Abstract

Introduction

Creation of novel ideas involves combining of separate concepts that occurs infrequently and unexpectedly (Turner and Fauconnier, 1999). Creative ideas emerge from the convergence of multiple concepts, with creativity being enhanced by the distinctiveness of the concepts (Nagai et al., 2009). On the other hand, the emergence of novel ideas is primed by signature processes, such as sudden reorganizations of mental representation (Bowers et al., 1990; Seifert et al., 1994). However, reflecting the unexpected nature of the emergence, little is known about the temporally evolving processes underlying the generation of creative ideas.

Metaphor is a rhetorical expression that compares one concept to another concept to express the first concept in a concise and impressive manner (Lakoff and Johnson, 1980). Previous theoretical works have postulated the comprehension of metaphors in terms of the attributions of the concepts [e.g., Class Inclusion Theory (Glucksberg and Keysar, 1990); Structure Alignment Theory (Gentner et al., 2001)]. Importantly, however, metaphor comprehension requires the combining of different concepts, which involves cognitive processes that are comparable to those involved in the creation of novel thoughts (Becker, 1997; Gineste et al., 2000). One notable phenomenon is the “emergence” of interpretations (Tourangeau and Rips, 1991), where an interpretation of a metaphor is not related to either of the original concepts, but a combination of the two concepts provides for a novel interpretation to emerge¹. These characteristics of metaphor interpretation can provide a unique opportunity for examining the emergence of creative thoughts, and possibly uncovering an alternative mechanism for metaphor comprehension.

The present study asks what specific situations enhance creative interpretations, and what processes precede their emergence. Previous studies suggest that inspection of visually presented items reflects attentional shifts and plays critical roles for higher-level cognitive functions, such as decision-making (Glaholt and Reingold, 2011), problem solving (Knoblich et al., 2001), and sentence comprehension (Rayner, 1998). For example, during the reading of visually presented sentences, the gaze fixates longer on a word when extra semantic attention is needed (Ehrlich and Rayner, 1981; Frazier and Rayner, 1990), and spatial shifts in visual attention precede insights in problem solving (Knoblich et al., 2001). These empirical evidences suggest that tracking visual attention is potentially beneficial for examining the cognitive processes underlying the emergence of creative interpretations.

In order to address the issue above, the present study examines the temporal dynamics of visual attention while participants freely generated interpretation for visually presented metaphorical sentences (Figure ​Figure11). Responses were made vocally, and eye gaze was continuously monitored during the task. The sentences were manipulated such that they involved either semantically unrelated or related concept combinations (Figure ​Figure1A1A). We then examined the temporal patterns for gazes prior to the generation of creative interpretations.

Open in a separate window

FIGURE 1

Experimental design. (A) Visual stimulus set during the present behavioral task. Participants were visually presented by sentences of the form “ ‘topic’ is ‘vehicle’ ”, and instructed to think of interpretations of the sentences. (B) Temporal sequence of the task. Participants freely generated interpretations about the sentences and made spoken responses. They made a manual button press when they could not come up with more responses.

Materials and Methods

In order to address the issue above, the present study examines the temporal dynamics of visual attention while participants freely generated interpretation for visually presented metaphorical sentences (Figure ​Figure11). Responses were made vocally, and eye gaze was continuously monitored during the task. The sentences were manipulated such that they involved either semantically unrelated or related concept combinations (Figure ​Figure1A1A). We then examined the temporal patterns for gazes prior to the generation of creative interpretations.

Results

We first sought to verify the classifications of the E and NE responses by analyzing semantic unrelatedness ratings for the presented topic/vehicle words and the generated responses. Sampled responses (see Materials and Methods for details) were rated by an independent participant group (N = 102; see also Materials and Methods for details; Figure ​Figure2A2A). A repeated measures two-way ANOVA with response type (E, NE) and sentence (N, F, L) as factors revealed main effects of both response type [F(1,101) = 87.7, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.46] and sentence [F(2,202) = 40.0, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.28]. Post hoc t-tests further revealed that the unrelatedness scores for E responses were higher than those for the NE responses in all sentence conditions [E-N > NE-N: t(101) = 4.6, p < 0.01, corrected for multiple comparisons based on Holm’s method, r = 0.42; E-F > NE-F: t(101) = 7.5, p < 0.01, corrected, r = 0.59; E-L > NE-L: t(101) = 6.6, p < 0.01, corrected, r = 0.55]. The unrelatedness scores for interpretations of the N sentences were higher than those for both F and L sentences [N > F: t(101) = 7.3, p < 0.001, corrected, r = 0.59; N > L: t(101) = 6.7, p < 0.001, corrected, r = 0.56]. These results from the independent participant group validate the classifications of the E and NE responses and the subjective unrelatedness scores of the participants in the interpretation task (Tourangeau and Rips, 1991).

Open in a separate window

FIGURE 2

(A) Average ratings for the degree of semantic unrelatedness between generated responses and presented words (7: very unrelated; 1: very related) within the evaluations from the independent participant group (N = 102). (B) Average frequencies of responses in the free interpretation experiment. (C) Average durations to responses (DR) defined as the duration between the onsets of serial responses. Error bars represent the SEM across participants. p < 0.05; p < 0.01.

Next, we analyzed all the responses generated in the free interpretation experiment, and calculated the frequencies of responses (Figure ​Figure2B2B). A two-way repeated measures ANOVA was conducted with sentence conditions (N, F, L) and response types (E, NE) as factors. Significant main effects were observed for both sentences [F(2,44) = 24.3, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.52] and response types [F(1,22) = 33.7, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.61], as well as significant interaction [F(2,44) = 33.0, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.60]. Post hoc t-tests further revealed that E responses were less frequently given than NE responses in all sentence conditions [NE-N > E-N: t(22) = 3.2, p < 0.01, corrected, r = 0.56, NE-F > E-F: t(22) = 5.9, p < 0.01, corrected, r = 0.78, NE-L > E-L: t(22) = 6.8, p < 0.01, corrected, r = 0.83]. Moreover, E responses were generated more frequently for the N and F sentences than for the L sentences [E-L < E-N: t(22) = 3.8, p < 0.01, corrected, r = 0.63; E-L < E-F: t(22) = 3.7, p < 0.01, corrected, r = 0.62]. These findings suggest that E responses were more infrequent, although they were mainly observed in the metaphorical (N, F) conditions. On the other hand, NE responses for N sentences were generated less frequently than such responses for the F and L sentences [NE-N < NE-F: t(22) = 7.8, p < 0.01, corrected, r = 0.87; NE-N < NE-L: t(22) = 5.9, p < 0.01, corrected, r = 0.78], suggesting that NE responses were generated more from conventional expressions (F, L) than from novel expressions (N), and, thus, that the N sentences were more challenging to interpret.

We then analyze temporal data for serial responses in order to examine the duration to generate interpretations. Durations to responses (DR) were defined as the interval between the onset of the response and the onset of the immediately prior response (regardless of response type). DRs thus reflect the time required to produce a response (Figure ​Figure2C2C). A two-way repeated measures ANOVA was conducted with response type (E, NE) and sentence condition (N, F, L) as factors. A significant main effect of sentence was observed [F(2,44) = 17.1, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.44]. Post hoc t-tests further revealed that DRs were longer during N sentences than both L and F sentences [N > F: t(22) = 5.3, p < 0.01, corrected, r = 0.75; N > L: t(22) = 3.9, p < 0.01, corrected, r = 0.64], suggesting that N sentences required longer durations to produce interpretations, which is consistent with the lower frequency of responses for N sentences.

Next, the eye tracking data was analyzed in order to examine visual attention toward the sentences. We first calculated the CGDs for both the topic and vehicle words during response generation, as the overall average DR period (12.3 s; see Materials and Methods for more details). As shown in Figure ​Figure3A3A, CGDs are assumed to reflect the total gaze durations when generating responses.

Open in a separate window

FIGURE 3

(A) Cumulative gaze durations (CGDs) to generate responses. Error bars represent SEM. p < 0.05; p < 0.01. (B) The time courses of x-axis gaze points within the N condition. The horizontal and vertical axes indicate time and normalized gaze points, respectively, and the time 0 indicates the onset of a spoken response. The solid lines (t < 0) below the time courses indicate the point when the gaze was shifted toward the topic and vehicle words and where the difference in gaze points (E vs. NE) was significant (p < 0.05).

We first analyzed CGDs for topic words. A two-way repeated measures ANOVA for CGDs was conducted with response type (E, NE) and sentence condition (N, F, L) as factors. A significant main effect of sentence [F(2,44) = 3.4, p < 0.05, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.13] was observed, as well as an interaction between sentence and response type [F(2,44) = 5.5, p < 0.01, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.20]. Post hoc t-tests further revealed longer CGD for the N sentence than F sentence [N > F: t(22) = 2.9, p < 0.05, corrected, r = 0.53]. Most importantly, production of E responses within the N sentence condition were preceded by longer CGDs than within F sentence condition [E-N > E-F: t(22) = 3.4, p < 0.01, corrected, r = 0.58]. These results suggest that greater visual attention was given to the topic word during the generation of E responses within the N sentence, which is consistent with the suggestion that E responses were challenging to generate.

Next, for vehicle words, a similar two-way repeated measures ANOVA was conducted. A significant main effect of sentence was observed [F(2,44) = 11.9, p < 0.01, ${\mathrm{\eta }}_{\mathrm{p}}^{\mathrm{2}}$ = 0.35], although the interaction effect failed to reach significance. Post hoc t-tests revealed longer CGDs for the N sentence condition [N > F: t(22) = 6.3, p < 0.01, corrected, r = 0.80, L > F: t(22) = 2.9, p < 0.05, corrected, r = 0.52], suggesting that interpretation generation within the N sentence conduction is associated with greater visual attention to vehicle words in addition to topic words.

Figure ​Figure3B3B presents the time courses of gaze point for both E and NE responses within N sentences. A sliding-window t-test (window size = 10 sample frames = 0.17 s) revealed significant gaze shifts to the topic word around -8, -7, and -2 s prior to E responses, but such shifts were not observed prior to NE responses. Accordingly, differences in gaze points were significant for E responses compared to NE responses at -11, -9, -8 s, shifting toward the topic words. These results suggest that the topic words received more visual attention during an earlier period in the generation of E responses, possibly reflecting the processes underlying the generation of creative interpretations.

Discussion

Previous theoretical studies of metaphor comprehension have postulated that the interpretation of metaphors involves a conceptual interaction between topic and the vehicle representations (Black, 1962), and a combination of the topic and vehicle concepts brings about emergent interpretations (Gineste et al., 2000). Class Inclusion Theory (Glucksberg and Keysar, 1990) further extended this theoretical account by introducing a selective representation of vehicle characteristics that guides the attribution dimension related to the topic concept, and the theory was verified experimentally (Glucksberg et al., 1997; McGlone and Manfredi, 2001). It has also been hypothesized that the category representation of the vehicle is formed in an ad hoc manner (so-called ad hoc category), and then transformed into topic attribution during the generation of emergent interpretations (Vega Moreno, 2004). The present study provides experimental evidence for these hypotheses by demonstrating that sentence words received prolonged visual attention prior to the generation of emergent interpretations.

In novel metaphors, because the topic and vehicle words were semantically dissimilar, a broader semantic search is required to construct the ad hoc category for the vehicle information (Stringaris et al., 2007). This idea is compatible with the present findings that novel metaphors were associated with longer durations in the generation of interpretations. On the other hand, the enhanced visual attention prior to emergent interpretations focused on the topic word. The shifted visual attention to topic words supports the transformational account of emergent interpretations, in that the topic words were gazed at longer when the ad hoc category information for the vehicle characteristics was transformed to topic-related information. Previous neuroimaging studies have suggested that searching for a wider range of semantic relationships within metaphor comprehension elicits activation of the right inferior frontal gyrus (Stringaris et al., 2006; Shibata et al., 2007a). Greater activations were observed in the right precentral gyrus and anterior cingulate when reading novel metaphors compared to literal sentences. Moreover, the right precentral activation was greater compared to when reading familiar metaphors, which suggest that the involvement of precentral was specific to reading novel metaphors (Ahrens et al., 2007). Collectively, such neuroimaging evidence might suggest a right hemisphere contribution in metaphor comprehension, which contrasts with conventional knowledge about left-hemisphere dominant language processing. It is suggested that the application of neuroimaging methods to the current task might allow an examination of dynamic aspects (e.g., Jimura and Braver, 2010; Jimura et al., 2010, 2013) of this hypothesis.

In the present study, emergent interpretations were generated more frequently in the metaphor sentence conditions than in the literal condition. Moreover, the emergent interpretations were more unrelated to the topic and vehicle words for the novel metaphor than for the familiar metaphors and the literal sentences. These results suggest that novel metaphors, where the degree of dissimilarity is greater between the topic and the vehicle, yields interpretations that are more creative in terms of both their quantity and their quality. On the other hand, within prior studies, it has been suggested that one important factor of creative insight is the distinctiveness of the two concepts blended when insight occurs (Nagai et al., 2009).

Attentional shifts toward the topic words were observed about 8 s prior to the generation of emergent interpretations. It is unlikely that visual attention directly reflects semantic processing of the vehicle and topic words, such as semantic attention, formation of ad hoc categories, or combining of the distinct concepts. Nonetheless, it is not unreasonable to assume that the enhanced visual attention reflects some cognitive processes specifically associated with the generation of creative interpretations, such as semantic encoding (Huettig and Altmann, 2005), or representational changes (Knoblich et al., 2001).

The present study has quantitatively identified a period that may be important for creative insights. At the same time, the present study arises one critical question: what is occurring during the period starting with the attentional shift and ending when a response has been generated? One possible approach to finding an answer would be to monitor brain activity during this period, and to examine the functional dynamics that bring about the creative thoughts.

Conflict of Interest Statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Acknowledgments

Footnotes

References