On understanding creative language: The late positive complex and novel metaphor comprehension

Highlights

• Electrophysiological correlates of novel metaphor comprehension.

• A graded N400 effect was found for literal, novel metaphoric and anomalous sentences.

• Reduced LPC amplitudes were found for novel metaphoric sentences.

• Both novelty of the metaphoric sentence and task demands affect the LPC pattern.

Abstract

Novel metaphoric sentences have repeatedly evoked larger N400 amplitudes than literal sentences, while investigations of the late positive complex (LPC) have brought inconsistent results, with reports of both increased and reduced amplitudes. In two experiments, we examined novel metaphor comprehension in Polish, using the same set of literal, novel metaphoric, and anomalous sentences. The first aim of the study was to test whether novel metaphors would evoke larger or smaller late positivity complex (LPC) amplitudes compared to literal and anomalous sentences. Some earlier studies have shown that whether increased LPC amplitudes are observed or not may be related to the task participants are asked to perform, with explicit acceptability judgments being more likely to evoke the LPC effect. The second aim of the study was, thus, to test whether the observed LPC pattern would be the same across two tasks, the semantic decision task (Experiment 1) and the reading task (Experiment 2). Our results replicated the N400 effect observed in earlier studies on metaphor in both experiments. Most importantly, a reduction in late positivity to novel metaphors relative to anomalous sentences was found in both experiments. Additionally, this difference was broadly distributed over parietal sites in Experiment 1, and clearly left-lateralized in Experiment 2, which might imply differences in the involvement of recollection and semantic processes. Overall, these results seem to indicate that both conventionality and task demands modulate the LPC pattern.

Introduction

Novel metaphors reflect the human drive to go beyond routine in communication. Although the common association they evoke might be with novels and poetry, they seem to play an important role in everyday communication, by evoking interest, surprise, and thus attracting our attention. Many theories of metaphor and models of metaphor comprehension have pervaded the fields of linguistics, philosophy and psychology, as scholars investigated the linguistic and cognitive structure of metaphors, features that make a metaphoric sentence plausible and apt, as well as processes involved in metaphor comprehension. The latter will be the main focus of the current paper. We will attempt to test the extent to which processes involved in metaphor comprehension and indexed by two event-related potential (ERP) components, i.e., the late positive complex (LPC) and the N400, are influenced by stimulus related aspects such as novelty of the metaphoric sentence, and task related aspects, such as potential differences between a semantic decision and a reading task. Before specific questions are raised, however, we will shortly outline previous behavioral and ERP findings related to the current study.

One of the first models of metaphor comprehension, the standard pragmatic model (Grice, 1975), stipulated that the literal meaning of a metaphoric utterance needs to be rejected before the figurative meaning is selected. Contrary to these predictions, authors of early behavioral studies on metaphor comprehension argued that the stage of literal meaning rejection is not a necessary step in arriving at non-literal meaning, by showing comparable reaction times to metaphoric and literal sentences (Gibbs et al., 1997, Inhoff et al., 1984, Janus and Bever, 1985, Onishi and Murphy, 1993, Pynte et al., 1996). This was further corroborated by results of studies that used the metaphor interference technique, in which participants are asked to classify three types of statements (true class-inclusion statements, e.g., Some birds are robins, false class-inclusion statements, e.g., Some birds are apples, and metaphoric statements, e.g., Some jobs are jails) as either literally true or literally false. Several studies have shown that participants take more time to reject a metaphoric statement than a false inclusion statement, and concluded that metaphoric meaning is activated before literal meaning processing has been completed (Glucksberg et al., 1982, Kazmerski et al., 2003, Wolff and Gentner, 2000). It is, however, important to bear in mind that early behavioral studies largely did not distinguish between different levels of conventionality of the experimental sentences, while results of recent research clearly show that conventionality impacts comprehension. This is manifested in longer reaction times and lower accuracy rates to novel metaphors as compared to conventional metaphoric or literal sentences (Arzouan et al., 2007a, Lai et al., 2009). Novel metaphors are frequently perceived as somewhat meaningful, as their meaning has no representation and needs to be constructed based on available semantic information.

This impact of conventionality has also been noted in the pattern of the N400 effect in research on metaphor comprehension. The N400, a negative-going wave observed between 300 and 500 ms after critical stimulus onset, was initially observed in response to semantic anomaly (Bentin et al., 1985, Kutas and Hillyard, 1984, Rugg and Nagy, 1987). One of the widely confirmed results is that novel metaphoric utterances evoke larger N400 amplitudes than literal utterances (Arzouan et al., 2007a, Arzouan et al., 2007b, Coulson and Van Petten, 2002, Coulson and Van Petten, 2007, De Grauwe et al., 2010, Goldstein et al., 2012, Lai et al., 2009, Pynte et al., 1996, Tartter et al., 2002). This increase in the N400 amplitude has been interpreted as reflecting difficulty of conceptual retrieval and integration (Arzouan et al., 2007a, Coulson and Van Petten, 2002, Goldstein et al., 2012), effort related to the integration of the unrelated concepts leading to conceptual expansion (Rutter et al., 2012), processing effort in lexical access (Weiland et al., 2014), or the extent of inferences that can be projected from one concept to another (Lai and Curran, 2013). Overall, these interpretations seem in line with the view of the N400 as reflecting the intersection between bottom-up processing of a stimulus and top-down activity in semantic memory, the result of which is an initial, flexible conceptual representation that might, however, need refinement at later stages (Kutas and Federmeier, 2011).

Interestingly, the N400 has shown sensitivity to the degree of conventionality of a metaphoric utterance, with novel metaphors (e.g., ripe dream) evoking somewhat larger N400 amplitudes than conventional metaphors (e.g., transparent intention), and somewhat smaller than anomalous utterances (e.g., indirect blanket) (Arzouan et al., 2007a, Goldstein et al., 2012). This graded N400 effect with similar latency, topography and morphology of the N400 component seems to provide some evidence for the claim that the difference between figurative and literal language is quantitative rather than qualitative, and stems from more complex mapping processes in the case of metaphoric than literal language, which is additionally modulated by the novelty of the mapping.

One model that accounts for both quantitative differences between literal and metaphoric language comprehension and the effects of conventionality of the metaphoric utterances observed in behavioral and electrophysiological studies is the Career of Metaphor Model (Bowdle and Gentner, 2005). This account proposes that metaphor is “a species of analogy” (Bowdle and Gentner, 2005, p. 196), and, thus, comparison is the main mechanism involved in novel metaphor comprehension. In the case of conventional metaphors, the model proposes that comparison is replaced by a faster and more automatic process of categorization in the course of conventionalization.

Two recent ERP studies have lent strong support to the Career of Metaphor Model. Lai and Curran (2013) used sentence-primes and simile-primes preceding metaphoric and literal sentences, and found that comparison induced by simile-primes facilitated mappings in novel metaphors producing reduced N400 amplitudes. This finding suggests that comparison is necessary in understanding novel metaphors. For conventional metaphors, the N400 priming effect was marginal, but it was significant in reaction times at later stages of processing. Thus, the authors concluded that comparison is vital in novel metaphor comprehension, and can be facilitative at later stages in conventional metaphor processing. Goldstein et al. (2012) asked their participants to explain half of the novel and conventional metaphoric word dyads presented in the exposure phase, followed by a test phase, in which participants performed a semantic judgment task for all the word pairs. The results showed that novel metaphoric word dyads that were explained in the exposure phase evoked smaller N400 amplitudes than those that were not explained, while conventional metaphoric word dyads that were explained in the exposure phase evoked larger N400 amplitudes than those that were not explained. This modulation of the N400 amplitudes by recent explanation was interpreted as indicative of reactivation of mappings available for conventional metaphors and sense retrieval for novel metaphors, showing that the semantic representations of both novel and conventional metaphors are dynamic and flexible.

These results clearly indicate that novel metaphor comprehension requires meaning construction rather than retrieval, which points to the creative nature of novel metaphor. One ERP study has recently compared the N400 responses to common and creative uses of objects and found a similar N400 effect to the one reported in studies on metaphor comprehension, with larger N400 amplitudes for creative than common uses (Kröger et al., 2012). This similarity in results could be seen as a link between different instances of creative thinking, and shows that investigating novel metaphor comprehension can bring us closer to understanding some aspects of creative cognition.

Although the N400 pattern observed in studies on metaphor comprehension seems to be quite consistent, it remains debatable that a single component could index all aspects involved in the comprehension of metaphoric sentences, as is evident from research on the LPC and sentence processing (Steinhauer et al., 2010). The LPC, known also as the P600, is a positive-going wave with a peak between 600 and 800 ms after stimulus onset. Although it was initially linked to the processing of syntactic anomalies (Hagoort et al., 1993, Osterhout et al., 1994), as well as to syntactic reanalysis (Friederici and Mecklinger, 1996), or syntactic integration (Kaan et al., 2000), results of later studies revealed its sensitivity to semantic anomalies, which gave rise to research on the semantic P600 (for a review see Bornkessel-Schlesewsky and Schlesewsky (2008)).

The P600 following the N400, also dubbed the biphasic N400-P600, has been observed for sentences which involve non-reversible semantic anomalies (van Herten et al., 2005). Brouwer et al. (2012) interpreted this pattern as reflecting retrieval of lexical and semantic information from long term memory (the N400), and the integration of this retrieved information into the mental representation of the sentences (the P600). Within this view, novel metaphors should evoke increased N400 amplitudes compared to literal language, as the retrieval poses greater demands related to meaning construction, and increased late positivity, as the integration of the retrieved information regarding two previously not associated terms is more difficult than in the case of literal language.

Results concerning the LPC in response to metaphoric sentences are, however, far from consistent. First, several studies on metaphor comprehension that reported the N400 effect did not examine the LPC time window. Furthermore, of those that did, some showed the expected biphasic pattern with increased N400 and LPC amplitudes to metaphoric sentences (Coulson and Van Petten, 2002, De Grauwe et al., 2010, Weiland et al., 2014), while others revealed the opposite pattern, with increased N400 but reduced LPC amplitudes to novel metaphors relative to literal and anomalous utterances (Arzouan et al., 2007a, Goldstein et al., 2012). Aspects such as task type (including semantic judgment tasks, reading tasks, or delayed response procedures) or stimulus selection criteria (with levels of meaningfulness, metaphoricity, familiarity and cloze probability controlled to various degrees) may have contributed to those inconsistencies (De Grauwe et al., 2010, Kuperberg, 2007). Below we discuss specific implications arising from these methodological differences.

The semantic decision task and the reading task have been most widely used in ERP studies on metaphor comprehension. In the semantic decision task, participants are asked to decide whether a sentence makes sense or not. This task was used in studies reported by Arzouan et al. (2007a) (Experiment 1 and 3), and Goldstein et al. (2012), which revealed a reduced LPC to novel metaphors, and by De Grauwe et al. (2010), who observed an increased LPC to conventional metaphors. In the reading task, participants are asked to read the sentences presented on the computer screen. This task was used by Arzouan et al. (2007a) (Experiment 2) and Coulson and Van Petten, 2002, Coulson and Van Petten, 2007. Additionally, in some studies (Coulson and Van Petten, 2002, Coulson and Van Petten, 2007), participants were asked to answer follow-up true-false comprehension questions. The use of the reading tasks has also resulted in inconsistent findings, with increased (Coulson and Van Petten, 2002) and reduced (Arzouan et al., 2007a, Coulson and Van Petten, 2007) LPC amplitudes to metaphoric sentences. Overall, the semantic decision task and the reading task have revealed both patterns of the LPC effect, with smaller and larger amplitudes to metaphors than literal and anomalous utterances.

Another factor that might be crucial for the LPC findings is the extent to which conventionality was manipulated and controlled in the process of material selection. While Arzouan et al., 2007a, Goldstein et al., 2012 employed both conventional and novel metaphoric word dyads, De Grauwe et al. (2010) tested exclusively conventional metaphors, Weiland et al. (2014) employed metaphors that were neither familiar nor unfamiliar, and Coulson and Van Petten, 2002, Coulson and Van Petten, 2007 did not directly refer to normative studies on familiarity or conventionality. Overall, results of studies in which novel metaphors were pretested on familiarity rating scales showed decreased LPC to novel metaphors compared to literal or conventional metaphoric utterances.

As the above discussion demonstrates, it remains difficult to account for the inconsistencies in LPC results in terms of task demands or novelty of metaphoric sentences, but novelty of the metaphoric meaning seems to modulate the LPC amplitudes in a more consistent manner. At the same time, the influence of task demands is not reflected in a clear LPC pattern. For these reasons, we designed a study in which we used the same set of novel metaphoric, literal, and anomalous sentences, for which the critical word was the same across the three conditions. For these sentences, we examined the ERP responses in the semantic decision task and the reading task. This design allowed us to tap into the differences in the LPC response between the two tasks, while the differences arising from possible stimulus characteristics were minimized. We expected that the results would provide support for one of the two interpretations of the LPC effect proposed in studies on metaphor comprehension, which we present below.

First, authors who have reported increased late positivity following metaphoric sentences interpreted it as indexing recovery and integration of additional material from semantic memory (Coulson and Van Petten, 2002), enhanced costs due to pragmatically or inferentially driven mapping processes involving two unrelated domains (Weiland et al., 2014), or selection of the metaphoric and suppression of the literal meaning (De Grauwe et al., 2010). In line with the discussion presented by Brouwer et al. (2012), we would argue here for a more parsimonious interpretation that associates an increase in LPC amplitudes with greater effort related to the integration of semantic information retrieved within the N400 time window with the preceding context. In line with research on recognition memory (Paller and Kutas, 1992), it could also index recollection of the entire sentence, which might be more demanding in the case of novel metaphors due low familiarity.

On the other hand, attenuated LPC amplitudes to metaphors were observed in the studies reported by Arzouan et al., 2007a, Goldstein et al., 2012, but only in response to novel and not conventional metaphoric word pairs. Such reduced LPC amplitudes remain difficult to account for within the interpretation of the LPC as reflecting semantic integration, as novel metaphors would be expected to require more integration than literal sentences and, consequently, should evoke increased LPC amplitudes. One possible explanation postulated by Arzouan et al., 2007a, Goldstein et al., 2012 is that an overlapping negativity attenuates late positivity amplitudes in the case of novel metaphoric utterances, and might be indicative of access to the non-literal route in the course of novel metaphor comprehension (Arzouan et al., 2007a, Goldstein et al., 2012). Following this interpretation, we would argue that novel metaphors require prolonged retrieval of information from long term memory (possibly related to access to the non-literal route) and mapping between two semantically distant concepts, that is initiated within the N400 time window and later overlaps with the LPC time window. Such overlapping sustained negativity is then reflected in a reduced LPC amplitude to novel metaphors.

This interpretation could be partly supported by a recent study that found sustained negativity between 500 and 900 ms after critical word onset in response to novel metaphors (Rutter et al., 2012). The design of the study differed from other studies on metaphor in several aspects. Neither the semantic decision nor the reading task was used. Instead, participants were asked to decide whether or not a sentence was unusual (question 1) and appropriate (question 2). Also, a delayed response procedure was used. Although sustained negativity was interpreted as reflecting ongoing difficulty in integrating sentence meaning, it could also be related to high demands on working memory (Steinhauer and Drury, 2012, Steinhauer et al., 2010), which in this paradigm could have been triggered by the need to maintain the activated information in memory due to the delayed response procedure, and by task complexity. This interpretation could further be supported by the fact that sustained negativity was associated with all sentence types (literal, metaphoric and anomalous), with amplitude differences consistent with the N400 pattern.

In the current study, we aimed to test whether increased semantic integration (increased LPC amplitudes) or overlapping sustained negativity associated with access to non-literal meaning or continued difficulty in the mapping process (reduced LPC) would be associated with novel metaphor comprehension. Furthermore, we planned to investigate whether the obtained LPC effect would be present regardless of the task the participants were asked to perform. Additionally, we predicted that the N400 effect would be replicated, with novel metaphors evoking larger N400 amplitudes than literal, and smaller than anomalous sentences. Since this effect has been found in several earlier studies that used different task procedures, we expected that it would be present in both experiments.

To these aims, we designed two experiments with the semantic decision task (Experiment 1) and the reading task (Experiment 2). To ensure that the sentences in Experiment 2 were read for comprehension, we used yes-or-no comprehension questions on 30% of the trials. The same set of materials including novel metaphoric, literal and anomalous sentences was used in both experiments in order to minimize differences in results that might arise from the different degrees of meaningfulness, familiarity, metaphoricity and cloze probabilities in the experimental sentences. Unlike the materials used in the studies by Arzouan et al., 2007a, Arzouan et al., 2007b, Kuperberg, 2007, Goldstein et al., 2012, who used semantically related, unrelated, conventional and novel metaphoric two-word expressions, our stimuli included novel metaphoric, literal, and anomalous sentences. Lai et al. (2009) also used sentences, but the critical words included in their study were verbs, adjectives, and nouns, while our critical words were nouns. Finally, unlike De Grauwe et al. (2010), we did not use nominal (A is B) metaphors. Instead, our critical nouns were preceded by nouns, verbs, or adjectives. Similarly to all of the studies mentioned above (with the exception of Experiment 1 in De Grauwe et al. (2010)), the ERPs were time-locked to the sentence-final word.