Edwin den Boer, IGEL Utrecht 1998 (handout/paper)

The Frequency of Original Metaphors in Literary and Nonliterary Texts

1. Texts

Metaphors were tallied in a corpus of 30 short texts (10585 words in total, average text length: 353 words), representing four main genres:

      Poetry (P):     10 (lyrical) poems
      Fiction (F):     8 fragments of prose fiction (short stories)
      News (N):        5 frontpage newspaper reports
      Science (S):     7 fragments of scientific (psychological) articles

These texts were taken at random from recent (1994, early 1995) periodical publications in English (American, British and a few from other varieties), each from a different title with the exception of three fiction fragments. Among the fiction fragments, final passages were over-represented. These contained more metaphors than other fragments. Yet the results for fiction were still a bit lower than expected. Comparisons between male and female authors, and between Britons and Americans, corrected for genre effects, did not show appreciable differences in metaphoricity. In contrast, texts with abstract themes (philosophy, finance, mathematics) were found to be about twice as metaphorical as other texts.

2. Results

1a. Metaphor types per 10,000 words
Genre Moderately conventional Moderately original Very original metaphors
Poetry 248 192 48
Fiction 200 72 29
News 256 84 0
Science 225 32 0
1b. Metaphor tokens per 10,000 words
Genre Moderately conventional Moderately original Very original metaphors
Poetry 312 233 71
Fiction 225 80 58
News 358 84 0
Science 468 35 0

Table 1 shows the mean scores of the relative frequency of metaphors for these genres, separately for types and tokens. See also Section 3 for a description of the levels of originality. Table 2 below shows those differences between genres that are statistically significant. None were found for conventional metaphors. Moderately original metaphors were clearly more frequent in poetry than in other genres, while very original metaphors were only found in literary texts.

Table 2. Significant differences between genres: Results of Mann-Whitney-Wilcoxon tests [N > S = more metaphors in news than in science, p < .05] [P >> F = more metaphors in poetry than in fiction, p < .01] MODERATELY ORIGINAL METAPHORS (e.g., The known, blazing like a headlamp): Types: P >> F P >> S F > S N > S Tokens: P >> F P > N P >> S F > S N > S VERY ORIGINAL METAPHORS (e.g., Life bleeds language): Types: P > N P > S F > N F > S Tokens: P > N P > S F > N F > S TOTAL NUMBER OF METAPHORS: Types: P > F P >> S Tokens: P > F

2. Results (continued)

Very conventional metaphors were not counted, because they are very frequent and sometimes difficult to distinguish from literal expressions, and because they will have no noticeable effect on the reader, although they might influence us indirectly or on a subliminal level. Moderately conventional metaphors were found in all the texts; more than twice as many tokens were found in scientific articles as in fiction, but this difference was not statistically significant, due to great variance. Moderately original metaphors were found in almost all texts, with the exception of three science fragments. Very original metaphors were found in half of the poems and half of the fiction fragments.

The overall effect for each category of metaphors was also computed, by running a Spearman rank-order correlation with the polynomial variable Literariness, i.e., the four genres ranked P, F, N, S in descending order. As shown on the poster, this correlation was very significant (p < .01) for all measurements of original metaphors, and for the total amount of metaphor types. Furthermore, there was also a significant correlation between Literariness and two other variables, which are a bit speculative but theoretically interesting: a weighed total, in which the number of moderately original metaphors was doubled and the number of very original metaphors was multiplied by four, and a ratio of originality, computed by dividing the weighed total by the unweighed total.

3. Method of analysis

The texts were analyzed by myself. However, the metaphors were not just classified intuitively, but according to specific criteria. These were not measured by separate scores, but they were taken into account when judging the originality of metaphors and they were often indicated explicitly in the analysis as reported in Den Boer (1998). The criteria were imagery value, novelty, and complexity. The rationale for combining these criteria was that they were all considered aspects of a single underlying variable, namely the relative concreteness of the source domain of the metaphor, compared with the target domain. Anyway, they all tend to promote a concrete, fictional interpretation, as opposed to the immediate construction or retrieval of an abstract meaning.

Generally speaking - but this is a simplification - the four levels of originality were contrasted as follows. Moderately conventional metaphors have more imagery value than very conventional metaphors. If the latter contain any imagery at all, it is only schematic, as in pay rise, full details, a probability distribution, right away or heavy fighting. The former are more specific and more graphic, as in: the brittle cheerfulness; his scholarship's been cut; I tagged along with my dad; a small circle around rebel leader Dzokhar Dudayev. However, they are not as novel as moderately original metaphors, such as The rainy trek from clothes-line to airing cupboard; allowing the markets to burn themselves out; or the known, blazing like a headlamp. These in turn are less complex than very original metaphors, like the following: Life bleeds language; my behindscenes beetle-on-its-back-grief waves its arms; my pleasures have wandered off in time.

It was hard to find a ground for determining whether different metaphorical utterances belonged to the same type of metaphor. Verbal repetition would not be a good criterion. A single word can have different metaphorical meanings, depending on the context, even if it only has a single literal meaning. Conversely, a single metaphor can be expressed in different words by means of synonymy, antonymy, derivation and composition. On the other side of the scale, Lakovian root metaphors could not be used either. They can underlie both conventional and original metaphors, which were to distinguished. Besides, it would often be difficult or impossible to assign very original metaphors to a certain root metaphor.

Therefore, a more limited conceptual criterion was chosen. Utterances were considered to belong to the same type if they contained basically the same source and target concepts, and produce the same associations and images. This criterion could be made more precise if we would specify the level of generality of these concepts, e.g. the 'basic', i.e. intermediate, level of Lakoff (1987), or if we would describe all metaphors by a proposition according to certain guidelines, and require tokens of the same type to be described by an identical proposition. In my study, the broadest application of this criterion was a case in a scientific text, where the statistical terms truncation and cutoff points were counted as tokens of the same type. As very original metaphors were generally very complex, too complex to be fully expressed in a single sentence, partial expressions of them were also counted as tokens of the same type.

The clearly distinct results for the different levels of originality suggest that they may correspond to empirical categories that are processed differently in metaphor understanding. I hope that this method will be developed further in experimental research, or a more strictly controlled form of text analysis. There are reasons for doubting the reliability of the method used, but in general, rationally argued text analysis cannot be dismissed as unempirical.

4. Implications for metaphor theory

The method of analysis was loosely based on Lakoff and Johnson (1980) and Lakoff and Turner (1989). The results clearly corroborate one of their less controversial assertations, namely that literature, and poetry in particular, is marked by a great amount, not of metaphors per se, but of unconventional metaphors. This idea was also confirmed by Steen (1994): in two of his experiments, experts rated metaphorical extracts from fiction and from newspapers on a number of scales; factor analyses indicated that the literary extracts were significantly less conventional in the first experiment and more difficult in the second, compared with the journalistic extracts.

But the results are in contrast with Katz et. al. (1988), who found no significant difference between literary (i.e., poetic) and 'nonliterary' metaphors, neither on Degree of Metaphoricity nor on any of the ten related dimensions for which they collected scale ratings. However, their so-called 'nonliterary' metaphors were not taken from nonliterary texts, but invented by Katz, Marschark and other metaphorologists. One could argue that these metaphors must be literary by definition, being created only in order to produce an aesthetic effect, with little regard for communication. Moreover, both the literary and the artificial metaphors were (re)phrased in the nominative form (A is B, e.g., History is a magnet or Truth is a firefly, both supposed to be nonliterary) and presented without context, in order to avoid style effects. But this is more than a matter of style. Nominative metaphors tend to express original metaphors and tend to occur in literature. Normally, it would be most redundant to mention the target term explicitly and also relate it explicitly to the source term, but in these cases it is often necessary, because the semantic relation is obscure, or because the source term is formulated rather concisely. Although nominative metaphors seem omnipresent in metaphor studies, they appear to be very rare in real life. In my corpus of over 10,000 words I only found two cases: Waste is my enemy, the foe (in a poem) and 'Love is a many splendoured thing' (a song title quoted in a short story). So we may conclude that the material used by Katz et. al. was not nearly representative for nonliterary metaphors and too uniform to allow substantial differences in degree of originality.


Den Boer, Edwin (1998). Concrete metaforen: Conceptualisme, concreetheid en de metaforiciteit van literaire teksten. Master thesis in Literary Studies, Utrecht University, in Dutch. [Concrete metaphors: conceptualism, concreteness, and the metaphoricity of literary texts.]

Katz, Albert N., A. Paivio, M. Marschark & J.M. Clark (1988). "Norms for 204 literary and 260 nonliterary metaphors on 10 psychological dimensions". Metaphor and Symbolic Activity 3: 4, 191-214.

Lakoff, George (1987). Women, fire and dangerous things: What categories reveal about the mind. Chicago: University of Chicago Press.

Lakoff, George & Mark Johnson (1980). Metaphors we live by. Chicago: University of Chicago Press.

Lakoff, George & Mark Turner (1989). More than cool reason: A field guide to poetic metaphor. Chicago: University of Chicago Press.

Steen, Gerard (1994). Understanding metaphor in literature: An empirical approach. London: Longman.


Written and presented by: Edwin den Boer, currently at NIPO, the Netherlands Institute for Public Opinion and Market Research in Amsterdam, graduate of Utrecht University. Research supervised by Will van Peer, now at Ludwig-Maximillians-Universität in Munich.

All data have been taken from Den Boer (1998). Examples are quoted from the text corpus analyzed in that study, except two artificial metaphors from Katz et. al. (1988).

This paper is based on a handout that was distributed at Poster session 1: Reading processes / Conditions of production and reception (August 28) of the VIth Biannual Conference of the Internationale Gesellschaft für Empirische Literaturwissenschaft, August 26-29, 1998, at Utrecht University.