32 target pictures for each of two target verb types: alternating (Fig. 1a) and non-alternating (Fig. 1b). Each alternating target verb was paired with a non-alternating target verb (e.g., show was paired with donate), and the respective pictures for the two verbs always depicted the same entities. There were four target pictures for each of eight alternating verbs (shows, hands, sends, chucks, offers, sells, throws, and loans) and for each of eight non-alternating verbs (donates, returns, displays, describes, delivers, demonstrates, reveals, and conveys). The eight prime verbs were all alternating verbs (passes, tosses, gives, lends, mails, brings, rents, and flings) and were always different from the target verbs. Each prime verb always occurred with the same pair of target verbs (e.g., pass always occurred with show and donate). Care was taken to avoid semantic similarity as much as possible between the verbs in such triples. In addition, the entities used in the target pictures were always different from the entities in the prime sentences. The 32 prime sentences occurred in either a Prepositional Object or a Double Object form in the Non-alternating

Author's personal copy 359

I. Ivanova et al. / Journal of Memory and Language 67 (2012) 355–370

Target Verb condition (Examples 5b–c), and in a Double Object form only, in the Alternating Target Verb (Dummy) condition (Examples 5a). 5a.

5b.

5c.

Alternating Target Verb (Dummy), Double Object Prime sentence: The waitress gives the monk the book. Target picture: DANCER SOLDIER APPLE SHOWS Non-alternating Target Verb, Prepositional Object Prime sentence: The waitress gives the book to the monk. Target picture: DANCER SOLDIER APPLE DONATES Non-alternating Target Verb, Double Object Prime sentence: The waitress gives the monk the book. Target picture: DANCER SOLDIER APPLE DONATES

The Dummy condition was included in order to increase the overall exposure of participants to the double object construction and thus encourage as much as possible the production of this construction with non-alternating target verbs. Thus, the double object construction occurred in three-fourths of all ditransitive sentences (rather than in half, as in typical structural priming experiments). However, since it is meaningful to talk about a priming effect only when there are at least two alternative prime structures for a given condition, and at present there was only one (Double Object) in the Alternating Target Verb (Dummy) condition, this condition was excluded from analyses. Note that we used a dummy condition rather than include additional double object sentences as fillers in order to keep as close as possible to the structure of the other experiments reported here (in particular, Experiments 2 and 4) (see Appendix A for the full list of experimental items). The 32 match pictures also depicted ditransitive events (Fig. 2). Half of the match pictures corresponded to the event described in the preceding prime sentence (and required a ‘‘yes’’ response), and half differed from the event described in the preceding prime sentence (and required a ‘‘no’’ response). Furthermore, to increase the demands

of the matching task, the 16 match pictures requiring a ‘‘no’’ response always differed from the preceding prime sentence by only one entity, such that six pictures had a different agent, five pictures had a different beneficiary, and the remaining five pictures had a different theme (e.g., for the prime sentence The waitress gives the book to the monk, the corresponding match picture was of a burglar giving a book to a monk). There were 96 sets of fillers. All of the filler sentences and 64 of the filler pictures for description made use of 18 monotransitive verbs. The remaining 32 filler pictures for description had lower-frequency verbs, mostly of Latin origin (e.g., consume, seize, discharge, reprimand). These were included to disguise the presence of non-alternating verbs (which tend to be of low-frequency and Latin origin) on half of the experimental target pictures. Finally, half of the filler match pictures required a ‘‘yes’’ response. With these materials, we constructed four lists of items, each containing one version of each item, and eight items from each condition, together with the 96 fillers. Order of presentation was randomized for each participant, with the constraint that between two and four fillers separated experimental items, and preceded the first item in the list. Acceptability and usage of non-alternating verbs with the prepositional object and double object constructions To obtain a measure of the acceptability and usage of the non-alternating verbs in the prepositional object and double object constructions for the population we were testing, we carried out the following pre-tests. Firstly, 12 additional participants rated how acceptable each of 32 sentences was on a scale from one (‘‘not an acceptable English sentence’’) to seven (‘‘perfectly acceptable’’). These 32 sentences included eight ditransitive sentences with nonalternating verbs (in four prepositional object and four double object sentences, counterbalanced across participants), eight correct prime sentences (counterbalanced in the same way), eight filler (monotransitive) sentences, and eight meaningless sentences (containing various semantic and syntactic violations; e.g., The galaxy arrives deeply the branch.). The mean acceptability ratings are listed in Table 1. Most importantly, the double object sentences with non-alternating verbs (M = 2.90, SD = .70) received lower ratings than their prepositional object equivalents (M = 5.67, SD = 1.37) [t1 (11) = 6.88, p < .001; t2 (7) = 5.29, p = .001]. However, the ratings for these sentences were also higher than those for the meaningless sentences Table 1 Mean acceptability ratings for double object and prepositional object ditransitive sentences with alternating and non-alternating verbs. Sentence type

Construction type

Mean rating

SD

Ditransitive (A verb)

DO PO DO PO – –

6.33 6.46 2.90 5.67 6.49 1.41

.57 .53 .70 1.37 .81 .37

Ditransitive (NA verb) Monotransitive Meaningless Fig. 2. An example of a match picture, corresponding to the prime sentences The waitress gives the book to the monk/the monk the book.

Note: A = Alternating; NA = Non-alternating; PO = Prepositional Object.

DO = Double

Object;

Author's personal copy 360

I. Ivanova et al. / Journal of Memory and Language 67 (2012) 355–370

(M = 1.41, SD = .54) [t1 (11) = 3.61, p < .01; t2 (7) = 5.96, p = .001]. Secondly, to obtain a measure of whether, and how often, non-alternating verbs were used with a double object construction, 36 further participants were presented with 72 sets of three, four, or five content words (e.g., jug soldier throw pirate) printed on a sheet of paper. These participants were requested to produce, in written form, an English sentence with each set of words without adding any additional content words. Eighteen critical sets of words contained two animate entities, one inanimate entity, and a ditransitive verb which was non-alternating in half of the sequences (donate, return, display, describe, deliver, demonstrate, reveal, convey, and present) and alternating in the other half (chuck, hand, loan, offer, pass, rent, send, show, and throw). The order of these 18 sets of words was varied for each verb and for each entity (e.g., animate-inanimate-animate-verb, verb-inanimate-animateanimate). The remaining 54 sequences for all participants were fillers and equally often consisted of three, four, or five words (e.g., artist recognise bottle clown empty). Participants produced many fewer double object sentences with the sequences containing non-alternating verbs (.05) than with the sequences containing alternating verbs (.38) [Estimate = 2.10, SE = .29, z = 7.37, p < .001] (see Table 2), thus confirming that the production of double object sentences with non-alternating verbs was very strongly dispreferred in English. Note also that they produced many more responses which were neither prepositional object nor double object sentences (labeled as ‘‘Other’’, and including both prepositional-object-like sentences with a preposition other than to, and sentences that were not ditransitive) with non-alternating verbs (.25) than with alternating verbs (.11). (For the proportions of prepositional object and double object sentences produced for each individual verb, see Appendix B.)

Procedure Each of the four lists of items was administered to seven participants. Each participant was seated in front of a PC and read written instructions, which stated that there were two alternating tasks, verifying whether pictures matched a previously presented sentence, and orally describing pictures. When describing pictures, participants were requested to use the exact form of the verb written under

Table 2 Mean proportions for sentence constructions produced with alternating and non-alternating verbs. Verb Cons.

Alternating

Non-alternating

DO PO Other PO_AT PO_Other

.38 .29 .11 .16 .06

.05 .68 .25 .00 .02

Note: Cons. = construction; DO = Double Object; PO = Prepositional Object; PO_AT = sentences otherwise similar to prepositional objects but containing the preposition at; PO_Other = sentences otherwise similar to prepositional objects but containing prepositions different from to or at.

the picture. Participants were told that they had five seconds to perform the task and requested to ensure they described the picture in this time. An experimental trial had the following sequence (Fig. 3). (1) A white asterisk appeared on a black background for 700 ms. (2) The prime sentence was presented for 1500 ms, again in white on a black background. (3) A match picture surrounded by a light green frame was shown, disappearing from the screen in 2500 ms or when a key-press response was given (the color of the frame served to remind participants of the appropriate task). Participants pressed the ‘‘M’’ key if the picture matched the prime sentence, and the ‘‘N’’ key if it did not. They were asked to perform this task as quickly as they could. (4) A target picture surrounded by a pink frame appeared for 5000 ms. (5) The next trial started. Before the experiment, each participant completed six practice trials. The experiment was presented with the DMDX software (Forster & Forster, 2003), and participants spoke in a headset microphone. The program recorded participants’ spoken responses. At the end of the experimental session, participants filled in a debriefing questionnaire, to ensure they were native speakers of English and were not aware of the experimental manipulation. Scoring All participants’ responses were transcribed and coded as Prepositional Object or Double Object sentences. A description was scored as a Prepositional Object sentence if the theme of the action immediately followed the verb and was followed by the preposition to and the beneficiary. A description was scored as a Double Object sentence if the beneficiary immediately followed the verb and was followed by the theme. Responses not scored as either Prepositional Object or Double Object sentences were scored as Other. These included the trials on which participants did not produce a description at all, used a wrong verb in their descriptions, or used prepositional-object-like constructions but containing the preposition at instead of to (admitted by the target verbs throw and chuck). Data analyses We analyzed the data using logit mixed effects (LME) modeling (Baayen, 2008; Baayen, Davidson, & Bates, 2008; Jaeger, 2008). We first centered the fixed predictors, Target Verb Type and Prime Construction, assigning numerical values with a range of .5 and a mean of 0 to levels within a predictor. We then fitted a model on the data from the Non-alternating Target Verb condition, with Prime Construction as a fixed predictor. The model also included random intercepts for participants and items, and random slopes for participants. All the analyses were done in the software R (Version 2.14.0). Results Out of the 896 produced responses, 711 (79.3%) were scored as Prepositional Object sentences, 144 (16.1%) were scored as Double Object sentences, and 41 (4.6%) were scored as Other (of which eight responses [0.9%] were sentences otherwise similar to prepositional object ones but

Author's personal copy 361

I. Ivanova et al. / Journal of Memory and Language 67 (2012) 355–370

*

the waitress gives the monk the book

700 ms

1500 ms

2500 ms

5000 ms Fig. 3. An example trial.

Table 3 Mean production of Double Object descriptions in all experiments. Experiment

Prime verb

Target verb

Prime construction

Experiment 1

Alternating (Dummy) Alternating Alternating Non-alternating Alternating (Dummy) Non-alternating Alternating Non-alternating

Alternating different Non-alternating different Alternating different Alternating different Alternating different Non-alternating different Non-alternating different Non-alternating same

DO

Experiment 2 Experiment 3 Experiment 4

.31 .03 .44 .44 .28 .01 .01 .09

Priming effect PO

(135) (6) (92) (93) (120) (3) (2) (20)

– .014 (3) .34 (75) .31 (66) – .01 (3) .00 (0) .00 (0)

– .016 .10 .13 – .00 .01 .09

Note. Raw numbers are given in brackets. Same = same verb between prime and target; Different = different verb between prime and target; DO = Double Object; PO = Prepositional Object.

containing the preposition at). The number of Other responses (Non-alternating Target Verb, Prepositional Object: 12 [5.36%]; Non-alternating Target Verb, Double Object: 13 [5.8%]; Dummy condition: 16 [3.57%]) did not significantly differ across conditions [both ps > .15]. The mean proportions of Double Object target responses in each condition together with the priming effect are reported in Table 3. The priming effect is calculated as the proportion of Double Object target responses following Double Object primes minus the proportion of Double Object target responses following Prepositional Object primes (the choice of Double Object responses was arbitrary). The statistical analyses are reported in Table 4. The intercept indicates that there were many more Prepositional Object (.98) than Double Object responses. More

Table 4 LME results for Experiment 1.

Intercept Prime Construction

Estimate

SE

4.76 .67

.55 .83

z

p 8.69 .81

.1]. An examination of Table 3 suggests that there was a priming effect in both the Alternating Prime Verb and Non-alternating Prime Verb conditions. The statistical analyses are reported in Table 5. The Prime Construction predictor was significant, indicating that more Double Object responses were produced following a Double Object prime (.44) than following a Prepositional Object prime (.33). However, the interaction was not a significant predictor. Furthermore, models fitted separately on the data from each of the two Prime Verb conditions indicated that there were significantly more Double Object responses produced after a Double Object prime than after a Prepositional Object prime, both in the Alternating Prime Verb condition [Estimate = .53, SE = .24, z = 2.25, p < .05] and in the Nonalternating Prime Verb condition [Estimate = .86, SE = .26, z = 3.35, p < .001]. Responses to the debriefing questionnaires indicated that six participants (21%) noticed something wrong in the phrasing or word order of the prime sentences (although most of them attributed it to the lack of connecting words and some incorrectly reported noticing primes such as The sailor throws the ball the clown; only one participant correctly remembered an example of an ungrammatical prime). To see whether ungrammaticality awareness influenced the results, we conducted analyses without these participants. We found robust priming effects both in the Non-alternating [Estimate = .58, SE = .64, z = .91, p = .36] and the Alternating Prime Verb conditions [Estimate = .96, SE = .30, z = 3.23, p < .01]. Thus, we can conclude that awareness of the ungrammaticality by some participants did not influence our results.

363

prime sentence is ungrammatical because the prime construction uses one form of the dative alternation (double object) and the verb is consistent only with the other form (prepositional object). Lastly, our results also tie in with findings that people can quickly learn to comprehend novel or ungrammatical sentences (e.g., Kaschak & Glenberg, 2004; Luka & Barsalou, 2005). Experiment 3: primes with non-alternating verbs, targets with different non-alternating verbs In Experiment 3, we explored the effect of ungrammatical prime sentences with non-alternating verbs on the production of ungrammatical targets. In particular, we investigated whether primes such as The waitress delivers the monk the book would lead to the production of similar ungrammatical targets with different non-alternating verbs (e.g., The dancer donates the soldier the apple). This experiment provides another test of the abstract structural persistence account: In this account, priming increases the accessibility of abstract structural rules so that they are more likely to be used with the verb selected for production. Because structural persistence would be based upon the accessibility of an abstract rule, it would generalize across different verbs. Thus, in this account, priming should occur between ungrammatical primes and targets with different non-alternating verbs. Conversely, in the alternative, lexically-driven persistence account, the production of sentences with ungrammatical verb-construction combinations is based on (temporary) associations between a specific verb and a specific construction. Thus, the structural rules associated with each verb after exposure should not generalize across different verbs. Hence, in this account, no priming should occur between ungrammatical primes and targets with different non-alternating verbs. As before, if structural persistence from brief exposure does not cause at all the production of ungrammatical verb-construction combinations, we would also predict no priming.

Discussion Method We observed a priming effect for the Alternating Prime Verb condition, and, more importantly, a priming effect for the Non-alternating Prime Verb condition. These results indicate that the presence of ungrammatical double object prime sentences with non-alternating verbs did not influence the magnitude of the priming effect in any way, and hence that such primes were processed in a similar way to well-formed primes. The results also indicate that people assign syntactic structure to ungrammatical sentences which is similar to the structure they assign to grammatical sentences, in line with Ivanova et al. (2012). Importantly, here we see that this also occurs even when the

Table 5 LME results for Experiment 2. Estimate Intercept Prime construction Prime verb type Interaction

.84 .70 .13 .26

SE .37 .19 .17 .35

z

p 2.24 3.75 .73 .76