Speedy Metonymy, Tricky Metaphor, Irrelevant Compositionality: How Nonliteralness Affects Idioms in Reading and Rating

Abstract

It is widely acknowledged that fixed expressions such as idioms have a processing advantage over non-idiomatic language. While many idioms are metaphoric, metonymic, or even literal, the effect of varying nonliteralness in their processing has not been much researched yet. Theoretical and empirical findings suggest that metonymies are easier to process than metaphors but it is unclear whether this applies to idioms. Two self-paced reading experiments test whether metonymic, metaphoric, or literal idioms have a greater processing advantage over non-idiomatic control sentences, and whether this is caused by varying nonliteralness. Both studies find that metonymic and literal idioms are read significantly faster than controls, while the advantage for metaphoric idioms is only tenuous. Only experiment 2 finds literal idioms to be read fastest of all. As compositionality of the idioms cannot account for these findings, some effect of nonliteralness is suggested, together with idiomaticity and the sentential context.

Appendices

Appendix 1

Theoretical Alternative for Building Matching Control Sentences

In building control sentences, constraints well-known to idiom experimenters arise: idioms require a fixedness of word choice such that synonyms or semantic neighbors can rarely replace words without rendering the idiom unrecognizable, marked, or changing the meaning. Some idioms require certain contexts to be understandable or unambiguous, some contain words rather rare in absolute terms yet very common within the idiomatic use, while the idioms themselves are actually very common or well-known themselves. However, exact overall match of length and frequency was done wherever possible. While it is most common to match control items as closely as possible to the test items by using as many of the same words in the same order as possible, this kind of “formal matching” causes an unavoidable difference in the meanings of sentences, which cannot be quantified. Size and effect of divergence or even fundamental difference in meanings of items remain unobservable and unmeasurable in participants’ minds. The problem of this “black box of meanings” could be dissolved by matching sentences for meaning. This would be especially profitable in the case of idioms as they can be translated into a fairly exact non-idiomatic meaning. Consequently, it could be argued that for idioms, it makes sense to form control sentences that are matched by meaning only. For example, the item For two hours, Marie’s stomach has been growling would be matched by For two hours, Marie has been very hungry. The advantage of matching by meaning is that it enables a direct comparison between the processing difficulty of an idiom and its meaning which in theory provides a more precise and pointed answer as to whether a fixed expression is indeed easier to process than its non-idiomatic sentence expressing the identical meaning. This matching, however, poses many challenges and leads to serious unavoidable differences in construction and word forms between sentence pairs: syntax, word categories, word number, and other grammatical differences can diverge to a degree where a control sentence cannot actually serve as true control anymore. Length and frequency matching become even more difficult as words from the idiom can hardly be repeated in its control. At the same time, the choice of potential words is automatically very constrained when the control sentence is meant to express a very particular meaning. Occasionally, an idiom’s meaning may be too ambiguous or complex to be fully captured in a control sentence with a fairly fixed required number of words. In sum, practical emerging difficulties and resulting differences between idiom and control sentences make matching by meaning very problematic and offer too many potential confounds, despite the undeniable advantage of a matched and controlled meaning and the theoretical strength of a very pointed answer to the question whether idioms are cognitively and semantically easier to process than non-idiomatic sentences. We conclude that “formal matching” is the sounder choice because it controls for more possible confounds, but matching by meaning could be valuable for a complementary study serving as a direct comparison for the current study.

Appendix 2

Prestudy on Ratings of Nonliteralness in Literal Idioms

Nonliteralness ratings on literal, metonymic, and metaphoric idioms were collected in two separate studies for two reasons: one, to test whether results from one study could be repeated for literal idioms, thus to decrease the chance of chance findings; second, it was attempted to make the studies as simple and outcomes as clear as possible by only demarcating two different types of idioms each time.

To check for significance of the effect of idiom type and to account for random individual differences of items and participants, an ordinal mixed effects regression was performed. Literalness ratings were fitted as a function of idiom type as a categorical fixed effect, random intercepts for items and random slopes for type by participant. Both studies revealed idiom type to have a significant effect. In study 1 containing the metaphoric idioms, the effect was stronger (b = − 2.11, z = − 14.43, p < 0.001 as opposed to b = − 0.53, z = − 4.56, p < 0.001 in study 2), which is expected, given that metaphoric idioms were rated as much more nonliteral than metonymic idioms. This indicates that literal idioms are indeed perceived as substantially more literal than both metonymic and metaphoric idioms. Furthermore, the difference in nonliteralness between these two groups confirm the results of the initial rating study (Michl 2019) on the nonliteralness of metonymic compared to metaphoric idioms and show that the effect can also be found on a less detailed rating scale.

Appendix 3

Effect of Transparency Ratings on Reading Times

Given that transparency can influence processing ease depending on the task, metonymic and metaphoric idioms were also rated on it by adult German native speakers (see, Michl 2019). Transparency is defined as the closeness of relation between what is said and what is meant (or the literal and the idiomatic meaning) in an idiom (see also Nippold and Taylor 2002; Titone and Connine 1999). 111 participants rated it on a 5-point Likert scale on which 5 indicated “completely transparent”. Metonymic idioms received a mean rating of 3.8 (sd = 0.18), metaphoric idioms received a mean rating of 2.8 (sd = 0.13). To control for potential transparency effects in the present reading experiments, median ratings for each idiom were once included in the final models. Results remained largely the same. Transparency itself was completely irrelevant in a non-biasing context (Table 7a, experiment 1), yet turned out to be a significant predictor when idioms were presented in a biasing context (Table 7b, experiment 2). One caveat needs to be borne in mind: literal idioms were not rated on transparency by multiple participants. Instead, literal idioms received ratings of “completely transparent” by the author. This was decided because literal idioms have only one meaning, so their literal and the idiomatic meaning are the same, so they should be completely transparent by definition. Due to this difference in data collection and for the sake of simplicity, transparency ratings are not considered in the final data analysis.

Table 7 Log-transformed reading times from self-paced reading (a) experiment 1, (b) experiment 2, lmer coefficients and standard error, t-value, confidence intervals, and p-value