Genericity and the N400 1 Characterizing kinds and instances of kinds: ERP reflections

Syntactic and semantic information are computed online in a manner such that electrophysiological methods can detect distinct processes within a few hundred milliseconds of a word. The amplitude of the N400 response has been shown to reflect semantic integration of a word in the context of a preceding word, sentence, and discourse. We show, in a combined behavioral and ERP study, that the N400 amplitude to the same word, in nearly identical sentential contexts, is modulated as a function of subtly different morphosyntactic environments that condition either a generic (grass is green) or nongeneric (the grass is green) reading. The results suggest that N400 amplitude reflects not only the existence of a semantic computation but can reflect processes relevant to the type of semantic relation being computed. Specifically, it is sensitive to whether a word is interpreted as characterizing a kind/type or an instance of a kind/token of a type. Genericity and the N400 3 A central problem in language comprehension involves when and how incoming information is incorporated into existing semantic representations. Event-related potentials (ERP) provide one tool for investigating this question. Findings indicate that the N400 response reflects the processing of semantic information, and that its amplitude varies with the degree to which a word fits the preceding word or sentential context (e.g. Federmeier & Kutas, 1999; Friederici, Pfeiffer, & Hahne, 1992; Hagoort & Brown, 1994; Holcomb, Kounios, Anderson, & West, 1999; Kutas, 1993; Kutas & Hillyard, 1980,1984; Neville, Nicol, Barss, Forster, & Garrett, 1991; Osterhout & Mobley, 1995; Van Petten, 1993,1995; among many others). Research by van Berkum, Hagoort, & Brown (1999) shows that this sensitivity to semantic fit extends to extra-sentential discourse contexts. In the present study, we investigate whether the type of semantic relation a given word has to the preceding linguistic context influences the N400. Specifically, we investigate how N400 amplitude varies when a critical word characterizes a kind (e.g., BANANA) or an instance of a kind (e.g., this banana). Kutas and Hillyard (1980) demonstrated that N400 amplitude varied with the plausibility and/or predictability of a word within a given context. Using sentences such as (1) and (2), they found that words such as beard in (1) elicited a reduced N400 in comparison to less plausible or predictable words such as city (2). (1) He shaved off his mustache and beard. (2) He shaved off his mustache and city. Genericity and the N400 4 This result has been extended in a number of ways. Studies consistently find that more plausible and more predictable words elicit smaller N400 amplitudes. These data suggest that N400 amplitude reflects processes involved in the integration of semantic information into the existing context. However, these studies do not indicate whether N400 amplitude also reflects the type of semantic relation a word has to its preceding context. A notable exception are studies which have investigated the is a relation in categorization (A banana is a fruit/A banana is not a fruit/A banana is a vehicle/A banana is not a vehicle) (e.g. Fischler, Bloom, Childers, Roucous, & Perry,1983). These studies show that the N400 is sensitive to whether the subject category is a member of the predicate category, providing evidence that N400 amplitude is sensitive to category inclusion. The present study investigated whether N400 amplitude is also sensitive to the difference between the relation involved in characterizing a kind (or type) and characterizing an instance of a kind (or token of a type). We held the critical word constant across sentences and manipulated its semantic relation to the preceding context. In one set of sentences, the critical word characterized a kind (3), whereas in another set, the same critical word characterized an instance of that kind (4). We chose such sentences because they have drastically different interpretations despite identical content words and virtually identical surface forms. The interpretations of these sentences differ along a number of (related) dimensions. First, whereas the predicate in (3) characterizes a kind, the predicate in (4) characterizes a specific instance of a kind. Second, whereas (3) has a determinate truth-value, (4) does not. It may be true Genericity and the N400 5 of some banana but not others, or a banana only at a given time. Finally, generics have discourse independent interpretations, whereas nongenerics do not. (3) Bananas are yellow. (4) This banana is yellow. (5) Bananas are green. (6) This banana is green. These differences in interpretation indicate that the critical word (yellow) must be integrated with representations constructed based on the preceding sentential context in different ways for generics and nongenerics. Furthermore, the manner in which the critical word is integrated depends not only on whether the subject refers to a kind or to instances of a kind, but also on the characterizing word. For example, abundant must be interpreted differently than yellow with respect to the kind BANANA. In sum, the interpretations of generic and nongeneric sentences differ in a number of ways. If N400 amplitude is sensitive to the type of semantic relation into which a critical word enters, then characterizing words in generic and nongeneric sentences may elicit distinct N400 responses. Comparing the processing of generic and nongeneric sentences also provides a theoretical basis for distinguishing the effects of predictability and genericity. In true generic sentences (3), the critical word must refer to a characteristic property of the kind. No such constraint applies to the critical word in the corresponding nongeneric sentence (4). In a nongeneric sentence, the critical word could refer to a characteristic property, or Genericity and the N400 6 an uncharacteristic or temporary property (e.g. green, wet). Thus, the critical word in true generic sentences (3) should be more predictable than the same word in corresponding nongeneric sentences (4). In contrast, critical words cannot refer to an uncharacteristic property if a sentence is generic, but may do so if the sentence is nongeneric. Consequently, critical words that refer to an uncharacteristic property should be more predictable in nongeneric sentences (6) than in generic sentences (5). These predictions concerning genericity and predictability were tested behaviorally in Experiments 1A and 1B and the reflections of genericity and predictability were tested electrophysiologically in Experiment 2. Experiment 1A In Experiment 1A we used the Cloze procedure to determine if the predictability of characteristic and uncharacteristic properties varies in accordance with constraints on the interpretation of generic and nongeneric sentences.