In the eye of the beholder
An eye-tracking study on Deliberate Metaphor TheoryName: Siméon T.A. Lahaije Student number: 6183743
Study/track: Research Master Brain and Cognitive Sciences, Cognitive Science Institute: Universiteit van Amsterdam, Amsterdam
Supervisors: dr. Roosmaryn Pilgram, prof. dr. Gerard J. Steen Co-assessor: dr. Sible J. Andringa
2 Abstract
Metaphors are very common in our everyday speech, appearing ‘all over the place’, yet we do not seem to cognitively process all linguistically identified metaphors as actual metaphors. This is what is called the “paradox of metaphor”. A proposed solution to this paradox is adding a third dimension of communication to the two-dimensional (thought and language) model of metaphor, as outlined in Deliberate Metaphor Theory (DMT). The present study looks into the cognitive processing of deliberate metaphors through eye-tracking and memory recall, by a robust set-up that can easily be reapplied to other metaphors and languages. Results confirm a cognitive demarcation between non-deliberate and non-deliberate metaphors, and thereby give insight into how non-deliberate metaphors are processed. These results provide interesting new possibilities for future studies. Based on these results, further considerations for metaphor theory in general and DMT in particular are provided. 1. Introduction
When Rey is taught in the ways of the Force by Luke Skywalker in Star Wars: The Last Jedi, she is told she must close her eyes and ‘reach out’. Instinctively, she reaches out her hand into the air before her. Luke teases her at first, making her think she is connecting with the Force, but eventually slaps her hand. Rey opens her eyes in shock, but quickly realizes her mistake: she places her hand on her chest and asks: “you meant…?” A stern look from Luke reveals that indeed, he meant for her not to literally reach out, but metaphorically, “with your emotions”. Metaphors like these are not just used on purpose by fictional Jedi masters, but quite often by teachers, scientists, psychologists, journalists and others alike to give greater insight into concepts and ideas that are at first (too) difficult to grasp. While metaphors are more common than may be expected in our everyday speech, they are often not used and discussed so explicitly, and are less notable unless specifically paid attention to, such as the verb “grasp” in the previous sentence (Lakoff & Johnson, 1980; Steen, 2008).
This understanding of the use of metaphors, and by that, the contemporary focus in metaphor research was greatly influenced by the work Metaphors We Live By by Lakoff & Johnson (1980), who provided extensive linguistic evidence to argue that metaphors as expressed are representative of metaphors of thought. That is, a comparison between two things is not just made linguistically, but also on the conceptual level, hence the name of their theory: conceptual metaphor theory (CMT). In this theory, metaphors are understood to not just be esthetical or convenient for understanding difficult concepts, but sometimes even necessary to understand some abstract notions and concepts, and ‘not just nice’ (Ortony, 1975). For example, the abstract concept of love is not easy to grasp, but by comparing it with something as concrete as a journey, through the conceptual metaphor LOVE IS A JOURNEY, expressed for example by saying “We are taking the next step in our relationship”, we are able to think and communicate effectively about love.
Throughout the following decades, CMT has guided metaphor research by enlarging its scope from language to thought, inciting researchers to conduct theoretical and experimental
psycholinguistic/cognitive linguistic research on metaphors. However, CMT is not without critique (Gibbs, 2011): in recent years, more attention has gone to the cognitive processing of metaphors (and whether or not they are all processed by comparison) and to the aspect of communication. For example, Bowdle & Gentner (2005) discuss how metaphors are understood by some theoretical accounts as comparisons, and by others as categorizations. They propose a combination of both accounts by the Career of Metaphor Theory, which states that novel metaphors are understood
3
through comparison because the base terms of the metaphor are not (yet) associated with a metaphorical category, whereas the base terms of a conventional metaphor are, and are thus understood through categorization (Bowdle & Gentner, 2005; Dulcinati et al., 2014), the latter of which is a cognitively less demanding task. Through repeated use, the base terms of a metaphor may become associated with the metaphorical category and this way, a novel metaphor may become conventional. This ‘career’ of a metaphor can already happen over the course of a single experiment, as Bowdle & Gentner (2005) found.
A second development is a focus on the communicational aspects of metaphor (Cameron, 2003; Reijnierse et al., 2017; Steen, 2008, 2015), which was not much focused on in the framework set by CMT, as CMT created a two-dimensional model of metaphor, basing it in language and thought (or concepts). However, there can be discrepancies between these two dimensions: some conventional metaphors are metaphorical linguistically and can be identified as such (for example by the MIP or MIPVU methods (Steen et al., 2010), but are not metaphorical in thought (so is argued), in the sense that no active cross-domain mapping is set up between the target and source domains of the metaphor. This came to light by analysis of the MIPVU text analyses results (Steen, 2008), that showed that the majority of conventional metaphors is expressed as a metaphor and not as a simile, which (by the Career of Metaphor Theory by Bowdle & Gentner (2005)) means they are processed through categorization, rather than comparison. This in turn means they are not actively processed as metaphors, yet they are metaphors linguistically, and are identified as metaphors as such, leading to a “paradox of metaphor” (Steen, 2008).
Since the paradox is caused by a discrepancy along the two dimensions (language and thought) of the two-dimensional model of metaphor, it is likely not to be solved along those lines. Rather, Steen (2008, 2011b) proposes a third dimension, that of communication, to be added to resolve the paradox. The communicative aspect or property of a metaphor explains how the user of the metaphor is using it: either deliberately, or not deliberately. By using a metaphor deliberately, a speaker actively uses the metaphor as such and invites the addressee of their words to switch to a different domain and compare that to the previous domain, making an active comparison and setting up a cross-domain mapping. Naturally, this means that novel metaphors are necessarily deliberate as the domains of the novel metaphor must be actively compared to understand the meaning of this new metaphor (in line with the Career of Metaphor Theory). However, conventional metaphors may either be used deliberately or not deliberately (table 1). When used deliberately, a conventional metaphor is expanded upon or played with (as in a pun) so that attention is drawn to the comparison between source and target domain. In the example of table 1, the conventional metaphors A BODY IS A TEMPLE is made deliberate by actively adding words from the source domain of “temple”, and using these in such a way they can also refer body actions, drawing an active comparison between “body” and “temple”. By this definition of deliberate metaphor, Steen’s DMT creates a clearer framework than the distinction that Cameron (2003) made between conventional and deliberate metaphors. While Cameron mentions briefly that conventional metaphors may be used deliberately, she mostly distinguishes between conventional and deliberate metaphors, and how a deliberate metaphor may become conventionalized (similar to the Career of Metaphor Theory (Bowdle & Gentner, 2005)). It is however the suggestion of DMT that conventional metaphors can be used deliberately that provides interesting insights into the role of communication, and creates possibilities for experimental research, such as this study.
4
Table 1. Overview of the interaction between the dimensions of communication and concept for metaphors according to DMT. This table is a simplified version of Table 1 of Steen (2011b), omitting the dimension of language. The metaphors used are A BODY IS A TEMPLE and A BOOK IS A PILL1.
Conventional Novel
Non-deliberate My body is a temple. I eat well and I exercise regularly.
-not possible- That new book is a pill. It will take at
least three full weekends to read it.
-not possible- Deliberate I treat my body like a temple,
cleansing it vigorously and keeping out any pagan filth.
The human body is basically a cactus: it can withstand harsh environments, but it always needs water.
You better have an appetite, because this book is a fat pill; don’t choke on it!
My new book is an elephant: big, loud, impossible to ignore, and will leave a big impression.
DMT is not without controversy: some argue that DMT sets back the progress that has been made in metaphor research in the past decades (Gibbs & Chen, 2017), in that DMT would make metaphors something special again, used only by a select few, against the insight (and subsequent progress) first made by Lakoff & Johnson (1980) that the use of metaphors is commonplace, widespread and egalitarian, and that DMT postulates that only some metaphors are truly metaphors. Steen (2017) responds to this that DMT does not consider metaphors ‘ornamental’ or ‘deviant’, but instead builds upon the framework of CMT by adding a third dimension, through which some metaphors are deliberately used as metaphors.
It stands to reason that more empirical research is needed to gain further insight in the existence, use and cognitive processing of deliberate metaphors (compared to non-deliberate metaphors). Steen himself agrees with this point (made by Gibbs (2015a) and Gibbs & Chen (2017)), but points out that such experiments need to be set up carefully and diligently, so as not to lead to ‘hasty and incorrect conclusions’ (Steen, 2017).
So far, a limited but growing number of studies have been carried out on DMT. These have mostly focused on examining the use of deliberate metaphor in various discourse settings, such as college lectures (Beger, 2011, 2016), political discourse (Perrez & Reuchamps, 2014) and corporate branding (Ng & Koller, 2013). Musolff (2016) conducted an experiment that found cross-cultural variation between participants when they interpreted NATION IS A BODY and NATION IS A PERSON
metaphors, showing a deliberate and not an automatic and subconscious interpretation of metaphor in specific circumstances. Reijnierse (et al., 2015) examined how extended metaphors (understanding extension as a form of deliberateness) influence political opinions. Tay (2017) examined deliberate metaphors in different discourse and found six categories of deliberate metaphor features
(elaboration, discourse signal, analogy, stark novelty, topic-triggering, and repetition) which were used differently across the various discourses.
1 There may be some discussion about whether or not A BODY IS A TEMPLE is a conventional metaphor in English (and Dutch) or not, yet because it provided a clear and unambiguous example (direct metaphor, and no metaphoric related words in the non-deliberate condition) we decided to use it as an example. Additionally, we included the metaphor A BOOK IS A PILL which a non-controversially conventional metaphor in Dutch.
5
However, since there is no uniform definition and consequently an objective identification procedure for deliberate metaphors, comparisons between studies on DMT and setting up experiments on DMT is difficult (Reijnierse, et al., 2017). A solution for this problem was has been provided by Reijnierse (et al., 2017) with DMIP, an extension of the MIP to identify potential deliberate metaphor. The authors specifically state that since the method is a semiotic approach, it can only identify potential deliberateness, as such an analysis cannot identify what happens in the speaker’s mind when producing the metaphor, or between users of the metaphor.
Gibbs (2015b) conducted an experiment were participants were questioned on (supposedly) deliberately used metaphors in a written exchange between friends on one’s marriage, such as how they thought the speaker was about his marriage. However, Steen (2015) has argued that this experiment did not correctly test DMT, as its assumptions about DMT were incorrect, in particular the way that the deliberate stimuli were set up: a conventional metaphor was made deliberate by adding a discourse signal such as ‘well’ or ‘it is like’. Steen (2015) argues that such discourse signals (at least not in the way they were used in this experiment) do not draw attention to the source domain, creating an active cross-domain mapping, which is at the core of deliberate metaphors. Furthermore, participants were interviewed on the possible intentions and perceptions of the speakers in the conversations in the stimuli, which does not provide insights into whether or not deliberate metaphors set up active cross-domain mappings cognitively. Finally, the use of a single metaphor may have been a confounding factor, and repeated use of the same metaphor (and text) could have created a repetition effect.
To learn more about a cross-domain mapping, that is, an online comparison between the target and source domains of the metaphor, and by that, test DMT, a sensible way would be to use a method that can measure participants’ active comparisons between words. Eye-tracking is a method that measures eye-movements (in and between words), which are considered to be indicative of
cognitive attention (Raney et al., 2014; Rayner, 1998) and is thus a natural method to test for active comparisons between words. Next to that, an eye-tracking study is relatively easy to set up, carry out and analyze, compared to for example EEG and fMRI. Finally, this (kind of) method does not depend on interpretation or feedback from participants, and therefore provides more objective results than subjective participant input. Eye-tracking has been used before as a method to develop a better understanding of the cognitive processing of metaphors (Ashby, 2017; Olkoniemi, 2016; Rubio-Fernández et al., 2016), metonymies (Frisson & Pickering, 1999), and reading times of metaphor (without eye-tracking) (Blank, 1988; Bowdle & Gentner, 2005; Gentner & Wolff, 1997; Janus & Bever, 1985; Rubio-Fernández et al., 2016). For an initial psycholinguistic study on DMT, we thus believe eye-tracking is a good way to gain new insights on the processing of deliberate metaphors.
In the present study, we will focus on the cognitive processing of deliberate metaphors compared to non-deliberate metaphors (and also source domain words in a literal context for controls) through eye-tracking and by memory recall. Because deliberate metaphors are theorized to have a higher load on working memory (Steen, 2011a, 2017) and because deliberate metaphors force the reader to make an active comparison (which takes more time and effort than categorization (Bowdle &
Gentner, 2005; Dulcinati et al., 2014), we expect participants to take more time to interpret and understand deliberate metaphors compared to literal statements and non-deliberate metaphors. This would be reflected by higher values for the different eye-tracking measurements (e.g. first gaze, total gaze) for the deliberate metaphors. Additionally, since deliberate metaphors require active
6
cross-domain mapping, we expect deliberate metaphors to have a more lasting impression on participants. This may be compared to how Krennmayr (et al., 2014) found how novel metaphors made a significantly stronger impression on participants compared to conventional metaphors due to the former requiring active comparisons. We therefore also expect to find a higher memory recall for questions on deliberate metaphors compared to literal statements and non-deliberate metaphors. We can thus formulate the following hypotheses:
H0: there is no difference in absolute and relative eye-movement behavior (first pass time, regressions, regression path duration, second pass time, total fixations and total gaze duration) for target word and target sentence, and significant differences in total trial time and memory recall in all three conditions (literal, non-deliberate metaphor, deliberate metaphor), indicating no difference in cognitive effort.
H1A: there is a significant difference in absolute and relative eye-movement behavior (first pass time, regressions, regression path duration, second pass time, total fixations and total gaze duration) for target word and target sentence, and significant differences in total trial time and memory recall between the literal and deliberate metaphor condition, indicating more cognitive effort to process deliberate metaphors compared to literal statements. H1B: there is a significant difference in absolute and relative eye-movement behavior (first pass time, regressions, regression path duration, second pass time, total fixations and total gaze duration) for target word and target sentence, and significant differences in total trial time and memory recall between the non-deliberate and deliberate metaphor condition, indicating more cognitive effort to process deliberate metaphors compared to non-deliberate metaphors.
2. Materials and methods 2.1 Participants
A total of 15 native Dutch participants (5 men, 10 women) were tested, aged 19 to 31 years (M = 24.4, SD = 3.44). Participants were recruited through posters, social media and personal invitation. All participants were receiving or had received higher education (HBO/WO) in the Netherlands. No participant reported suffering from dyslexia or had other problems with reading or language processing, nor had any serious eye-sight problems that could affect reading. A small compensation in the form of € 4,- was provided. The experiment was approved by the Ethics Committee of the Faculty of Humanities of the University of Amsterdam.
2.2 Stimuli
To select conventional metaphors, we created a questionnaire consisting of 35 existing (confirmed by the Van Dale dictionary, as per MIPVU (Steen et al., 2010)) and 35 newly created metaphors in Dutch, presented randomly, that 13 participants were asked to rate on familiarity on a seven-point scale (similar to Rubio-Fernández et al., 2016). Participants were all (former) college students between the ages of 20 and 30. The 27 highest ranked metaphors, and therefore most conventional ones, (all with a score of 3,5 or higher) were selected for the experimental stimuli (see appendix A).
7
Experimental texts consisted of three sentences, whereby the word of the source domain of the metaphor appeared in a sentence that was the same across all three conditions, while the word of the target domain was presented in the first sentence in the non-deliberate and deliberate metaphor conditions (see table 2 for an example). Additionally, words from the source domain of the metaphor also appeared in the first and third sentence in the deliberate metaphor condition. This set-up means, in terms of the six features of deliberate metaphor as identified by Tay (2017), that the stimuli in this experiment exclusively used the features of elaboration and topic-triggering. Stark novelty was not possible, due to the fact that all metaphors had to be conventional, and due to the structure of the stimuli, the features of analogy, discourse signal and repetition were not possible either or would be to out of line with the rest of the stimuli.
Table 2. Example of the set-up of stimuli per condition in Dutch with English translation in italics. This example is based on the conceptual metaphor FIGHT IS WAR, whereby the sentence ‘It is war’ is the middle sentence in all three conditions and therefore used for eye-tracking analysis. The word ‘war’ is similarly used for eye-tracking analysis. It should be noted that each sentence appeared fully on one line on screen in the experiment.
Literal Non-deliberate Deliberate
First sentence
De kanonnen knallen en er wordt geschoten.
The cannons are blasting and shots are fired.
De politieke leiders hebben grote ruzie. The political leaders are having a big fight.
Mijn moeder blijft neutraal in onze heftige ruzie. My mother stays neutral in our fierce fight.
Second sentence Het is oorlog. It is war. Het is oorlog. It is war. Het is oorlog. It is war. Third sentence
Er is overal groot gevaar. There is great danger everywhere.
Er wordt veel geschreeuwd.
There is a lot of yelling.
Ze eist een staakt-het-vuren.
She demands a ceasefire.
The 27 sentences that included the source domain word appeared as the middle sentences in all three conditions, leading to 81 short texts (see Appendix B). All texts were written in the present tense and easy to understand. They were checked for aptness and how natural they sounded by open commentary by four of the participants who filled out the conventionality questionnaire, and adapted based on their nearly unanimous commentary. The stimuli were judged on deliberateness by two experienced members of the Metaphor Lab department, and were identified as potentially deliberate by DMIP (Reijnierse et al., 2017).
To prevent a repetition effect a blocked design was used for the experiment, so that only nine sentences per condition were read per participant. This lead to three versions, which were all mirrored, for a total of six versions used. Sentences were set in a pseudorandom order so that no condition appeared twice in a row. Thus, the first four target sentences in version one appeared for example as literal/deliberate/literal/deliberate, in version two as
non-deliberate/deliberate/non-deliberate/literal, and in version three as
deliberate/literal/deliberate/non-deliberate. While the target sentences differed in word and syllable length, they were divided amongst conditions so that all conditions shared a similar number of words and syllables per version of the experiment, and so that the three longest sentences (four words, six syllables) appeared each in one condition. The majority (15) of the target sentences were of an “It is
8
X” structure, but these were distributed so that no more than two of these sentences appeared in a row. The average number of words and syllables per trial was nearly equal per condition.
2.3 Procedure
Eye movements were recorded with the Eyelink 1000 eye tracker by SR Research Ltd. Participants sat between 50-60 cm away from the screen, under which the eye tracker system was mounted.
After providing background information, participants were instructed on the experiment and told they had to casually read texts. They were instructed to make sure they understood what they read, and if so, move on to the next text by pressing a button. It was made clear they did not have to agree or disagree with what they read and simply read the texts casually, but that there would be a
questionnaire afterwards to measure what they had remembered passively. At no point prior to reading the stimuli did participants know the experiment was about the cognitive processing of (deliberate) metaphors, so as to make sure they would not process the non-deliberate metaphors deliberately.
After calibrating the eye-tracking equipment to the characteristics of the participant’s eyes,
participants once more received instructions by on-screen text for clarity and to adapt to the screen, followed by three practice texts and the experimental texts. All texts were presented in Arial, font size 18, in black color on a white background on a 1280x1024 screen. Texts were presented in full, with each sentence on a separate line, as Gong & Ahrens (2007) found that conceptual mapping only occurs when a text including a metaphor is presented paragraph-style, and not line by line. Between each text a ‘+’-sign was shown in the upper-left corner at the same position of the first letter of the next text for calibration and to prevent participants from reading ahead of the text.
Afterwards, participants were questioned orally on nine randomly selected sentences to see if they had payed attention as instructed, and to measure what they had remembered based on condition. These questions referred to the source domain word in the target sentence, for example: “The political leaders are having a big fight, it is?” or “The cannons are blasting and shots are fired, it is?” (War) (see table 2).
2.4 Analysis
We used six different measurements for eye-tracking analysis: first gaze time (duration of initial word processing), regressions (percentage of regression, which is the percentage of eye-movements to the left of the interest area, indicating re-reading earlier parts), regression path duration (time from first fixation up to the first movement right from the interest area, indicating time to integrate a specific part of the text), second pass time (re-reading), total gaze duration (total attention time) and total number of fixations (Clifton et al., 2007; Rubio-Fernández et al., 2016) (see figure 1). We measured both the source domain word in the middle sentence, as well as the entire middle sentence2. Measurements were analyzed both in absolute numbers, while the first and second pass time, regression path duration and total gaze time were also compared relatively (that is, the values of these measurements were divided by total trial time, in other words, we calculated the percentage of the total trial spent on a specific part of the trial (Vansteenkiste et al., 2015)) to control for
9
differences in reading styles between participants. Total trial time was measured and compared as well.
Figure 1. Overview of eye-tracking measurements. The sentence is “Look at that dog walking there.”, orange circles represent fixations, green lines the movements from one fixation to the next, and numbers the different fixations in order. Time in ms is not reported, but is not necessary to understand the calculation of measurements. For the word “dog”, the first gaze time is time of fixation 4, second gaze time is time of fixations 7 + 10, total gaze duration is time of fixations 4 + 7 + 10, total fixations is 3, regression path duration is time of fixations 4 + 5 + 6 + 7, and percentage of regressions is 1/3.
Analysis of the target word was only included when at least one fixation was made within the interest area of the target word, and analysis for the target sentence followed the same rule. Due to their separate analysis, it was possible that we had measurements for the target sentence, but not for the target word. Continuous fixations were combined into one, so as to measure gaze, which is common in linguistic eye-tracking research (Rayner, 1998; Rubio-Fernández et al., 2016).
Measurements could include 0, for example the second gaze of a word when a participant only fixated on it once, or total gaze duration if the participant did not fixate on that word at all. The regression path duration was per definition (Raney et al., 2014) only calculated if a regression path started after the first gaze of the word; thus, a regression path following a second or third run was not calculated and included for this measurement. Fixations shorter than 80 ms were excluded (Frisson & Pickering, 1999).
We compared the different conditions of sentences (similar to Janus & Bever (1985)) and the different condition results per participant with each other (Bowdle & Gentner, 2005; Rubio-Fernández et al. (2016)), using a repeated measures ANOVA (and possible post-hoc tests). All
measurements per sentence were checked for correlation with familiarity scores and word frequency of the target word3, to check for the influence of these factors on reading times.
For the questionnaire, we compared the total number of correct answers per condition, and also checked for correlations between total correct answers and the different eye-tracking measurements of the relevant trials, to see if differences between correct answers per condition were due to longer looking times instead of the conditions.
3
We used the DISCO (extracting DIstributionally related words using CO-occurrences) Java tool and the accompanying Dutch database to calculate word frequency.
10 3. Results4
Due to eye-tracking equipment failure, three participants were excluded from eye-tracking analysis, but their answers to the questionnaire were registered (analyses including and excluding these participants are provided). Each excluded participant was tested on a different version (the different ordered sets of stimuli), leaving four participants per version. Data on total trial time was measured for every participant; analyses including and excluding these participants are provided.
3.1 Absolute eye-movement behavior
No significant difference between conditions was found for any measurement for both the target word and target sentence on absolute times. First gaze target word: F(2, 22) = .069, p = .93; second gaze target word: F(2, 22) = .47, p = .63; total gaze target word: F(2, 22) = .38, p = .69; total fixations target word: F(2, 22) = .004, p = .996; regressions target word: F(2, 22) = .48, p = .63; regression path target word: F(2, 22) = .55, p = .58. First gaze target sentence: F(2, 22) = 1.21, p = .32; second gaze target sentence: F(2, 22) = 1.33, p = .28; total gaze target sentence: F(2, 22) = 2.17, p = .14; total fixations target sentence: F(2, 22) = .25, p = .78; regressions target sentence: F(2, 22) = .37, p = .698. The measurements per condition per sentence were also calculated and analyzed by a repeated measures ANOVA, but revealed no significant differences either: first gaze target word: F(2, 52) = .69, p = .51; second gaze target word: F(2, 52) = 1.12, p = .34; total gaze target word: F(2, 52) = .27, p = .77; total fixations target word: F(2, 52) = .080, p = .92; regressions target word: F(2, 52) = 1.15, p = .33; regression path target word: F(2, 52) = 1.08, p = .35. First gaze target sentence: F(2, 52) = .69, p = .51; second gaze target sentence: F(2, 52) = .74, p = .48; total gaze target sentence: F(2, 52) = 1.32, p = .28; total fixations target sentence: F(2, 52) = .29, p = .75; regressions target sentence: F(2, 52) = .895, p = .42.
A moderate correlation (Pearson) was found between conventionality scores and the total gaze for the target word in the non-deliberate metaphoric condition (r(25) = -.38, p = .049), first gaze for the target sentence in the non-deliberate metaphoric condition (r(25) = -.43, p = .027) and the total gaze for the target sentence in the non-deliberate metaphoric condition (r(25) = -.43, p = .025).
Spearman’s rho did not reveal any significant correlations between conventionality score and different eye-tracking measurements.
3.2 Relative eye-movement behavior
We also calculated relative eye-movement behavior times, by dividing first gaze, second gaze, total gaze and regression path times for the target word, and first gaze, second gaze and total gaze times for the target sentence by the total trial time. At word level, repeated measures ANOVA analyses revealed no significant differences between conditions for the first gaze target word: F(2, 22) = .297, p = .75; second gaze target word: F(2, 22) = .54, p = .59; total gaze target word: F(2, 22) = .81, p = .46; regression path target word: F(2, 22) = 1.93, p = .17; second gaze target sentence: F(2, 22) = 1.22, p = .31. However, at sentence level, a significant difference was found for the first gaze of the target sentence, F(2, 22) = 3.81, p = .038. Post-hoc comparisons revealed a significant difference between the non-deliberate metaphoric (M = 15.1, SD = 2.27) and deliberate metaphoric (M = 13.2, SD = 2.32) conditions (t(11) = 4.18, p = .002, d = 1.21) (figure 2). A significant difference was also found for the
11
total gaze of the target sentence, F(2, 22) = 5.15, p = .015. Post-hoc comparisons revealed a
significant difference between the literate (M = 15.9, SD = 2.25) and non-deliberate metaphoric (M = 17.7, SD = 2.03) conditions (t(11) = -2.39, p = .036, d = -.69) and between non-deliberate metaphoric (M = 17.7, SD = 2.03) and deliberate metaphoric (M = 15.3, SD = 2.43) conditions: t(11) = 3.97, p = .002, d = 1.15 (figure 3). This means that, participant-wise, a significantly larger portion of the trial was spent on the first gaze and the total gaze of the target sentence in the non-deliberate metaphor condition (compared to the deliberate metaphor condition (and literal condition)).
Figure 2. Average relative first gaze for target sentence in percentage of total trial time, participant-wise (n = 12). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*).
Figure 3. Average relative total gaze for target sentence in percentage of total trial time, participant-wise (n = 12). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*).
12
The relative measurements per condition per sentence revealed no significant difference between conditions for the first gaze target word: F(2, 52) = 1.49, p = .23; second gaze target word: F(2, 52) = 1.10, p = .34; total gaze target word: F(2, 52) = .84, p = .44; regression path target word: F(2, 52) = 2.76, p = .072; second gaze target sentence: F(2, 52) = .54, p = .59. A significant difference was found for the first gaze of the target sentence, F(2, 52) = 4.57, p = .015. Post-comparisons revealed a significant difference between non-deliberate metaphoric (M = 15.1, SD = 3.76) and deliberate metaphoric (M = 13.1, SD = 2.64) conditions (t(26) = 2.84, p =.009, d = .55) (figure 4). A significant difference was also found for the total gaze of the target sentence, F(2, 52) = 4.56, p = .015. Post-hoc comparisons revealed a significant difference between the non-deliberate metaphoric (M = 17.6, SD = 4.07) and deliberate metaphoric (M = 15.3, SD = 2.90) conditions (t(26) = 2.84, p = .009, d = .55) (figure 5). This means that, sentence-wise, a significantly larger portion of the trial was spent on the first gaze and the total gaze of the target sentence in the non-deliberate metaphor condition (compared to the deliberate metaphor condition).
Figure 4. Average relative first gaze for target sentence in percentage of total trial time, sentence-wise (n = 27). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*).
13
Figure 5. Average relative total gaze for target sentence in percentage of total trial time, sentence-wise (n = 27). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*). A moderate correlation was found between conventionality scores and the first gaze for the target word in the deliberate metaphoric condition (r(25) = -.39, p = .045), and between conventionality scores and the total gaze for the target sentence in the non-deliberate metaphoric condition (r(25) = -3.90, p = .044). Spearman’s rho revealed a significant correlation between conventionality scores and total gaze time for the target sentence in the non-deliberate metaphoric condition (rs(25) = -.45, p = .020.
3.3 Total trial time
We calculated total trial per condition per participant, and per condition per sentence, once including and once excluding the participants who were excluded from the eye-tracking analysis. Per
participant, excluding participants with missing eye-tracking data (n = 12), we found a significant difference between conditions: F(2, 22) = 8.25, p = .002. Post-hoc comparisons revealed a significant difference between literal (M = 4358, SD = 1150) and deliberate metaphoric (M = 4780, SD = 1261) conditions (t(11) = -.31, p = .010, d = -.89), and between non-deliberate metaphoric (M = 4454, SD = 1235) and deliberate metaphoric (M = 4780, SD = 1261) conditions (t(11) = -4.03, p = .002, d = -1.16) (figure 6). For all participants (n = 15), the assumption of sphericity was violated (2(2) = 19.7, p < .001), thus degrees of freedom were corrected by Greenhouse-Geisser estimates ( = .56), revealing a significant difference between conditions: F(2, 28) = 7.34, p = .013. Post-hoc comparisons revealed here too a significant difference between literal (M = 4537, SD = 1096) and deliberate metaphoric (M = 5240, SD = 1573) conditions (t(14) = -2.72, p = .017, d = -.70), and between non-deliberate
metaphoric (M = 4720, SD = 1242) and deliberate metaphoric (M = 5240, SD = 1573) conditions (t(14) = -3.08, p = .008, d = -.796) (figure 7). This means that, participant-wise, a significantly longer time was spent on trials in the deliberate condition compared to both the non-deliberate and literal condition.
14
Figure 6. Average total trial time in ms, participants-wise (n = 12). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = non-deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*).
Figure 7. Average total trial time in ms, participants-wise (n = 15). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = non-deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*).
Per sentence for n = 12, we did not find a significant difference (F(2, 52) = 1.62, p = .21). However, for all participants (n = 15) we found a significant difference between conditions: F(2, 52) = 4.34, p = .018. Post-hoc comparisons revealed a significant difference between literal (M = 4537, SD = 720) and deliberate metaphoric (M = 5240, SD = 955) conditions (t(26) = -2.59, p = .015, d = -.499), and
between non-deliberate metaphoric (M = 4720, SD = 835) and deliberate metaphoric (M = 5240, SD = 955) conditions (t(26) = -2.10, p = .045, d = -.40) (figure 8). This means that, sentence-wise, a
15
significantly longer time was spent on trials in the deliberate condition compared to both the non-deliberate and literal condition.
Figure 8. Average total trial time in ms, sentence-wise (n = 27), all participants (n = 15). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*).
3.4 Questionnaire
We found a significant difference between conditions on correct answers for n = 15: F(2, 28) = 7.31, p = .003). Post-hoc comparisons revealed significant differences between literal (M = 2.20, SD = .86) and non-deliberate metaphoric (M = 1.20, SD = .78) conditions (t(14) = 3.87, p = .002, d = 1), and between non-deliberate (M = 1.20, SD = .78) and deliberate metaphoric (M = 2.13, SD = .74)
conditions (t(14) = -3.29, p = .005, d = -.85) (figure 9). For n=12, we found similar results: a significant differences between conditions (F(2, 22) = 4.28, p = .027), with post-hoc comparisons revealing significant differences between the literal (M = 2.08, SD = .90) and non-deliberate metaphoric (M = 1.17, SD = .84) conditions (t(11) = 2.93, p = .014, d = .85) and between non-deliberate (M = 1.17, SD = .84) and deliberate metaphoric (M = 2.00, SD = .74) conditions (t(11) = -2.59, p = .025, d = -.75) (figure 10). This means that stimuli from both the deliberate metaphor and literal condition were
significantly better recalled than stimuli from the non-deliberate condition.
Checking for correlations between questionnaire scores and absolute eye-tracking measurements, we found a significant correlation between the number of correct answers on the literal condition questions and first gaze time for the target word for the literal condition (r(10) = .596, p = .041; rs(10) = .67, p = .018)). No correlations were found for the non-deliberate and deliberate conditions. For correlations between questionnaire scores and relative eye-tracking measurements, a significant correlation was found between the number of correct answers on the non-deliberate condition questions and the total gaze time for the target word on the non-deliberate metaphoric condition (r(9) = -.64, p = .033; rs(9) = -.698, p = .017). No correlations were found for the literal and deliberate conditions.
16
Figure 9. Average correct answers on questionnaire per condition (n = 15). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*).
Figure 10. Average correct answers on questionnaire per condition (n = 12). LIT = literal condition, NDM = non-deliberate metaphoric condition, DM = deliberate metaphoric condition. Error bars show +/- SD. Significant differences (p < .05) are indicated by (*).
3.5 Word frequency
We checked for correlations between eye-movement measurements and the frequency of the target word. For absolute measurements5, we found a significant correlations between first gaze time for the target word in the non-deliberate metaphoric condition and word frequency (rs(25) = -.49, p =
5
For n = 12. For n = 15, only total trial times were available, but no correlations were found between these and word frequency.
17
.009), between first gaze time for the target sentence in the non-deliberate metaphoric condition and word frequency (r(25) = -.50, p = .008; rs(25) = -.56, p = .003) and between second gaze time for the target sentence in the non-deliberate metaphoric condition and word frequency (r(25) = .46, p = .017). For relative measurements, we found significant correlations between first gaze for the target word in the literal condition and word frequency (r(25) = -.46, p = .015; rs(25) = -.48, p = .011), between first gaze for the target sentence in the non-deliberate condition and word frequency (r(25) = -.56, p = .003; rs(25) = -.48, p = .011), and between the second gaze for the target sentence in the non-deliberate metaphoric condition and word frequency (r(25) = .399, p = .039).
4. Discussion
In this research we aimed to test and gain more insight in DMT by comparing how participants read and recalled metaphors used in deliberate and non-deliberate ways, and the source domain word of the metaphors used in a literal sense. We expected to find higher values for the different eye-tracking measurements, and higher recall of deliberate metaphors. Not all results confirmed our hypotheses: for the relative first and total gaze of the target sentence the results were opposite our expectations, whereas the results of the total trial time did confirm our hypotheses. For the memory recall, the results both agreed and disagreed with our expectations. However, we are still able to reject H0, and we would argue that these results do confirm a cognitive demarcation between deliberate and non-deliberate metaphors, and in fact provide new insights into the cognitive processing of deliberate metaphors, opening new ventures for future research.
As for the results, we found significant differences on the relative first gaze of the target sentence, on the total gaze of the target sentence and on total trial time. For the first and total gaze, the results were opposite of what was expected: the target sentence in the non-deliberate condition was significantly longer gazed at first and in total compared to the deliberate metaphoric condition. We can understand these findings by looking at how the target sentence compares to the rest of the text. For the literal condition, the target sentence follows the rest of the text in that all words are used in a literal sense, and can therefore be said to be predictable. Predictability has an effect on looking time in that predictable words require less looking time compared to unpredictable words (Clifton et al., 2007). For the deliberate condition, it can be argued that the target sentence (or at least the use of a metaphor or source domain word) is predictable as well, as a source domain word, and thereby part of the metaphor, had already been introduced in the first sentence. However, this is not the case for the non-deliberate metaphoric condition, where the use of a source domain word in the target sentence (setting up a metaphor) could not be predicted, leading to a significantly longer relative looking time for the first gaze on the target sentence. Adding to that, we may draw a comparison with findings of Lai (et al., 2009), who studied the processing of novel and conventional metaphors, and literal sentences with EEG, and compared the results with different models of
metaphor compared to literal processing such as the direct and indirect access view, and the gradient salience model6. They found an initial difference in ERP’s between literal and metaphoric (both novel and conventional) expressions, whereas later ERP’s showed similar results between literal and conventional metaphor texts on one hand, and novel metaphors on the other. This could be
6
The direct access view holds that both the literal and metaphoric meanings of a word are both available at the same time, whereas the indirect access view holds that the metaphoric meaning only becomes accessible after the literal meaning is rejected (Lai et al., 2009). The graded salience model (Giora, 2003) states the ‘salience’ of an expression influences its processing speed, which means that conventional metaphors are processed faster than novel metaphors as they are more salient.
18
understood in that the literal meaning of a conventional metaphor is first rejected and a metaphoric meaning is then retrieved, which takes cognitive effort, or because multiple meanings were retrieved together at once instead of only the literal one. The initial discrepancy can then be understood as a difference in the meaning retrieval process, whereas the later discrepancy is due to a difference in the meaning integration process. Lai (et al., 2009) point out that these results are in agreement with the Career of Metaphor Theory (and the indirect access view) as is DMT. Their results are also in agreement with ours, in that meaning retrieval took more cognitive effort in the non-deliberate metaphor condition, as the metaphoric meaning of the word was unexpected, whereas in the deliberate metaphor condition the metaphor was already (partly) introduced in the first sentence, which made the metaphoric meaning retrieval of the target word easier.
For the relative total gaze time of the target sentence, we can again understand the found results by looking at the entire text and how the processing of the target sentence fits herein. As stated, part of the metaphor is already introduced in the first sentence in the deliberate metaphoric condition, and continues in the third and final sentence. Because the metaphor is more ‘spread out’ it is likely that the processing of the metaphor is spread out as well, and not limited to the target sentence. For the non-deliberate metaphor, it is limited to the target sentence however, and while the metaphors in this condition are processed through categorization instead of comparison (the latter of which takes more time (Bowdle & Gentner, 2005)), the fact this processing takes place in one sentence instead of the whole text can explain why, against expectations, the relative total gaze was longer for the non-deliberate than the non-deliberate condition. A confounding factor in these findings, however, is the fact that metaphor related words were spread over the text in the deliberate condition, specifically the third sentence. Because of this set-up, we did not just check whether or not the target sentence was processed differently (as a non-deliberate or deliberate metaphor) by the manipulation of the first sentence since the final sentence played a role in this manipulation as well. For a complete(r)
understanding, a future study may want to add a deliberate metaphor condition that is only different from the non-deliberate condition in the first sentence and compare that to a condition similar to the deliberate condition in this study, or completely replace the condition.
That the process of comparison takes longer than categorization and thereby that deliberate metaphors take more time to process than non-deliberate metaphors can be seen by the results of the total trial time, where a significantly longer time was found for the deliberate metaphoric condition, compared to both the literal and non-deliberate metaphoric condition. Combining these results with the findings that the relative first and total gaze time were significantly lower for the deliberate compared to the non-deliberate condition, gives further credit to the idea that the processing of the metaphor is ‘spread out’ for deliberate metaphors. This is not surprising, as every (source domain) word that links to the metaphor, and in fact, is what makes the metaphor
deliberate, would be relevant to process for understanding the expression. Consequently, that means the source domain words are not simply pragmatic signals that serve as an extension of the metaphor so as to make it deliberate and force the addressee of the expression to compare instead of categorize, but must necessarily be processed in combination with or as part of the metaphor to understand it. While this particular combination of results provides us with more insight into the processing of deliberate metaphors than anticipated, we cannot infer yet the form of processing. That is, we do not know if the source domain words outside of the target sentence (and in that sense, part of the deliberate metaphor, and in the source domain of the conventional metaphor it is based on, but not part of the conventional metaphor itself) are compared to the target domain word
19
through categorization or comparison, or are simply given more attention to. Alternatively, the source domain word in the target sentence may not be processed through comparison, but the metaphorical meaning of the word may be accessed, which makes accessing the metaphorical meaning of the source domain in the target sentence faster, either because the metaphorical meaning of previous word in the same source domain was accessed, or because that word hinted that the metaphorical meaning should be accessed for the current word. Additionally, we would also have to look into the possible confounding factor of the source domain in the final sentence, which may have manipulated the target sentence differently than if there was only a source domain word in the first sentence. Thus, future studies would have to look into differences in processing of a deliberate metaphor depending on the placements of the features (Tay, 2017) that make a metaphor deliberate. However, eye-tracking would be limited in the insight it can provide in these processes, but fMRI could shine more light on them. fMRI has been applied successfully before to understand the differences in processing metaphor and similes (Shibata et al., 2012), and the influence of context on a possibly metaphoric target sentence, somewhat comparable to the experimental set-up of this study (Prat et al., 2012). As noted earlier, EEG can be insightful as well, as the previously mentioned study of Lai (et al., 2009) showed results in line with ours. The present study therefore shows the limits of eye-tracking research, but still delivers new insights into the processing of deliberate metaphors and by showing its limits, also opens new ventures for future studies.
While significant differences between literal, non-deliberate and deliberate metaphoric stimuli were only found for the relative first and relative total gaze time of the target sentence, it can be argued that these measurements were the most informative, and therefore the most relevant ones in this experimental set-up. Because reading style varied between participants, a second (and sometimes third) gaze of target word and sentence was registered for some of them, while for others only a first gaze was registered: these participants read the target word and/or the target sentence only once. Similarly, a regression path for the target word was out of 27 trials only measured for an average of 7.45 trials (SD = 3.06) per participant. However, a first and total gaze time for the target sentence were calculated for virtually all trials (of all non-excluded participants). The difference in reading styles also means that the absolute gaze time is less informative than the relative gaze time, which eliminates those personal differences. Furthermore, while the assumption of eye-tracking
methodologies and experiments on reading is that the eyes are only focused on one word, and consequently, only process that single word, in practice it turns out that some information from words next to the word that is focused on is received and processed, called parafoveal processing (Schotter et al., 2012). This means that the target word measurements are less reliable than the target sentence measurements, as participants could have processed (in some form) (part of) the target word by focusing on another word in the target sentence. Next to that, it is known that length and frequency have a strong effect on the skipping rate of words, in that shorter words and high-frequency words are often skipped (Clifton et al., 2007). The target sentences consisted mostly of such words, such as ‘hij’ (he), ‘zijn’ (are) or ‘dat’ (that). Combining these findings, we may conclude that the target sentences were not read word by word, but rather at once, which means that (apart from total trial time) the relative measurements of the target sentence (apart from the total trial time) are the most informative eye-tracking measurements for the stimuli used in this experiment. For the memory recall, the results confirmed our hypothesis that questions about deliberate metaphoric stimuli were better recalled than non-deliberate metaphoric stimuli, but the hypothesis about questions regarding literal stimuli was not confirmed. Here, we found that literal stimuli were
20
significantly better recalled that the non-deliberate metaphoric stimuli, almost equal to the deliberate stimuli. A possible explanation for this could be that the literal stimuli that participants were questioned on were more ‘remarkable’, similar to the deliberate stimuli, and were therefore remembered better. While some of the literal stimuli were about animals that were named (so as to line up to the metaphoric stimuli in which an animal was compared to a person (e.g. a bear to a strong man) which could be considered more ‘remarkable’, these stimuli formed a minority of the memory recall, and we therefore do not believe this could explain the found results. Another explanation is that coherent texts are better recalled than non-coherent texts (Krennmayr et al., 2014; McNamara, 2001). While the metaphor used were conventional, their use might still be found to be less coherent compared to the literal texts, as the metaphoric meaning of the source domain word is used instead of the literal meaning. That is, the metaphoric meaning of a word, even if it has reached the point in its ‘career’ (Bowdle & Gentner, 2005) that it is commonly known and easily accessible, is still deviant from the literal meaning (and in for example the indirect access view, accessed after the literal meaning is rejected) which would make the entire text less coherent. For the deliberate stimuli, we may argue that deliberateness of the metaphor may the entire text more coherent, in that source domain words from the metaphor appeared throughout the text, creating a coherent story7. Moreover, a deliberate metaphor requires an active comparison and is thus likely to leave an impression on the participant, in this case on their memory (Krennmayr (et al., 2014). It is also possible that deliberate metaphors are not considered coherent, but that the high scores on this condition are only because of the impression they left. An interaction between these two factors is also possible, but would require further testing with a different experimental set-up. It is possible that there was a ceiling effect for the memory recall; future studies may therefore opt to select more or all stimuli for the memory recall.
We checked for correlations between absolute and relative eye-tracking measurements and conventionality scores, between those measurements and word frequency, and between correct answers on the memory recall and eye-tracking measurements. As far as correlations were found for conventionality scores and word frequency, the results were non-controversial, in that more
conventional metaphors took less time to read that less conventional ones, and that high frequency words were read faster than low frequency words. As only a limited number of non-controversial correlations were found, we do not believe further discussion is required. For the memory recall questionnaire, a significant correlation was found between the scores for the questions on the non-deliberate texts and the relative total gaze of the target word in the non-non-deliberate condition. However, since this is the only correlation of possible value found and also of moderate strength, it may be regarded as an incidental result and we therefore do not consider this relevant to discuss for the results of this study.
For the set-up of this study, we had to select conventional metaphors for the stimuli of this study, at to that goal we had participants fill out a questionnaire in which they rated the conventionality of metaphors on a seven-point scale, similar to Rubio-Fernández (et al., 2016). However, one may question whether or not this method correctly measures the conventionality of metaphors in
7
However, the repeated appearance of a source domain word in the final sentence may have also been a confounding factor, and therefore we may not have just measured the influence of deliberateness of the metaphor in the target sentence (by manipulation through the first sentence) on memory recall of the target word. A repetition of this study may want to include a condition with only a source domain word in the first sentence (as described earlier for the results of the target sentence and total trial time as well).
21
everyday speech, or if it is too subjective. Thibodeau & Durgin (2011) found a high correlation between conventionality scores and frequencies of metaphoric sentences based on participants’ input and the Google search engine respectively, showing that this rating method can be considered objective. It must be noted, however, that we did not find a correlation between conventionality scores and frequency. It should also be noted that for example the metaphor FEAR IS COLD was scored 2.92 out of 7, which means participants regarded it as not conventional or unfamiliar, yet this is considered a familiar conceptual metaphor. However, a conceptual metaphor is not what is used in everyday speech, but rather what is underlying expressions such as ‘That scare gave me chills’, which explains the low score. Nevertheless, we do believe this method is reliable, especially when
depending on critical participants with high education, such as in our study. That said, future studies may want to look into and compare different methods of determining conventionality, such as questionnaires, dictionary analyses, and corpora studies.
We also tested the stimuli using MIPVU (Steen et al., 2010), by which all non-deliberate and
deliberate metaphoric stimuli were marked as metaphoric, whereas the literal stimuli were not. Next to that, DMIP (Reijnierse et al., 2017) marked all deliberate metaphoric but not the non-deliberate metaphoric stimuli as potentially deliberate. Thus, the used stimuli were correctly constructed according to these objective identification procedures, and we would recommend other
(psycholinguistic) studies to use the same procedures so as to have meaningful comparisons between them. However, these methods still warrant some discussion. One common critique of the MIPVU method is that it is a time-intensive procedure, while another is that it depends on dictionary definitions. It may be unclear how these definitions came to be, and it may sometimes be difficult to determine whether or not a word as it is used fits the first and most basic meaning of that word. This was discussed by for example Dorst & Reijnierse (2015) who argue that dictionaries are descriptive, but not prescriptive, and that researchers should ultimately make systematic decisions based on their own intuitions. It was not necessary for us to do this, as there was no conflict for the stimuli in this experiment: all literal and metaphoric uses of the source domain word were created to be clearly distinct from one another, and all metaphoric uses were defined by the Van Dale dictionary. This also showed that all metaphors used were conventional, in that the metaphorical meaning has become so attached to the source domain word that it reached the point in its ‘career’ (Bowdle & Gentner, 2005) of being defined in the dictionary. Interestingly, some of the metaphors that received low scores on the conventionality questionnaire were also not defined by the dictionary (for example A LAND IS A FORT), showing a possible correlation between subjective ratings and dictionary
definitions. Furthermore, some metaphors used were defined under the first definition as figurative (for example TIME IS MONEY), possible indicating that the metaphor is so conventional it warranted a mention under the first definition. There seemed to be no link between these definitions and conventionality scores however.
A special mention should go to the metaphor A BODY IS A TEMPLE: according to the Van Dale dictionary it is an outdated metaphor, yet it scored relatively high: 5.83 out of 7 points. However, it can be argued that in a modern Western society such as the Netherlands, which is mostly non-religious but shares the contemporary Western focus on healthy eating and positive body image, the metaphoric use of the word would in fact be quite common, and may be even more common than the literal use of the word. This could have interesting implications for theories on metaphor, for example the Career of Metaphor Theory (Bowdle & Gentner, 2005): could a metaphor reach a point in its ‘career’ where it surpasses the literal meaning of a word? Other words and metaphors could be
22
discussed similarly: “peulenschil” (peel of pods, used in this sense similar to “a piece of cake”) is used to describe something easy, and a quick look at the Google search engine results indicates that the word is used more often metaphorically than literally. While it is likely that such an expression is still processed through categorization (or comparison) as it is still metaphoric, frequency of words is known to be a factor in gaze time, and thereby processing speed (Clifton et al., 2007). Additionally, we may consider the Graded Salience model of Giora (2003), which states that the ‘salience’ of an expression determines how rapidly it is processed and understood, and that the most salient interpretation of a word or expression always comes first. This is used to explain how conventional metaphors and literal expressions take (in some studies) equal processing time. However, the results of Lai (et al., 2009) (which agree with our results) disagree with the Graded Salience model, as they found a difference in initial processing between conventional metaphors and literal statements, which were of equal salience (and should therefore be processed similarly according to the Graded Salience model). The influence of the metaphoric or literal use of words on (perceived) frequency, the interaction between frequency and categorization/comparison, and between salience (and frequency) and processing is as of yet unknown, but could and should be further explored in future studies.
Due to the set-up of this experiment, we only used the two of the six features of deliberate
metaphor, as distinguished by Tay (2017). While these features are considered equal to one another, in the sense that DMT does not distinguish between levels of deliberateness, other researchers may want to carry out (eye-tracking) experiments with stimuli based on the other four features as well, so as to form a complete understanding of the processing of (deliberate) metaphors. Furthermore, not all deliberate metaphors using any of the six features identified by Tay (2017) may be identified with DMIP (Reijnierse et al., 2017), such as the feature of repetition, either hinting at a misidentification of the features or at an incompleteness of DMIP. Further theoretical and empirical research would be highly recommended, as a uniform definition and identification procedure is necessary to perform relevant experiments that could successfully be compared.
While we took great care to select conventional metaphors for use as stimuli for this experiment, and selected participants in the same age range and with similar educational backgrounds for the
questionnaire and experiment, it is still possible that some participants in the experiment were not familiar with the presented metaphors. It would thus be a novel metaphor for them, and
consequently, a deliberate metaphor. This would be difficult to prevent; a questionnaire before the experiment would likely influence the results and a questionnaire afterwards would vice versa be influenced by the stimuli in the experiment. Regardless, future studies on the effects of
conventionality on the (perceived) deliberateness of metaphors, and the differences between deliberate conventional and novel metaphors are recommended. Finally, we encourage the reuse of the experimental set-up of this study with different metaphors and in different languages, which should be relatively easy to set up, but would provide great initial insight (to be further refined by fMRI and/or EEG) in the processing of (deliberate) metaphors within and between languages. In conclusion, we believe that the present study is relevant for the development of DMT and metaphor research in general as we have created and successfully tested a robust experimental set-up for DMT that is easy to implement and apply to other languages and metaphors (with notes on its pitfalls and possible improvements). Next to that, we have found the first physiological evidence of a cognitive demarcation between non-deliberate and deliberate metaphors, through memory recall and online processing by results with large effect sizes. Finally, these results also provide insights into
23
how this processing takes place, which additionally creates new ventures for future theoretical and empirical studies.
5. References
Ashby, J., Roncero, C., de Almeida, R. G., & Agauas, S. J. (2017). The early processing of metaphors and similes: Evidence from eye movements. The Quarterly Journal of Experimental Psychology, 1-9. Blank, G. D. (1988). Metaphors in the lexicon. Metaphor and Symbol, 3(3), 21-36.
Bowdle, B. F., & Gentner, D. (2005). The career of metaphor. Psychological review, 112(1), 193. Cameron, L. (2003). Metaphor in educational discourse. London/New York: Continuum.
Clifton, C., Staub, A., & Rayner, K. (2007). Eye movements in reading words and sentences. Eye Movements, 341-371.
Dorst, A. G., & Reijnierse, W. G. (2015). A dictionary gives definitions, not decisions: On using a dictionary to identify the basic senses of words. Metaphor and the Social World, 5(1), 137-144. Dulcinati, G., Mazzarella, D., Pouscoulous, N., & Rodd, J. (2014). Processing metaphor: The role of conventionality, familiarity and dominance. UCL Working Papers in Linguistics, 72-88.
Frisson, S., & Pickering, M. J. (1999). The processing of metonymy: Evidence from eye movements. Journal of Experimental Psychology: Learning, Memory, and Cognition, 25(6), 1366.
Gentner, D., & Wolff, P. (1997). Alignment in the processing of metaphor. Journal of memory and language, 37(3), 331-355.
Gibbs, R. W. (2015a). Does deliberate metaphor theory have a future? Journal of Pragmatics, 90, 73-76.
Gibbs Jr, R. W. (2015b). Do pragmatic signals affect conventional metaphor understanding? A failed test of deliberate metaphor theory. Journal of Pragmatics, 90, 77-87.
Gibbs, R. W., & Chen, E. (2017). Taking metaphor studies back to the Stone Age: A reply to Xu, Zhang, and Wu (2016). Intercultural Pragmatics, 14(1), 117-124.
Giora, R. (2003). On our mind: Salience, context, and figurative language. Oxford University Press. Gong, S. P., & Ahrens, K. (2007). Processing conceptual metaphors in on-going discourse. Metaphor and Symbol, 22(4), 313-330.
Janus, R. A., & Bever, T. G. (1985). Processing of metaphoric language: An investigation of the three-stage model of metaphor comprehension. Journal of Psycholinguistic research, 14(5), 473-487. Krennmayr, T., Bowdle, B. F., Mulder, G., & Steen, G. J. (2014). Economic competition is like auto racing: Building metaphorical schemas when reading text. Metaphor and the Social World, 4(1), 65-89.
Lai, V. T., Curran, T., & Menn, L. (2009). Comprehending conventional and novel metaphors: An ERP study. Brain Research, 1284, 145-155.
24
Lakoff, G., & Johnson, M. (2008). Metaphors we live by. University of Chicago press.
McNamara, D. S. (2001). Reading both high-coherence and low-coherence texts: Effects of text sequence and prior knowledge. Canadian Journal of Experimental Psychology/Revue canadienne de psychologie expérimentale, 55(1), 51.
Musolff, A. (2016). Cross-cultural variation in deliberate metaphor interpretation. Metaphor and the Social World, 6(2), 205-224.
Ng, C. J. W., & Koller, V. (2013). Deliberate conventional metaphor in images: The case of corporate branding discourse. Metaphor and Symbol, 28(3), 131-147.
Olkoniemi, H., Ranta, H., & Kaakinen, J. K. (2016). Individual differences in the processing of written sarcasm and metaphor: Evidence from eye movements. Journal of Experimental Psychology: Learning, Memory, and Cognition, 42(3), 433.
Ortony, A. (1975). Why metaphors are necessary and not just nice. Educational theory, 25(1), 45-53. Perrez, J., & Reuchamps, M. (2014). Deliberate metaphors in political discourse: the case of citizen discourse. Metaphorik, 7-41.
Prat, C. S., Mason, R. A., & Just, M. A. (2012). An fMRI investigation of analogical mapping in
metaphor comprehension: The influence of context and individual cognitive capacities on processing demands. Journal of Experimental Psychology: Learning, Memory, and Cognition, 38(2), 282.
Raney, G. E., Campbell, S. J., & Bovee, J. C. (2014). Using eye movements to evaluate the cognitive processes involved in text comprehension. Journal of visualized experiments: JoVE, (83).
Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological bulletin, 124(3), 372.
Rayner, K., Slattery, T. J., Drieghe, D., & Liversedge, S. P. (2011). Eye movements and word skipping during reading: effects of word length and predictability. Journal of Experimental Psychology: Human Perception and Performance, 37(2), 514.
Reijnierse, W. G., Burgers, C., Krennmayr, T., & Steen, G. J. (2015). How viruses and beasts affect our opinions (or not): The role of extendedness in metaphorical framing. Metaphor and the Social World, 5(2), 245-263.
Reijnierse, W. G., Burgers, C., Krennmayr, T., & Steen, G. J. (2017). DMIP: A Method for Identifying Potentially Deliberate Metaphor in Language Use. Corpus Pragmatics, 1-19.
Rubio-Fernández, P., Cummins, C., & Tian, Y. (2016). Are single and extended metaphors processed differently? A test of two Relevance-Theoretic accounts. Journal of Pragmatics, 94, 15-28.
Schotter, E. R., Angele, B., & Rayner, K. (2012). Parafoveal processing in reading. Attention, Perception, & Psychophysics, 74(1), 5-35.
Shibata, M., Toyomura, A., Motoyama, H., Itoh, H., Kawabata, Y., & Abe, J. I. (2012). Does simile comprehension differ from metaphor comprehension? A functional MRI study. Brain and language, 121(3), 254-260.
