Predicting inflectional morphology from context

0 downloads 0 Views 246KB Size Report
Feb 10, 2015 - This article may be used for research, teaching, and private study purposes. ..... word would likely carry an inflectional suffix but not the.
This article was downloaded by: [University of Illinois at Urbana-Champaign] On: 12 February 2015, At: 13:03 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Language, Cognition and Neuroscience Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/plcp21

Predicting inflectional morphology from context a

Steven G. Luke & Kiel Christianson a

b

Department of Psychology, Brigham Young University, Provo, UT, USA

b

Beckman Institute for Advanced Science and Technology, University of Illinois at UrbanaChampaign, Champaign, IL, USA Published online: 10 Feb 2015.

Click for updates To cite this article: Steven G. Luke & Kiel Christianson (2015): Predicting inflectional morphology from context, Language, Cognition and Neuroscience, DOI: 10.1080/23273798.2015.1009918 To link to this article: http://dx.doi.org/10.1080/23273798.2015.1009918

PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions

Language, Cognition and Neuroscience, 2015 http://dx.doi.org/10.1080/23273798.2015.1009918

Predicting inflectional morphology from context Steven G. Lukea* and Kiel Christiansonb a

Department of Psychology, Brigham Young University, Provo, UT, USA; bBeckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Champaign, IL, USA

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

(Received 4 November 2013; accepted 9 January 2015) The present studies investigated the influence of the semantic and syntactic predictability of an inflectional morpheme on word recognition and morphological processing. In two eye-tracking experiments, we examined the effect of syntactic and semantic context on the processing of letter transpositions in inflected words. Participants experienced greater and earlier disruption from cross-morpheme letter transpositions when target verbs appeared in a context that syntactically predicted the presence of a past-tense suffix. Further, internal transpositions caused greater and earlier disruption even in monomorphemic verbs when syntactic context created an expectation of morphological complexity. No effect of semantic predictability was observed, potentially because the semantic manipulation was insufficiently strong. The results reveal that syntactic contexts typical of most English sentences can lead readers to make predictions about the morphological structure of upcoming words. Keywords: prediction; context; morphology; inflection; transposed-letter effect; eye tracking

Words like throwing, arrowhead and rancher have a complex internal structure. The word recognition system may use this structure, recognising subcomponents of the word separately (Rastle & Davis, 2008; Schreuder & Baayen, 1995; Taft & Forster, 1975). It is also conceivable that complex words are recognised holistically, with structure playing no role (Butterworth, 1983; McClelland & Patterson, 2002; Seidenberg & Gonnerman, 2000). Whether and when structure is involved in lexical access has been debated extensively. This debate has been so long-lasting because many different factors apparently influence when and how morphological structure is involved in lexical access. Most of these factors, such as frequency, semantic transparency and family size, are characteristics of the complex words themselves (see Amenta & Crepaldi, 2012, for a thorough review). It may be that morphological processing is sensitive not only to these word-internal factors but also to word-external factors. A few studies have investigated the effect of task constraints on morphological processing and have found that changes in task can lead to significant changes in complex word processing. For example, Luke and Christianson (2011) observed a strong stem frequency effect in a lexical decision task using inflected verbs. When these same verbs were embedded in sentences as part of a self-paced reading study, an interaction of wholeword and stem frequency was observed, indicating that the verbs were processed differently. Other studies have similarly observed task-dependent morphological processing (Bertram, Hyönä, & Laine, 2000; Bertram, Laine,

*Corresponding author. Email: [email protected] © 2015 Taylor & Francis

Baayen, Schreuder, & Hyönä, 2000; Hyönä, Vainio, & Laine, 2002; Taft, 2004). While these studies indicate that morphological processing is sensitive to task demands, a more important question is how morphological processing might be influenced by sentence context. Context is known to influence the processing of monomorphemic words (DeLong, Urbach, & Kutas, 2005; Fine, Jaeger, Farmer, & Qian, 2013; Laszlo & Federmeier, 2009; Luke & Christianson, 2012; Rayner & Well, 1996; Van Berkum, Brown, Zwitserlood, Kooijman, & Hagoort, 2005; Wicha, Moreno, & Kutas, 2004). Indeed, many studies show that highly constraining contexts lead the language comprehension system to make specific predictions containing precise information about upcoming words, such as animacy and gender (Otten, Nieuwland, & Van Berkum, 2007; Van Berkum et al., 2005; Wicha et al., 2004) as well as phonological (DeLong et al., 2005), orthographic (Laszlo & Federmeier, 2009) and letter position (Luke & Christianson, 2012) information. Similar predictions may also be made about the morphological structure of upcoming words. If so, then context could have a significant influence on how complex words are processed. Only a few studies have systematically investigated the influence of linguistic context on complex word processing. Most of these focused on semantic context. For example, in an eye-tracking study, Juhasz (2012) observed an effect of semantic predictability on compound word processing. This influence was strongest in the earliest

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

2

S.G. Luke and K. Christianson

reading time measures. Other studies have also investigated this issue, but the results have proven inconsistent (de Almeida & Libben, 2005; Gagné, Spalding, & Gorrie, 2005; Pollatsek, Drieghe, Stockall, & de Almeida, 2010; Zwitserlood, Bolwiender, & Drews, 2005). It is thus unclear whether predictions of morphological structure are made from semantic context. Even fewer studies have investigated the influence of syntactic context on morphological processing. An eyetracking study conducted in Hebrew by Deutsch, Frost, Pollatsek, and Rayner (2005) investigated both semantic and syntactic context effects. They found that, whereas a semantically incongruent context did not reduce facilitation from a prime previewed in the parafovea during sentence reading, a syntactically biasing context (i.e., the prime was a verb, when the syntax required a noun) eliminated any morphological priming effect, at least at later stages of lexical access. This suggests that syntactic structure might influence morphological processing, although the morphological properties of Hebrew make it difficult to generalise to other languages. The lack of research on syntactic context is particularly unfortunate when one considers inflected words, which carry morpho-syntactic information and thus could be particularly susceptible to syntactic influence. Tyler and Marslen-Wilson (1986) provide evidence that syntactic context has precisely this effect in spoken word recognition. Using a gating procedure in which different portions of inflected words were presented in syntactically and semantically constraining and non-constraining contexts, Tyler and Marslen-Wilson found that, when the syntax was strongly constraining, the full form of the word (including the suffix) could be identified at the same point that the base of the word could be identified. When syntactic constraint was weak, the full form was not identified consistently until about 200 ms after the base. On the other hand, semantic constraint moved the identification point of the base earlier but had no influence on recognition of the full form. Thus, it appears that syntactic constraint can perhaps permit predictions of an upcoming suffix, while semantic constraint may not. Of course, context, especially syntactic context, should be expected to influence the processing of inflected and derived words differently. Morphological processing is known to differ in significant ways as a function of word formation type (Bertram, Laine, & Karvinen, 1999; Tyler & Cobb, 1987). Further, while semantic context might make a particular derived or compound word predictable, syntactic context cannot; at best, syntax can lead readers to expect a noun [as in the Deutsch et al.’ s (2005) study], but that noun could be complex (shoelace) or not (hat). However, syntax can predict inflectional suffixes, which carry morpho-syntactic information. For example, There were several could be followed by almost any count noun, but the inflectional morphology predicted is

overwhelmingly likely to be the plural -s, whether or not the noun is also derivationally complex. On the other hand, There was a great deal of could be followed by almost any mass noun, but that noun could completely lack derivational morphology (e.g. sand), or be morphologically complex via many possible derivational suffixes, such as complexity, jingoism, excitement, pollution, failure, crying, denial, etc. Thus, it is more likely that predictions can be made reliably about inflection than about derivation. Inflection can also be predicted by semantic context; the presence of a time adverbial (e.g. Last week, I) predicts the past tense. Thus, inflected verbs are potentially open to the influence of both syntactic and semantic context. For this reason, the present study will focus on inflectional suffixes, as Tyler and MarslenWilson (1986) did. A further reason to focus on inflectional morphology is that in most studies investigating contextual influence on word processing the context is manipulated, so that a particular word is expected (but cf. Szewczyk & Schriefers, 2013, who showed that Polish readers appear to make predictions regarding animacy independently of word identity). This is true whether the studies focused on monomorphemic or on complex words. The use of inflected verbs here allows us to deliberately avoid this approach, focusing instead on more weakly constraining contexts where the inflection is predictable but the word itself is not. Such contexts are potentially much more common than highly constraining contexts where a particular word is expected. Indeed, if contextual influence is only felt in those rare instances when context is highly constraining, then context is a negligible influence on word recognition that only affects a small subset of words. Instead, a goal of the present study is to see if more mildly constraining contexts that predict morpho-syntactic information but not lexical identity can also influence word recognition. In order to investigate the influence of semantic and syntactic context on inflectional processing in English, the present study exploited a phenomenon known as the transposed-letter (TL) effect. Using masked priming, many studies have shown that a non-word constructed by transposing two letters of a real word (like jugde, a non-word constructed from judge) can facilitate recognition of that word compared to unrelated or orthographic controls (Forster, Davis, Schoknecht, & Carter, 1987; Perea & Lupker, 2003a, 2003b; Schoonbaert & Grainger, 2004). Transpositions involving internal letters are generally more facilitative than external-letter transpositions (e.g. judeg; see Chambers, 1979; Luke & Christianson, 2012; Perea & Lupker, 2003b). Even when words containing transpositions are presented fully visible to readers, internal transpositions are less disruptive than external ones (White, Johnson, Liversedge, & Rayner, 2008). In sum, the research on the TL effect shows that,

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

Language, Cognition and Neuroscience while external-letter transpositions are often disruptive, internal-letter transpositions are usually not, or in the case of unmasked presentation, are less so. One potential exception to the greater flexibility of internal letter position encoding involves complex words. In a masked priming naming study, Christianson, Johnson, and Rayner (2005) observed that, for compound words, masked primes containing a letter transposition within one of the constituents (e.g. sunhsine) facilitated naming when compared to substitution primes (sunsbine), consistent with previously observed TL effects. However, primes containing transpositions that crossed the morpheme boundary (e.g. susnhine) produced significantly less facilitation. The same pattern was observed for derived -er words. These findings indicate that the morpheme boundary may be privileged, in much the same way that the external letters are. Christianson et al. interpreted this effect as evidence for early morphological decomposition, consistent with other, non-TL masked priming studies (Marslen-Wilson, Bozic, & Randall, 2008; Rastle & Davis, 2003; Rastle, Davis, & New, 2004). This decrease in facilitation when transpositions occur across morpheme boundaries was partially replicated by Duñabeitia, Perea, and Carreiras (2007, 2014) in Basque and Spanish and has also been observed for inflected verbs in English (Luke & Christianson, 2013). The TL effect has also been used to investigate morphological structure in eye-tracking studies. Velan, Deutsch, and Frost (2013) presented Hebrew words containing transpositions in sentences, similar to White et al. (2008). Velan et al. observed that transpositions were more disruptive when they occurred within complex words compared to morphologically simple controls. Importantly, these differences were observed in the earliest reading time measure, suggesting that morphological structure has an early arising influence on letter-position encoding. Luke and Christianson (2013) used a similar methodology to investigate the effect of frequency on the processing of inflected English verbs. The eye-tracking experiments reported below use the morphological TL effect to investigate the influence of semantic and syntactic context on the processing of regular past-tense English verbs. This influence will be measured as change in the onset and/or magnitude of disruption caused by cross-boundary transpositions. Such a change would indicate that context can lead the language comprehension system to make predictions about the morphological structure of upcoming words. Such predictions should have an early influence on word processing. Instead of highly constraining contexts that require particular lexical items, the contexts used here were designed to lead readers to predict only that an upcoming word would likely carry an inflectional suffix but not the identity of that word. That is, they might expect that an upcoming verb will carry the past-tense -ed suffix but not

3

Table 1. Example sentences for Experiments 1 and 2. Experiment 1a – syntactic context Expected Suffix The man would have burned the letter, but – No TL his lawyer wanted him to keep it. Unexpected At the hospital, the burned patient was Suffix – No TL treated and given skin grafts. TL Morph At the hospital, the burend patient was treated and given skin grafts. TL Suffix At the hospital, the burnde patient was treated and given skin grafts. Experiment 1b – semantic context Expected Suffix Last week, the bear sniffed the campers’ – No TL backpack, which was full of snacks. Unexpected In Canada, the bear sniffed the campers’ Suffix – No TL backpack, which was full of snacks. TL Morph In Canada, the bear snifefd the campers’ backpack, which was full of snacks. TL Suffix In Canada, the bear sniffde the campers’ backpack, this was full of snacks. Experiment 2a – semantic context with plural NP Expected Suffix Last week, the bears sniffed the campers’ – No TL backpack, which was full of snacks. Unexpected In Canada, the bears sniffed the campers’ Suffix – No TL backpack, which was full of snacks. TL Morph In Canada, the beasr snifefd the campers’ backpack, which was full of snacks. TL Suffix In Canada, the beasr sniffde the campers’ backpack, this was full of snacks. Experiment 2b – syntactic context with monomorphemic verbs Expected Suffix The worried parents attend the school – No TL board meetings about the budget problems. Unexpected The parents will attend the school board Suffix – No TL meetings about the budget problems. TL Internal The parents will attned the school board meetings about the budget problems. TL Final The parents will attedn the school board meetings about the budget problems. Note: The target words are in italics, and the transposed letters are in bold; target words appeared in normal font in the experiments.

that it will be wilted. The particular contextual manipulations used are shown in Table 1. The primary question under investigation is whether participants will be more sensitive to cross-boundary letter transpositions (i.e. the transpositions will cause more and earlier disruption) if they are led by context to expect such a boundary.

Experiment 1a Experiment 1a explored the effect of syntactic context on morphological processing. If readers use context to predict particular suffixes, they might look for the internal structure of a word sooner when such predictions are possible. If this is the case, cross-morpheme transpositions

4

S.G. Luke and K. Christianson

(TL Morphs) should be disruptive earlier when a suffix is expected than when it is not.

Method Participants

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

Thirty-six people from the University of Illinois at Urbana-Champaign community participated. Most participants were recruited from the Educational Psychology subject pool. All had normal or corrected-to-normal vision and received either course credit or $7. Apparatus Eye movements were recorded via an SR Research Eyelink 1000 eye tracker (spatial resolution of 0.01°) sampling at 1000 Hz. Text was displayed in 12-point Courier New font. Participants sat 69 cm away from a 20inch monitor, so that approximately 3.5 characters subtended 1° of visual angle. Chin and head rests minimised head movements. Although viewing was binocular, eye movements were recorded from the right eye. Materials Experiment 1a used 42 regular verbs carrying an -ed suffix. The verbs were carefully evaluated to ensure that they could be used in sentences both as main verbs and to modify nouns. For each of the 42 verbs, two sentence frames were constructed: a version in which a suffix was expected and a version in which a suffix was unexpected (see Table 1). In the Expected Suffix condition, the words would have preceded the target verb. In the Unexpected Suffix condition, the verb appeared as a modifier of a noun. In these structures, the verbs were not technically past tense but were rather past participle forms. These words are still complex, however, and while nominal modifiers will likely be processed more slowly than verbs, no differences in the processing of the morphological structure should be expected unless, as we hypothesise, context leads readers to make predictions about word structure. To test the strength of this context manipulation, 20 native speakers of English completed a sentence fragment completion task. These participants did not participate in any of the experiments reported here. Participants read target sentences up to but not including the target word and completed the sentence in a way that makes sense and creates a grammatically correct sentence. Half of the target sentences each participant saw were the Expected Suffix version, and the other half were not (order was randomised), and this was counterbalanced across participants. Participants completed the sentences with a word with an -ed suffix 87% of the time in the Expected Suffix condition, but only 1.4% of the time in the Unexpected Suffix condition (p < .0001). This indicates that readers should have a much greater expectation of

morphological structure in the Expected Suffix than in the Unexpected Suffix condition. Participants produced the target verbs only about 1% of the time, indicating that, while participants would have a greater expectation of morphological structure in the Expected Suffix condition, they would not be expecting particular verbs. Each verb appeared in one of three TL conditions: the target verb contained either no transposition (No TL), a cross-morpheme transposition (TL Morph), or a transposition of the word-final suffix letters (TL Suffix). This yielded a 2 (Context: Expected Suffix vs. Unexpected Suffix) × 3 (TL Type: No TL, TL Morph, TL Suffix) within-subjects and within-items design. Table 1 shows an example of these manipulations. The No TL condition should produce no disruption. The TL Suffix condition, which involves word-final letters, should produce significant disruption, and the TL Morph should not produce significant disruption unless readers expect complexity, in which case it should be comparable to the TL Suffix condition (Luke & Christianson, 2013; White et al., 2008). Sentences were counterbalanced in a Latin Square design across six lists. In addition to the 42 sentences from Experiment 1a, each list contained the 42 items from Experiment 1b and 84 filler items with other past-tense verbs for a total of 168 sentences. Procedure Participants were told that many of the sentences they would read contained misspelled words, and that the purpose of the experiment was to see how well people are able to read and understand sentences containing such errors. Participants were further told that the misspelled words should still be understandable, and that they should try to read the sentences as normally as possible and to understand them. Each trial began with a gaze trigger, a black circle presented in the position of the first character of the text. Once a stable fixation was detected on the trigger, the sentence was presented in full. The participant pressed a button on a standard game controller to indicate that s/he had finished reading the sentence. The sentence then disappeared. After this, a question about the sentence sometimes appeared (after 1/3 of trials), which participants answered by pressing Yes or No on the controller. Then the next trial began. Sentence order was randomised for each participant. Results Before analysis, the data were trimmed, with fixations 800 ms excluded (7% of fixations). Trials where the comprehension question was answered incorrectly and trials where participants reported not being familiar with the target word in a post-session questionnaire were also excluded (5% of trials).

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

Language, Cognition and Neuroscience Three different reading time measures for the target words were analysed: first fixation duration, gaze duration and total time. First fixation duration, the duration of the very first fixation the eyes make on the target word, reflects the earliest moments of foveal processing. This measure should reveal any early implicit contextual effects. Gaze duration, the sum of all fixations on the target word before the eyes move away from it, reflects a more intermediate processing stage. Finally, total time is the sum of all fixations on the target word, including rereading, and so reflects later stages of lexical access and integration. Descriptive statistics for each reading time measure are shown in Table 2. Each reading time measure was analysed separately using linear mixed models (Baayen, Davidson, & Bates, 2008) in R (R Development Core Team, 2012), using the lme4 package (Bates, Maechler, & Bolker, 2011). The primary effect of interest is whether syntactic context can create an expectation of morphological complexity. This expectation should in turn influence the onset and amount of disruption caused by the different transposition conditions, especially the TL Morph condition. For this reason, the Expected Suffix vs. Unexpected Suffix conditions were compared directly within each transposition condition. These comparisons were performed by using contrast coding to create a set of planned comparisons that compared each Expected Suffix TL condition to its Unexpected Suffix counterpart (e.g. Expected Suffix TL Morph was compared to Unexpected Suffix TL Morph). These contrasts will show in which TL conditions an effect of Context is observed. Additionally, two contrasts that collapse across Context conditions to explore the TL effect were coded: reading times in all words containing transpositions were compared to the No TL condition, and the TL Morph condition was compared with the TL Suffix condition. Trial Number was also included in the models when it interacted with the Context contrasts to reveal strategic effects arising throughout the course of the experiment. Mixed-effects models do not test for main effects and interactions in the same way as analyses of variance. They

5

skip directly to testing simple effects or, as in this case, planned comparisons. When a significant interaction is reported, this indicates that the effect in a second condition differed statistically in magnitude from whatever simple effect was observed in the first condition. A simple sum of the coefficients gives the estimated coefficient in the second condition. Models were fitted using a stepwise selection procedure. Random effects structures were fitted using likelihood ratio tests. Random intercepts for Participant and Item were included, but no random slopes contributed to the models (all ps > .99). Only effects that were significant (i.e. p < .05) were retained in the model. p-values were obtained using Markov chain Monte Carlo sampling. Response times were log transformed. First fixation duration Participants fixated words containing transpositions longer than untransposed controls (coeff. = −0.1, SE = 0.02, t = −4.02, p < .0001) and TL Suffix words longest of all (coeff. = −0.058, SE = 0.02, t = −2.72, p = .0065). They also read words in the Expected Suffix conditions faster than words in the Unexpected Suffix conditions (both ts > 2.04, both ps < .05) in both the No TL and TL Suffix conditions but not in the TL Morph condition, where the two context conditions were equal (p > .75). Figure 1 reveals the reason for this pattern of results: in the Expected Suffix condition, the TL Morph and TL Suffix conditions were both disruptive, and equally so, relative to the No TL control condition, whereas in the Unexpected Suffix condition only the transposition of the final letters was disruptive. In other words, when morphological complexity was expected, TL Morphs were disruptive, but when complexity was not expected, TL Morphs were not disruptive, indicating that the target word was processed more like a monomorphemic word (cf. White et al., 2008). The model was refit using different contrasts, and the results confirmed that the TL Suffix condition was not statistically different from the TL Morph condition in the Expected Suffix condition (t < 1.05, p > .29), but that there was a significant difference between them in the

Table 2. Means (and standard deviations) in milliseconds for the three dependent measures in Experiment 1a.

First fixation duration Expected Suffix (would have burned) Unexpected Suffix (the burned patient) Gaze duration Expected Suffix (would have burned) Unexpected Suffix (the burned patient) Total time Expected Suffix (would have burned) Unexpected Suffix (the burned patient) Note: Transposed letters are in italics for emphasis.

No TL (burned)

TL Morph (burend)

TL Suffix (burnde)

221.08 (69.84) 232.87 (85.23)

238.91 (94.92) 237.52 (83.72)

244.22 (83.88) 261.66 (102.72)

267.56 (123.69) 298.48 (140.42)

318.73 (191.15) 367.34 (249.78)

347.54 (191.29) 434.78 (253.35)

395.52 (237.87) 469.91 (307.71)

510.30 (436.7) 634.27 (546.22)

509.42 (378.07) 677.84 (521.55)

6

S.G. Luke and K. Christianson

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

(both ts < 0.76, both ps > .43). This again indicated that in the first half of the experiment the TL Morph condition was as disruptive as the TL Suffix condition in the Expected Suffix context (t = 0.81, p > .41) but not in the Unexpected Suffix condition (t = 4.07, p < .0001). As with gaze duration, the two TL conditions were equivalent in both context conditions in the second half of the experiment (both ts < 0.07, both ps > .94). This result confirms the findings from gaze duration that participants adapted to the context manipulation as the experiment progressed. Figure 1. Mean first fixation durations in Experiment 1a. Error bars represent standard error.

Unexpected Suffix condition (t = 2.85, p = .0044). Thus, it appears that participants detected the morphological structure of the inflected words more quickly when syntactic context led them to expect an inflected suffix. Gaze duration Again, participants spent more time on words containing transpositions than on untransposed controls (coeff. = −0.27, SE = 0.035, t = −7.8, p < .0001) and TL Suffix words longest of all (coeff. = −0.13, SE = 0.03, t = −4.42, p < .0001). Expected Suffix targets were read more quickly in all TL conditions (all ts > 2.61, all ps < .01). However, Trial Number interacted with the context difference in the TL Morph condition (coeff. = −0.0019, SE = 0.0009, t = −2.18, p = .029) but not in the other conditions (both ts < 1.5, both ps > .13). Follow-up tests revealed that, in the first half of the experiment, gaze duration showed the same pattern as first fixation duration; the TL Morph condition was as disruptive as the TL Suffix condition in the Expected Suffix context (t = 1.1, p > .27) but not in the Unexpected Suffix condition (t = 4.81, p < .0001). In the second half of the experiment, the influence of context had disappeared; the two TL conditions were equivalent in both context conditions (both ts < 1.48, both ps > .13). This suggests that participants adapted to the context manipulation as the experiment progressed. Total time The pattern of results for total time was similar to the pattern for gaze duration. Again, participants spend more time on words containing transpositions than on untransposed controls (coeff. = −0.33, SE = 0.04, t = −8.11, p < .0001) and TL Suffix words longest of all (coeff. = −0.076, SE = 0.03, t = −2.19, p = .029). Expected Suffix targets were read more quickly in all TL conditions (all ts > 3.87, all ps < .0001). Again, Trial Number interacted with the TL Morph contrast (coeff. = −0.002, SE = 0.001, t = −1.91, p = .056) but not with the other conditions

Frequency analysis To further confirm the observation that readers expected morphological complexity in the Expected Suffix but not in the Unexpected Suffix condition, the first fixation and gaze duration data were reanalysed to look for the onset of stem frequency effects in the two conditions. As these items were not selected specifically with a frequency analysis in mind, the stem and whole-word frequencies were highly collinear. This means that any frequency effect or interaction of effects observed for these items might be difficult to interpret, and so the analyses were restricted to investigating whether stem frequency had any effect on reading times over and above whole-word frequency. This was accomplished by residualising stem frequency on whole-word frequency. The data were analysed using linear mixed models, with Context and residual Stem Frequency as fixed effects and Participant and Item as random effects. For first fixation duration, the effect of residual Stem Frequency was significant in the Expected Suffix condition (Estimate = −0.039, SE = 0.014, t = −2.77, p = .0056), but a significant interaction with Context (Estimate = 0.043, SE = 0.02, t = 2.19, p = .029) indicated that there was no effect of residual Stem Frequency in the Unexpected Suffix condition. For gaze duration, residual Stem Frequency was not significant in the Expected Suffix condition (t < −0.03), but an interaction with Context (Estimate = −0.06, SE = 0.029, t = −2.09, p = .037) indicated that residual Stem Frequency was a significant predictor of reading times in the Unexpected Suffix condition. The fact that stem frequency effects appeared at first fixation in the Expected Suffix condition but not until gaze duration in the Unexpected Suffix condition supports the conclusion that the morphological structure of the word was recognised later in the Unexpected Suffix condition. Discussion Experiment 1a used TL effects to investigate the influence of syntactic context on readers’ sensitivity to morpho‐ logical structure. At first fixation, participants were disrupted by transpositions across the morpheme boundary only when the presence of a suffix was syntactically

Language, Cognition and Neuroscience predictable. This general pattern was also evident early on in the experiment for the gaze duration and total time measures, and indicates that syntactic context can lead readers to predict that an upcoming word will be inflected.

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

Experiment 1b Experiment 1b explored whether semantic context informs readers’ expectations about the morphological structure of upcoming words. The same transposition conditions were used as in Experiment 1a, but semantic context was manipulated by adding a time adverbial (e.g. ‘Last week,’) to create the expectation of a past-tense suffix (see Table 1).

7

The transposition conditions used in Experiment 1b were the same as those used in Experiment 1a. This yielded a 2 (Context: Expected Suffix vs. Unexpected Suffix) × 3 (TL Type: None, TL Morph, TL Suffix) design. Thus, 33% of the sentences did not contain any transposition, while the other 66% did contain transpositions. An example of these manipulations can be found in Table 1. The sentences were counterbalanced in a Latin square design across six lists. In addition to the 42 sentences from Experiment 1b, each list contained the 42 items from Experiment 1a and 84 filler items for a total of 168 sentences. Apparatus and procedure The apparatus and procedure were the same as in Experiment 1a.

Method Participants The participants were the same as in Experiment 1a. Materials Experiment 1b used 42 regular past-tense verbs as targets. None of these verbs were used in Experiment 1a. For each of the 42 verbs, a sentence was constructed with two versions. The Expected Suffix version begins with an adverbial phrase pointing to the past, such as yesterday or last year. The Unexpected Suffix version of the sentence begins with a prepositional or other phrase that conveys no time-related information. These sentences were normed as described in Experiment 1a to test the strength of the predictability manipulation. Participants completed the sentences using a past-tense verb about 88% of the time, indicating that readers tend to expect the past-tense form (and thus a suffixed verb) to follow a singular noun phrase (NP). Participants were significantly more likely to complete the sentence with a past-tense or other suffixed verb when the sentence was preceded by a time adver‐ bial (Expected Suffix condition: 96% vs. 80% in the Unexpected Suffix condition; p < .0001). Participants produced the actual target verbs less than 1% of the time.

Results Data were trimmed and analysed as described in Experiment 1. About 7% of fixations were excluded, along with about 5% of trials due to incorrect comprehension question answers. Descriptive statistics can be found in Table 3. First fixation duration The No TL condition was faster than the TL conditions (coeff. = −0.11, SE = 0.03, t = −4.13, p < .0001), and the TL Morph condition was faster than the TL Suffix condition (coeff. = −0.054, SE = 0.02, t = −2.5, p = .012). No difference between the Context conditions was evident in any TL condition (all ps < .44). Gaze duration The No TL condition was faster than the TL conditions (coeff. = −0.37, SE = 0.06, t = −6.02, p < .0001), and the TL Morph condition was faster than the TL Suffix condition (coeff. = −0.21, SE = 0.05, t = −3.96, p < .0001). No effect of Context was evident (all ps > .12).

Table 3. Means (and standard deviations) in milliseconds for the three dependent measures in Experiment 1b.

First fixation duration Expected Suffix (Last week, the bear sniffed) Unexpected Suffix (In Canada, the bear sniffed) Gaze duration Expected Suffix (Last week, the bear sniffed) Unexpected Suffix (In Canada, the bear sniffed) Total time Expected Suffix (Last week, the bear sniffed) Unexpected Suffix (In Canada, the bear sniffed) Note: Transposed letters are in italics for emphasis.

No TL (sniffed)

TL Morph (snifefd)

TL Suffix (sniffde)

240.19 (88.06) 238.26 (78.85)

254.68 (101.78) 253.64 (89.18)

269.15 (108.13) 274.09 (111.57)

296.19 (136.52) 286.10 (120.84)

361.68 (225.49) 341.47 (166.56)

377.95 (190.74) 385.39 (196.2)

401.81 (243.68) 405.29 (224.3)

504.58 (349.51) 469.75 (258.62)

528.43 (387.73) 510.59 (275.68)

8

S.G. Luke and K. Christianson

Total time The No TL condition was faster than the TL conditions (coeff. = −0.3, SE = 0.07, t = −4.22, p < .0001), and the TL Morph condition was faster than the TL Suffix condition (coeff. = −0.18, SE = 0.06, t = −2.81, p = .005). No effect of Context was evident (all ps > .6).

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

Discussion Experiment 1b investigated whether the semantic context can make an upcoming suffix predictable. Even though the norming data indicated that the contextual manipulation significantly increased expectation of the past tense, no effects involving Context were significant. Instead, TL Morphs were disruptive in both Context conditions starting at first fixation, matching the results for the Expected Suffix condition in Experiment 1a. This pattern suggests that the expectation of complexity was strong for all verbs in all conditions and is consistent with the norming results, which showed that participants still had a strong expectation of a past-tense verb even in the Unexpected Suffix condition. These findings suggest that a singular NP preceding a verb is sufficient to establish an expectation that the verb will be suffixed. Thus, the results of Experiment 1b provide further evidence that syntactic context influences morphological processing but are inconclusive regarding semantic context. It may be that, if the influence of syntactic context were further reduced, effects of semantic context will appear. Experiment 2a tests this possibility.

sniffed became the bears sniffed; see Table 1). Changing the NP to plural allows the verb to legally appear without a suffix (e.g. the bears sniff), which should decrease the syntactic expectation of morphological complexity, thereby potentially permitting effects of semantic context on morphological processing to appear. These sentences were re-normed, using a different set of participants. Participants completed the sentences with a past-tense word 100% of the time in the Expected Suffix condition, but only 67% of the time in the Unexpected Suffix condition (p < .001). Thus, changing the NPs to plural maintained the high expectation of complexity in the Expected Suffix Condition but reduced it somewhat in the Unexpected Suffix condition (from 80% in Experiment 1b). Participants produced the target verbs less than 1% of the time. The transposition conditions in Experiment 2a were the same as those in Experiments 1–2. This yielded a 2 (Context: Expected Suffix vs. Unexpected Suffix) × 3 (TL Type: None, TL Morph, TL Suffix) design. Thus, 33% of the sentences did not contain any transposition, while 66% did contain transpositions. The sentences were counterbalanced in a Latin square design across six lists. In addition to the 42 sentences from Experiment 2a, each list contained the 42 items from Experiment 2b and 42 filler items containing transpositions in derived words for a total of 126 sentences. Apparatus and procedure The apparatus and procedure were the same as in Experiment 1a.

Experiment 2a

Results

In Experiment 2a, the items from Experiment 1b were changed, so that all of the NPs were plural instead of singular nouns. As a plural noun can be followed by an unsuffixed verb in English (e.g. the squirrels climb), changing the NPs to plural should reduce the expectation that the following verb will carry a suffix, potentially revealing effects of semantic context on morphological processing that were overshadowed by the overall strong expectation of complexity in Experiment 1b.

Data were trimmed and analysed as described in Experiment 1a. About 6% of fixations were excluded, along with 4% of trials due to incorrect comprehension questions answers. Descriptive statistics are found in Table 4.

Participants Forty-eight people from the University of Illinois at Urbana-Champaign community participated. Most participants were recruited from the Educational Psychology subject pool. All had normal or corrected-to-normal vision and received either course credit or $7. Materials In Experiment 2a, the 42 items from Experiment 1b were changed, so that the NP was a plural noun (e.g. the bear

First fixation duration Participants fixated words containing transpositions longer than untransposed controls (coeff. = −0.037, SE = 0.008, t = −4.02, p < .0001), but unlike in Experiment 1b, no difference between the TL Morph and TL Suffix conditions was observed (p > .51). As in Experiment 1b, no difference between the Context conditions was evident in any transposition condition (all ps > .36). Gaze duration Participants fixated words containing transpositions longer than untransposed controls (coeff. = −0.082, SE = 0.01, t = −7.95, p < .0001). Unlike in first fixation duration, the TL suffix condition was now more disruptive than the TL Morph condition (coeff. = −0.031, SE = 0.011, t = −2.7, p = .0069). Again, no effect of Context was evident (all ps > .14).

Language, Cognition and Neuroscience

9

Table 4. Means (and standard deviations) in milliseconds for the three dependent measures in Experiment 2a.

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

First fixation duration Expected Suffix (Last week, the bears sniffed) Unexpected Suffix (In Canada, the bears sniffed) Gaze duration Expected Suffix (Last week, the bears sniffed) Unexpected Suffix (In Canada, the bears sniffed) Total time Expected Suffix (Last week, the bears sniffed) Unexpected Suffix (In Canada, the bears sniffed)

No TL (sniffed)

TL Morph (snifefd)

TL Suffix (sniffde)

250.82 (92.57) 246.63 (88.21)

271.48 (106.18) 269.46 (109.17)

276.48 (119.46) 268.74 (118.76)

325.04 (155.56) 308.40 (138.04)

385.72 (238.5) 383.86 (217.05)

414.65 (251.79) 405.25 (268.07)

713.73 (477.56) 663.26 (455.52)

550.17 (457.13) 531.54 (356.96)

599.05 (446.44) 614.75 (558.99)

Note: Transposed letters are in italics for emphasis.

Total time Suffix transpositions were again more disruptive than morpheme boundary transpositions (coeff. = −0.045, SE = 0.014, t = −3.27, p = .0011), but unlike in the other reading measures, total time was shorter in these conditions than in the No TL condition (coeff. = 0.09, SE = 0.012, t = 7.33, p < .0001), perhaps because participants spent more time on the TL targets during their initial reading, so they were less likely to re-read these words later. Again, no effect of Context was evident (all ps > .1). Discussion Experiment 2a investigated whether the semantic predictability of the suffix influences readers’ expectations about morphological complexity. Even though the norming data indicated that the influence of syntactic context was weakened by the introduction of plural NPs, semantic predictability still did not appear to influence morphological processing. Instead, transpositions across the morpheme boundary were disruptive starting at first fixation, suggesting that no semantic effects were observed because the influence of syntactic context was still too strong. Together with Experiment 1b, these findings suggest that any NP, even a plural one, constitutes sufficient syntactic context to establish the expectation of a suffix on the upcoming verb. Experiment 2b Experiment 1a explored whether syntactic context can produce specific predictions about the structure of upcoming verbs. Context clearly had an effect; TL Morphs caused earlier disruption when syntactic context made a suffix predictable. Thus, when syntax predicts morphological complexity, readers recognise it more quickly when it is present. Experiments 1b and 2a suggest that simply encountering a subject NP while reading is sufficient to create an expectation of complexity in the upcoming verb. In all these previous experiments, context led participants

to expect a complex word and then they encountered a complex word. However, if truly top-down predictions about morphological structure are generated from context, then contexts should influence word processing even when the word that is expected to be complex turns out not to be complex. Experiment 2b explored the consequences of expecting complexity where none exists. To accomplish this, we employed sentences with plural NPs, which permit the use of uninflected base forms of verbs (e.g. the parents attend; see Table 1). Experiment 2a showed that even with plural NPs there is still a strong early expectation of morphological complexity; TL Morphs were as disruptive as word-final transpositions starting at the first fixation. To eliminate this expectation, versions of each sentence were created that included the word will after the NP (e.g. the parents will attend). In this context, only the infinitival uninflected form of the verb is licensed, so there should be no expectation of complexity. If internal transpositions are more disruptive, relative to controls, when morphological complexity is expected even when no complexity is present, this will provide strong evidence that top-down predictions of morphological structure are generated from syntax. Participants The participants were the same as in Experiment 2a. Materials Experiment 2b used 42 regular uninflected verbs. None of these verbs were used in Experiment 2a. For each of the 42 verbs, two sentence frames were constructed: a version in which context would cause participants to expect a suffix and a version in which no suffix was expected. In the Expected Suffix condition, the verb appeared immediately after the plural NP. Experiment 2a showed that this structure was sufficient to create an expectation of complexity, although the plurality of the NP makes it possible to use non-complex unsuffixed verbs. Note that,

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

10

S.G. Luke and K. Christianson

in this experiment, the expectation of an upcoming suffix is misleading, as the upcoming verb does not carry a suffix. Thus, if internal transpositions are disruptive in this condition, it cannot be because of the word itself – internal transpositions in monomorphemic words are not typically disruptive, especially in the early stages of word recognition (White et al., 2008) – so the disruption must be attributable solely to the influence of context. In the Unexpected Suffix condition, the word will preceded the target verb, so that only the unsuffixed infinitive verb form could be expected to follow. These sentences were normed to test the strength of the context manipulation, as described for Experiment 2a. Participants completed the sentences with a word in the past tense and/or carrying the -ed suffix 60% of the time in the Expected Suffix condition, but never (0%) in the Unexpected Suffix condition (p < .0001). This indicates that readers should have a much greater expectation of morphological structure in the Expected Suffix than in the Unexpected Suffix condition. Participants produced target verbs less than 1% of the time. The transpositions used in Experiment 2b were in the same letter positions as those used in previous experiments. However, since the verbs in Experiment 2b were uninflected, the transpositions did not disrupt a morpheme boundary or suffix. This yielded a 2 (Context: Expected Suffix vs. Unexpected Suffix) × 3 (TL Type: None, TL Internal, TL Final) design. The TL Internal condition is analogous to the TL Morph condition from the previous experiments, and the TL Final condition to the TL Suffix condition. As in previous experiments, 33% of the sentences did not contain any transposition, while 66% did contain transpositions. See Table 1 for examples. Sentences were counterbalanced in a Latin square design across six lists. In addition to the 42 sentences from Experiment 2b, each list contained the 42 items from Experiment 2a and 42 filler items, making 126 sentences total. Apparatus and procedure The apparatus and procedure were the same as in Experiment 1a.

Results Data were trimmed and analysed as described in Experiment 1a. About 6% of fixations were excluded, along with 4% of trials due to incorrect comprehension question answers. Descriptive statistics are found in Table 5. First fixation duration A significant effect of Context was observed in all TL conditions, with shorter fixations in the Unexpected Suffix condition when the infinitive form was morpho-syntactically required (all ts > 3.22, all ps < .005). The model coefficients reveal that this difference was greatest in the TL Internal condition (0.086 vs. 0.042 and 0.057 for the No TL and TL Final conditions, respectively). The different condition means are shown in Figure 2. This figure shows clearly that, in the Unexpected Suffix condition, when context did not permit an inflected verb, the internal transposition was not disruptive. In the Expected Suffix condition, when morphological complexity was expected, the internal transposition was as disruptive as the TL Final condition. To confirm this statistically, the model was refit using different contrasts, and the results confirmed that the TL Final condition was significantly different from the TL Internal condition in the Unexpected Suffix condition (coeff. = 0.027, SE = 0.013, t = 2.12, p = .034), but that there was no difference between them in the Expected Suffix condition (t < .18). These results show that participants were more disrupted by an internal transposition when they expected that transposition to cross a morpheme boundary, even when no boundary was present. Additionally, fixation durations were shorter when No TL was present (coeff. = −0.032, SE = 0.008, t = −4.07, p < .0001), but the difference between the two TL conditions was not significant when collapsing across predictability (t = −1.38, p = .17). Gaze duration Participants spent more time on words containing transpositions than on untransposed controls (coeff. = −0.093, SE = 0.011, t = −8.44, p < .0001) and TL Final words

Table 5. Means (and standard deviations) in milliseconds for the three dependent measures in Experiment 2b.

First fixation duration Expected Suffix (parents attend) Unexpected Suffix (parents will attend) Gaze duration Expected Suffix (parents attend) Unexpected Suffix (parents will attend) Total time Expected Suffix (parents attend) Unexpected Suffix (parents will attend) Note: Transposed letters are in italics for emphasis.

No TL (attend)

TL Internal (attned)

TL Final (attedn)

239.56 (74.79) 220.78 (75.56)

268.96 (101.75) 224.76 (81.32)

268.43 (106.46) 238.27 (90.53)

291.92 (140.33) 262.67 (122.01)

368.15 (233.07) 297.43 (206.3)

384.58 (225.16) 374.25 (247.07)

814.67 (495.37) 779.44 (503.52)

559.71 (421.79) 490.93 (434.92)

608.38 (475.21) 595.12 (590.88)

Language, Cognition and Neuroscience

11

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

General discussion

Figure 2. Mean first fixation durations in Experiment 2b. Error bars represent standard error.

longest of all (coeff. = −0.061, SE = 0.012, t = −4.96, p < .0001). Unexpected Suffix targets were read more quickly in the No TL condition (coeff. = −0.055, SE = 0.018, t = −3.02, p = .0025) and in the TL Internal condition (coeff. = −0.12, SE = 0.018, t = −6.56, p < .0001) but only marginally in the TL Final condition (coeff. = −0.032, SE = 0.018, t = −1.83, p = .07). There was a greater effect of Context in the TL Internal condition than in the other TL conditions (TL Internal vs. TL Final in Unexpected Suffix condition: t = 5.86, p < .0001; in Expected Suffix condition: t < 1.14). Total time The effect of Context was only present in the TL Internal condition (coeff. = −0.096, SE = 0.022, t = −4.26, p < .0001; in other conditions both ts < 1.62, both ps > .1), with longer total reading times in the Expected Suffix condition (TL Internal vs. TL Final in Expected Suffix condition: t = 4.25, p < .0001; in Unexpected Suffix condition: t < 1.62). Participants spent less time on words containing transpositions than on untransposed controls (coeff. = 0.2, SE = 0.014, t = 14.41, p < .0001) and more time on TL Final words than on TL Internal words (coeff. = −0.066, SE = 0.016, t = −4.16, p < .0001).

Discussion Experiment 1a showed that when syntactic context predicts morphological complexity, readers show greater and earlier sensitivity to TL Morphs. Experiment 2b showed that when readers expect morphological complexity, they are more disrupted by internal transpositions, even when those transpositions do not cross a morpheme boundary because there is no boundary present. Together with the previous experiments, these results reveal clear top-down effects of syntactic context on morphological processing and letter-position encoding, suggesting that readers use syntax to make predictions about the inflectional structure of upcoming words.

The present experiments show that syntactic context affects morphological processing. When readers expected a complex word, they were disrupted by internal crossmorpheme transpositions but when syntax did not predict morphological complexity, cross-morpheme transpositions produced no disruption. This effect appeared in first fixation duration in Experiment 1a, suggesting that syntax has an early influence on morphological processing (cf. Deutsch et al., 2005). Experiment 2b showed that internal transpositions were more disruptive when a complex word was expected, even when the transposition occurred within a monomorphemic word and thus did not cross a morpheme boundary. These results suggest that readers can use syntactic context to predict the presence of a suffix on an upcoming verb, and that this prediction influences early processing of that verb. The results of Experiment 2b suggest that this effect of syntactic context is truly top-down and not contingent on the presence of a morpheme boundary. No evidence of a semantic influence on morphological processing was observed in Experiment 1b or 2a. This may be because morphological processing is insensitive to semantic context, but there are reasons to be cautious about such an interpretation. The sentence completion task results for the items in Experiment 1b revealed that the semantic manipulation, although significant, was not particularly strong, primarily because participants had a strong default tendency to expect a past-tense verb. Pluralising the subject NP (Experiment 2a) weakened the expectation of the past tense somewhat, but this did not appreciably change the results; participants still appeared to expect an inflected verb, perhaps because most of the verbs encountered during the experiment were inflected. Thus, it appears that the strong effect of syntactic context overrode any effects of semantics. The verbs in Experiments 1b and 2a all appeared as main verbs immediately following the sentential subject, so readers almost certainly were unsurprised by the appearance of a verb, which in English typically carries a suffix (if it is regular), and always so when the NP is singular. Cross-morpheme transpositions were disruptive for first fixation duration in Experiments 1b and 2a, as they were in the Expected Suffix but not the Unexpected Suffix condition in Experiment 1a, suggesting that, at least in English, an NP wanting a verb is sufficient syntactic context to predict the presence of a suffix. Experiment 2b showed that an NP was indeed sufficient context to create an expectation of verb complexity, as evidenced by an increased sensitivity to internal letter transposition. In recent years, many researchers have suggested that the language comprehension system actively predicts upcoming words. Studies have shown that highly constraining semantic contexts can lead readers or listeners to

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

12

S.G. Luke and K. Christianson

make specific, detailed predictions about upcoming words (DeLong et al., 2005; Fine et al., 2013; Laszlo & Federmeier, 2009; Luke & Christianson, 2012; Van Berkum et al., 2005; Wicha et al., 2004). However, for most words in most sentences, semantic context is mildly constraining at best. If useful predictions can only be made when contexts are highly constraining then prediction is likely not a central feature of the language comprehension system. Szewczyk and Schriefers (2013) observed that semantic contexts that are not sufficiently constraining to require a particular lexical item can still lead to precise predictions about the animacy of the upcoming word. Their findings suggest that semantic context provides much more opportunity for prediction than might be supposed from previous studies. The findings of the present study indicate that syntactic context is similarly useful; participants were able to use syntactic context to make highly specific predictions about the presence of particular suffixes even when the upcoming word itself was not predictable. Further, the level of syntactic constraint employed in these experiments occurs in almost every English sentence. This means that the influence of syntactic context on word recognition may be quite pervasive, even ubiquitous, permitting readers to make morpho-syntactic predictions that guide word recognition. Combined with the findings of Szewczyk and Schriefers, these results indicate that sentential context provides a wealth of information that the language comprehension system can exploit to predict upcoming words. If, as these data suggest, prediction is not limited to situations where the sentence is unusually constraining, then the language comprehension system can gain significant benefit from actively predicting features of upcoming words. The fact that morphological processing is sensitive to linguistic context and task requirements has important implications for complex word recognition research. Much of the literature on morphological processing consists of lexical decision/masked priming studies in which target words are presented in isolation. The present results, along with previous research showing context- and task-based differences in morphological processing (Bertram, Hyönä, et al., 2000; Bertram, Laine, et al., 2000; Juhasz, 2012; Luke & Christianson, 2011), should lead researchers to be cautious about relying exclusively on such methods, at least when studying inflection. Studying complex word recognition in context has certain advantages. One is that deliberate control of context may help clarify certain inconsistent findings in the literature and may help to adjudicate between different points of view about morphological processing (see Introduction). Another is that studying word processing in context would make it much simpler to link complex word recognition research to other aspects of language processing and reading, such as

syntactic parsing and eye-movement control. In particular, situating words in contexts is the only way to determine what role expectation (e.g. Fine et al., 2013; Levy, 2008) plays in morphological processing in particular and visual word recognition in general. These same concerns apply to all visual word recognition research, not just research with complex words. With regard to the TL effect, current models of visual word recognition explain the TL effect, and word recognition generally, in largely bottom-up terms. Luke and Christianson (2012), however, presented evidence that a highly constraining semantic context eliminates the TL effect, suggesting that context can influence even the earliest stages of word recognition. The present results provide further evidence that the transposability of some letters is contingent on top-down influence; in both Experiments 1a and 2b, the expectations created by the previous syntactic structure made participants more sensitive to the positions of internal letters. These results suggest strongly that morpho-syntactic expectations generated online can prompt readers to attend more and earlier to internal letters, thereby making internal letter transpositions much more disruptive. Under these circumstances, it seems more important than ever for visual word recognition researchers to study how word recognition interacts with the broader context of reading. Disclosure statement No potential conflict of interest was reported by the authors. References Amenta, S., & Crepaldi, D. (2012). Morphological processing as we know it: An analytical review of morphological effects in visual word identification. Frontiers in Psychology, 3, 232. doi:10.3389/fpsyg.2012.00232 Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixedeffects modeling with crossed random effects for subjects and items. Journal of Memory and Language, 59, 390–412. doi:10.1016/j.jml.2007.12.005 Bates, D., Maechler, M., & Bolker, B. (2011). Lme4: Linear mixed-effects models using S4 classes. R package (Version 0.999375-42) [Computer software]. Retrieved from http:// CRAN.R-project.org/package=lme4 Bertram, R., Hyönä, J., & Laine, M. (2000). The role of context in morphological processing: Evidence from Finnish. Language and Cognitive Processes, 15, 367–388. doi:10.1080/ 01690960050119634 Bertram, R., Laine, M., Baayen, R. H., Schreuder, R., & Hyönä, J. (2000). Affixal homonymy triggers full-form storage, even with inflected words, even in a morphologically rich language. Cognition, 74(2), B13–B25. doi:10.1016/S00100277(99)00068-2 Bertram, R., Laine, M., & Karvinen, K. (1999). The interplay of word formation type, affixal homonymy, and productivity in lexical processing: Evidence from a morphologically rich language. Journal of Psycholinguistic Research, 28, 213– 226. doi:10.1023/A:1023200313787

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

Language, Cognition and Neuroscience Butterworth, B. (1983). Lexical representation. In B. Butterworth (Ed.), Language production (Vol. 2, pp. 257–294). London: Academic Press. Chambers, S. M. (1979). Letter and order information in lexical access. Journal of Verbal Learning and Verbal Behavior, 18, 225–241. doi:10.1016/S0022-5371(79)90136-1 Christianson, K., Johnson, R. L., & Rayner, K. (2005). Letter transpositions within and across morphemes. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31, 1327–1339. doi:10.1037/0278-7393.31.6.1327 de Almeida, R. G., & Libben, G. (2005). Changing morphological structures: The effect of sentence context on the interpretation of structurally ambiguous English trimorphemic words. Language and Cognitive Processes, 20, 373– 394. doi:10.1080/01690960444000232 DeLong, K. A., Urbach, T. P., & Kutas, M. (2005). Probabilistic word pre-activation during language comprehension inferred from electrical brain activity. Nature Neuroscience, 8, 1117– 1121. doi:10.1038/nn1504 Deutsch, A., Frost, R., Pollatsek, A., & Rayner, K. (2005). Morphological parafoveal preview benefit effects in reading: Evidence from Hebrew. Language and Cognitive Processes, 20, 341–371. doi:10.1080/01690960444000115 Duñabeitia, J. A., Perea, M., & Carreiras, M. (2007). Do transposed-letter similarity effects occur at a morpheme level? Evidence for morpho-orthographic decomposition. Cognition, 105, 691–703. Duñabeitia, J. A., Perea, M., & Carreiras, M. (2014). Revisiting letter transpositions within and across morphemic boundar‐ ies. Psychonomic Bulletin & Review, 21, 1557–1575. doi:10. 3758/s13423-014-0609-2 Fine, A. B., Jaeger, T. F., Farmer, T. A., & Qian, T. (2013). Rapid expectation adaptation during syntactic comprehension. PloS One, 8, e77661. Forster, K. I., Davis, C., Schoknecht, C., & Carter, A. (1987). Masked priming with graphemically related forms: Repetition or partial activation? Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 39, 211–251. Gagné, C. L., Spalding, T. L., & Gorrie, M. C. (2005). Sentential context and the interpretation of familiar open-compounds and novel modifier-noun phrases. Language & Speech, 48, 203–221. Hyönä, J., Vainio, S., & Laine, M. (2002). A morphological effect obtains for isolated words but not for words in sentence context. European Journal of Cognitive Psychology, 14, 417–433. Juhasz, B. J. (2012). Sentence context modifies compound word recognition: Evidence from eye movements. Journal of Cognitive Psychology, 24, 855–870. doi:10.1080/20445911. 2012.706602 Laszlo, S., & Federmeier, K. D. (2009). A beautiful day in the neighborhood: An event-related potential study of lexical relationships and prediction in context. Journal of Memory and Language, 61, 326–338. doi:10.1016/j.jml.2009.06.004 Levy, R. (2008). Expectation-based syntactic comprehension. Cog‐ nition, 106, 1126–1177. doi:10.1016/j.cognition.2007.05.006 Luke, S. G., & Christianson, K. (2011). Stem and whole-word frequency effects in the processing of inflected verbs in and out of a sentence context. Language and Cognitive Processes, 26, 1173–1192. doi:10.1080/01690965.2010.510359 Luke, S. G., & Christianson, K. (2012). Semantic predictability eliminates the transposed-letter effect. Memory & Cognition, 40, 628–641. doi:10.3758/s13421-011-0170-4 Luke, S. G., & Christianson, K. (2013). The influence of fre‐ quency across the time course of morphological processing:

13

Evidence from the transposed-letter effect. Journal of Cognitive Psychology, 25, 781–799. Marslen-Wilson, W. D., Bozic, M., & Randall, B. (2008). Early decomposition in visual word recognition: Dissociating morphology, form, and meaning. Language and Cognitive Processes, 23, 394–421. doi:10.1080/01690960701588004 McClelland, J. L., & Patterson, K. (2002). Rules or connections in past-tense inflections: What does the evidence rule out? Trends in Cognitive Sciences, 6, 465–472. doi:10.1016/ S1364-6613(02)01993-9 Otten, M., Nieuwland, M. S., & Van Berkum, J. J. (2007). Great expectations: Specific lexical anticipation influences the processing of spoken language. BMC Neuroscience, 8(1), 89. doi:10.1186/1471-2202-8-89 Perea, M., & Lupker, S. J. (2003a). Does jugde activate COURT? Transposed-letter similarity effects in masked associative priming. Memory & Cognition, 31, 829–841. doi:10.3758/ BF03196438 Perea, M., & Lupker, S. J. (2003b).Transposed-letter confusability effects in masked form priming. In S. Kinoshita & S. J. Lupker (Eds.), Masked priming: The state of the art (pp. 97– 120). New York, NY: Psychology Press. Pollatsek, A., Drieghe, D., Stockall, L., & de Almeida, R. G. (2010). The interpretation of ambiguous trimorphemic words in sentence context. Psychonomic Bulletin & Review, 17, 88– 94. doi:10.3758/PBR.17.1.88 R Development Core Team. (2012). R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. ISBN 3-900051-07-0. Retrieved from http://www.R-project.org Rastle, K., & Davis, M. H. (2003). Reading morphologicallycomplex words: Some thoughts from masked priming. In S. Kinoshita & S. J. Lupker (Eds.), Masked priming: State of the art (pp. 154–168). Hove: Psychology Press. Rastle, K., & Davis, M. H. (2008). Morphological decomposition based on the analysis of orthography. Language and Cognitive Processes, 23, 942–971. doi:10.1080/0169096080 2069730 Rastle, K., Davis, M. H., & New, B. (2004). The broth in my brother’s brothel: Morpho-orthographic segmentation in visual word recognition. Psychonomic Bulletin & Review, 11, 1090–1098. doi:10.3758/BF03196742 Rayner, K., & Well, A. D. (1996). Effects of contextual constraint on eye movements in reading: A further examination. Psychonomic Bulletin & Review, 3, 504–509. doi:10.37 58/BF03214555 Schoonbaert, S., & Grainger, J. (2004). Letter position coding in printed word perception: Effects of repeated and transposed letters. Language and Cognitive Processes, 19, 333–367. doi:10.1080/01690960344000198 Schreuder, R., & Baayen, R. H. (1995). Modeling morphological processing. In L. B. Feldman (Ed.), Morphological aspects of language processing (pp. 131–154). Hillsdale, NJ: Lawrence Erlbaum Associates. Seidenberg, M. S., & Gonnerman, L. M. (2000). Explaining derivational morphology as the convergence of codes. Trends in Cognitive Sciences, 4, 353–361. Szewczyk, J. M., & Schriefers, H. (2013). Prediction in language comprehension beyond specific words: An ERP study on sentence comprehension in Polish. Journal of Memory and Language, 68, 297–314. doi:10.1016/j.jml.2012.12.002 Taft, M. (2004). Morphological decomposition and the reverse base frequency effect. The Quarterly Journal of Experimental Psychology Section A, 57, 745–765. doi:10.1080/02724980 343000477

Downloaded by [University of Illinois at Urbana-Champaign] at 13:03 12 February 2015

14

S.G. Luke and K. Christianson

Taft, M., & Forster, K. I. (1975). Lexical storage and retrieval of prefixed words. Journal of Verbal Learning and Verbal Behavior, 14, 638–647. doi:10.1016/S0022-5371(75)80051-X Tyler, L. K., & Cobb, H. (1987). Processing bound grammatical morphemes in context: The case of an aphasic patient. Language and Cognitive Processes, 2, 245–262. doi:10.10 80/01690968708406934 Tyler, L. K., & Marslen-Wilson, W. D. (1986). The effects of context on the recognition of polymorphemic words. Journal of Memory and Language, 25, 741–752. doi:10.1016/0749596X(86)90047-1 Van Berkum, J. J. A., Brown, C. M., Zwitserlood, P., Kooijman, V., & Hagoort, P. (2005). Anticipating upcoming words in discourse: Evidence from ERPs and reading times. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31, 443–467. doi:10.1037/0278-7393.31.3.443 Velan, H., Deutsch, A., & Frost, R. (2013). The flexibility of letter-position flexibility: Evidence from eye movements in reading Hebrew. Journal of Experimental Psychology: Human

Perception and Performance, 39, 1143–1152. doi:10.1037/ a0031075 White, S. J., Johnson, R. L., Liversedge, S. P., & Rayner, K. (2008). Eye movements when reading transposed text: The importance of word-beginning letters. Journal of Experimental Psychology: Human Perception and Performance, 34, 1261–1276. doi:10.1037/0096-1523.34.5.1261 Wicha, N. Y., Moreno, E. M., & Kutas, M. (2004). Anticipating words and their gender: An event-related brain potential study of semantic integration, gender expectancy, and gender agreement in Spanish sentence reading. Journal of Cognitive Neuroscience, 16, 1272–1288. doi:10.1016/S0010-9452(08) 70260-0 Zwitserlood, P., Bolwiender, A., & Drews, E. (2005). Priming morphologically complex verbs by sentence contexts: Effects of semantic transparency and ambiguity. Language and Cognitive Processes, 20, 395–415. doi:10.1080/01690960 444000160