Floating or Suggesting Ideas? A Large-Scale Contrastive Analysis of Metaphorical and Literal Verb-Object Constructions
Authors: Prisca Piccirilli, Alexander Fraser, Sabine Schulte im Walde
Organizations: Institute for Natural Language Processing (IMS), University of Stuttgart, Germany · Technical University of Munich (TUM), Germany · Munich Center for Machine Learning (MCML), Germany
Metaphor pervades everyday language, allowing speakers to express abstract concepts via concrete domains. While prior work has studied metaphors cognitively and psycholinguistically, large-scale comparisons with literal language remain limited, especially for near-synonymous expressions. We analyze 297 English verb-object pairs (e.g., float idea vs. suggest idea) in ~2M corpus sentences, examining their contextual usage. Using five NLP tools, we extract 2,293 cognitive and linguistic features capturing affective, lexical, syntactic, and discourse-level properties. We address: (i) whether features differ between metaphorical and literal contexts (cross-pair analysis), and (ii) whether individual VO pairs diverge internally (within-pair analysis). Cross-pair results show literal contexts have higher lexical frequency, cohesion, and structural regularity, while metaphorical contexts show greater affective load, imageability, lexical diversity, and constructional specificity. Within-pair analyses reveal substantial heterogeneity, with most pairs showing non-uniform effects. These results suggest no single, consistent distributional pattern that distinguishes metaphorical from literal usage. Instead, differences are largely construction-specific. Overall, large-scale data combined with diverse features provides a fine-grained understanding of metaphor-literal contrasts in VO usage.
Figures & tables
Figure 1: Pipeline of our approach: based on an existing set of metaphorival vs. literal VO pairs, we extract sentences containing these VOs. We then run five NLP tools on all sentences (masking the VO, see Section 3 ) to extract more than 2,200 cognitive and linguistic features in order to analyze the contrast between metaphorical and literal language usage from (i) a cross-pair and (ii) a within-pair perspective.
Figure 2: Metaphorical vs. literal VO preferences against total frequencies on a logarithmic scale, with points color-coded by dominance (green for literal-dominant, orange for metaphorical-dominant). The ratio ranges from 0 (entirely literal usage) to 1 (entirely metaphorical usage). Labels indicate the dominant VO followed by its corresponding counterpart in parenthesis, For example, met. VO breathe life (lit. VO instill life ) is metaphorical dominant with a ratio of 1.0. For better readability, we report in Table 2 the 10 most (i) frequency-balanced, (ii) literal-dominant and (iii) metaphor-dominant VO pairs.
Figure 3: Metaphorical vs. literal verb-only preferences (independently of their direct objects) against total frequencies on a logarithmic scale, with points color-coded by dominance (green for literal-dominant, orange for metaphorical-dominant). The ratio ranges from 0 (entirely literal usage) to 1 (entirely metaphorical usage). Labels indicate the dominant verb followed by its corresponding counterpart in parenthesis.
Category
Count
Total number of sentences
2,058,787
Sentences with literal VO s
1,724,892
Sentences with metaphorical VO s
333,895
Total number of VO pairs
297
VO pairs literal > metaphorical
222
VO pairs metaphorical > literal
75
Table 1: Distribution of literal and metaphorical sentences across VO pairs in the dataset. The table reports the total number of sentences per category and the number of VO pairs for which one usage type predominates.
Metaphorical
Literal
Met Count
Lit Count
Ratio
Log 2
Top 10 frequency-balanced VO Pairs
absorb concept
assimilate concept
89
65
0.58
0.45
absorb experience
assimilate experience
109
91
0.55
0.26
frame idea
pose idea
147
120
0.55
0.29
mount production
organise production
322
305
0.51
0.07
mount production
organize production
322
316
0.51
0.03
Table 2: Top VO pairs across distributional profiles: (i) the 10 most frequency-balanced pairs, (ii) the 10 pairs where metaphorical usage is more frequent than their literal counterparts, and (iii) the 10 pairs where literal usage is more frequent. Met Count and Lit Count indicate the number of sentences in which the metaphorical and literal VO occur, respectively. Ratio corresponds to the proportion of metaphorical usage (cf. Figure 2 ). Log 2 represents the base-2 logarithm of the ratio between metaphorical and literal frequencies, capturing the direction and magnitude of the imbalance.
Figure 4: Tool-level aggregation of distributional dominance for 93 clearly-distinctive VO pairs (mean absolute difference ≥1 standard deviations in at least three tools, detailed methodology explained in Sections 3 and 4.3 ). Feature-level contrasts were collapsed into tool-level directional judgments (literal-dominant in green vs. metaphor-dominant in orange) for VO pairs meeting the distinctiveness criterion in at least three tools. VO pairs are sorted top to bottom from mostly literal-dominant to mostly metaphor-dominant. For example, the VO pair instill idea vs. breathe idea behaves significantly more "metaphorical-like" across features of four tools (SEANCE, TAACO, TAALES and TAASSC), i.e., the metaphorical variant breathe idea introduces robust distributional differences detectable across linguistic features (structurally, lexically, discourse-wise, etc.) Sections 3 and 4.2 describe the linguistic aspects captures by the tools.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Metaphorical VO
Literal VO
Met Count
Lit Count
absorb knowledge
assimilate knowledge
354
190
absorb information
assimilate information
1,154
511
absorb idea
assimilate idea
369
141
absorb culture
assimilate culture
308
127
absorb material
assimilate material
382
85
absorb lesson
assimilate lesson
366
50
Appendix
Table 3: Metaphorical vs. Literal Verb–Object Pairs with Corpus Frequencies (N=297)
Metonymy and metaphor often co-occur in natural language, yet computational work has studied them largely in isolation. We introduce a framework that transforms a literal sentence into three figurative variants: metonymic, metaphoric, and hybrid. Using this framework, we construct MetFuse, the first dedicated dataset of figurative fusion between metonymy and metaphor, containing 1,000 human-verified meaning-aligned quadruplets totaling 4,000 sentences. Extrinsic experiments on eight existing benchmarks show that augmenting training data with MetFuse consistently improves both metonymy and metaphor classification, with hybrid examples yielding the largest gains on metonymy tasks. Using this dataset, we also analyze how the presence of one figurative type impacts another. Our findings show that both human annotators and large language models better identify metonymy in hybrid sentences than in metonymy-only sentences, demonstrating that the presence of a metaphor makes a metonymic noun more explicit. Our dataset is publicly available at: https://github.com/cincynlp/MetFuse.
Conceptual metaphors guide our thinking and actions by allowing us to reason about more abstract experiences (e.g., paying taxes) in terms of more concrete or embodied experiences (e.g., carrying a physical load) (Lakoff and Johnson, 2011). It follows that different conceptual metaphors can result in different reasoning: framing paying taxes as an investment in a community rather than a physical load leads to a very different outlook on taxation. Identifying the conceptual metaphors guiding a speaker or writer thus helps to reveal their framing of events. Though these metaphors can't be observed directly, groups of linguistic metaphors, metaphorical expressions as they appear in language, serve as evidence for them. Motivated by this, we present an unsupervised method that extracts linguistic metaphors from a corpus and uses a structured clustering approach to form groups corresponding to conceptual metaphors. Using this method, we point to key topical and framing differences in left- vs. right-leaning podcasts. For example, left-leaning podcasts tend to conceptualize media stories as a weapon, while right-leaning sources commonly discuss the economy as a system subject to vertical changes.
Static concreteness ratings are widely used in NLP, yet a word's concreteness can shift with context, especially in figurative language such as metaphor, where common concrete nouns can take abstract interpretations. While such shifts are evident from context, it remains unclear how LLMs understand concreteness internally. We conduct a layer-wise and geometric analysis of LLM hidden representations across four model families, examining how models distinguish literal vs figurative uses of the same noun and how concreteness is organized in representation space. We find that LLMs separate literal and figurative usage in early layers, and that mid-to-late layers compress concreteness into a one-dimensional direction that is consistent across models. Finally, we show that this geometric structure is practically useful: a single concreteness direction supports efficient figurative-language classification and enables training-free steering of generation toward more literal or more figurative rewrites.