Linguistic Stimuli Design The set of linguistic stimuli used in the experiments presented in this thesis was held similar to the maximum possible degree (see Table 3.1 for an example) in order to keep results comparable. All experimental items were German independent main clauses, uniform in their syntactic structure (NP-V-ADV-NP) and designed so that sentences’ subject NPs did not contain any helpful information with respect to the predictability of the verb or the object noun phrases.
Experiment 1, Experiment 2, and Experiment 3 featured two different verb categories, that is, highly constraining2(spill), and unconstraining (order) verbs. These differences in verb constraints were used to manipulate the target noun’s predictability and surprisal.
The adverb following the verb in all stimuli (The man spillssoon the water) served as a padding region in order to give time for linguistic predictions or, in the later experiments, for visually informed expectations concerning the object noun.
In order to observe possible differences in predictability and surprisal of the target nouns in the first three studies, both verb categories were paired with two different object nouns (see Table 3.1, (1) and (2)), one of which is more plausible in the verb context (see Table 3.1, column 3). If the differences in verb constraint information – either alone as inExperiment 1 andExperiment 2, or in combination with visual context as in Experiment 3 – were enough to have more or less specific expectations about the target noun, surprisal and processing effort for the less plausible nouns were expected to increase.
All auditorily presented stimuli were recorded using Audacity (Version 2.0.6). Factors with potentially substantial influence on a word’s processing effort and (linguistic) surprisal, such as word length and frequencies (Schilling et al., 1998) were controlled for. In particular, noun frequencies were derived a priori from the word lists DeReWo3of the German research corpus (DeReKo) and held approximately constant within an item. Differences in word length were integrated in the analyses (by either including the length factor as predictor in the model, or using mid-word time windows, see “Analysis” subsections for details). All nouns were concrete and inanimate, in order to make them easily depictable and, especially in the ERP study, to prevent differences in the N400 amplitude caused by abstract words (West and Holcomb, 2000). Fillers were always plausible sentences with differing length and of differing syntactic structure in order to prevent fatigue effects. Half of the fillers were followed by yes/no comprehension questions (such as “Did the man spill the lemonade?”) to keep participants focused.
2for validation of the categories, see 3.2
3(DeReKo corpus size > 28 billion words. Source: Korpusbasierte Wortgrundformenliste
For the sake of observing the effect manipulations on merely the visual context level on word processing, the linguistic stimuli were altered for studiesExperiment 4, Experiment 5, Experiment 6, and Experiment 7. That is, only highly contraining verbs were used, paired with only the more plausible noun continuations in order to exclude any linguistic variation and hence its effect on target word predictability and processing effort.
Linguistic Stimuli Validation Both linguistic manipulations used in the first three studies, i.e. strength of verb constraint and the nouns’ plausibility in their contexts were validated offline prior to using the stimuli in the actual online experiments.
Verb constraint The more constraining a verb is, the fewer plausible continuations it al- lows. A classical sentence completion task for cloze probability hence assessed to what extent the verbal constraint (i.e. highly constraining (spill) or unconstraining (order)) increased the predictability of the target noun (3.1, column 4)3. 17 German native speakers participated voluntarily in this online questionnaire. All items were truncated prior to the target noun and presented in one list, containing 50 % fillers, shown in randomised order. Participants were asked to spontaneously complete the sentences with a noun best fitting the sentence context. Unique participation of each webform user was controlled for. Results from the Cloze task showed differences between nouns in the highly constraining verb context only (see Table 3.1, column 4, 1a) vs. 1b)). Cloze probabilities ranged from 4% to 55% for plausible nouns in high constraining verb contexts (spill water) and from 0% and 4% for the less plausible nouns in the same contexts (spill ice cream). Unconstraining verbs produced Cloze probabilities <.01 for all critical nouns. In sum, higher constraint led to higher cloze probabilities of the two subsequent nouns (see Table 3.1, column 4).
Verb-noun plausibility A plausibility rating on a seven-point Likert scale assessed how plausible participants would rate a target noun to be in its sentence context (3.1, column 3). 14 German native speakers participated voluntarily in the online questionnaire. Participants read the stimuli sentences in a webform and were asked to spontaneously judge the plausibility of each sentence combination, resulting from the Verb - Object manipulation, ranging from 1 (very plausible) to 7 (not plausible at all). Items were presented in randomised lists, containing 50 % filler sentences. Each participant had only a single access to the webform. Results plotted in Fig.3.1 show that plausibility – or thematic fit – of the nouns used in the stimuli differed in the context of high constraining verbs (see Table 3.1, column 3, 1a) vs.
3Note that the Cloze task was only used to assess the strength of verb constraints. All object nouns
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 spill order Verbs Rating
Object Water Ice cream Book
Rating Questionnaire
Figure 3.1Pre-test results (verb-noun plausibility). Participants rated how plausible a noun was in the context of the previous verb, using a 7-point Likert scale where 7 is not plausible at all and 1 is perfectly plausible. Error bars reflect 95% confidence intervals (CI).
Table 3.1Sample items and corresponding pretest results for the two nouns in each verb condition: highly constraining (1) and unconstraining (2).
Item noun plausibility cloze%
M (SD) M (SD)
(1) The man spills soon the a) water 1.12 (0.68) 13.67 (18.06)1
b) ice cream 2.76 (2.17) 0.16 (0.54)
(2) The man orders soon the a) water 1.65 (1.50) < 0.01 (0.0)
b) ice cream 1.90 (1.80) < 0.01 (0.0)
1b)), while both nouns were equally plausible in the unconstraining verb context (see Table 3.1, column 3, 2a) vs. 2b)), reflecting the intended manipulation.