Temporal word embeddings have been successfully employed in semantic change research to identify and trace shifts in the meaning of words. In a previous work, we developed an approach to study the diachrony of complex expressions, namely literary metaphors. Capitalizing on the evidence that measures of semantic similarity between the two terms of a metaphor approximate human judgments of the dificulty of the expression, we used time-locked measures of similarity to reconstruct the evolution of processing costs of literary metaphors over the past two centuries. In this work, we extend this approach previously used on Italian literary metaphors and we present a proof-of-concept study testing its crosslinguistic applicability on a set of 19th-century English literary metaphors. Our results show that the processing costs of metaphors changed as a function of textual genre but not of epoch: cosine similarity between the two terms of literary metaphors is higher in literary compared to nonliterary texts, and this diference is stable across epochs. Furthermore, we show that, depending on the metaphor structure, the diference between genres is afected by word-level variables, such as the frequency of the metaphor's vehicle and the stability of the meaning of both topic and vehicle. In a broader perspective, general considerations can be drawn about the history of literary and nonliterary English language and the semantic change of words.
terms of linguistic structure, metaphors normally involve its dificulty and we analyze how it varies across time two terms, the topic and the vehicle: for example, in and textual genres. We also consider the role of word frethe metaphor ‘Sally is a chameleon’, the topic Sally is quency (WF) and vector coherence (VC), two widely used described by the broadened vehicle chameleon, to indi- measures in the study of semantic change [29, 30], as well cate a person who changes attitude/behavior to fit their as semantic neighborhood density (SND) in shaping the surroundings. While metaphors are broadly used in ev- dificulty of the expression. WF and VC were considered eryday communication, they are certainly a distinctive to assess the efect of the semantic change of the single feature of literary texts, as long evidenced in stylistics word on the evolution of whole metaphor understand[15]. Past studies on literary metaphors, however, report ing, while SND was considered to analyze the impact of mixed results. The rating study by Katz et al. [16] found a measure known to synchronically impacts metaphor no diference between literary and everyday metaphors, understanding [31, 24] on its diachronic unfolding. while other studies showed that the former type is less familiar and more open-ended than the latter [17], but literary metaphors are rated as less dificult and more 2. Methods familiar when presented together with their original context [18]. Moreover, the processing of literay metaphors 2.1. Dataset of metaphors seems to be particularly efortful, given the multitude The study focuses on “classic” literary metaphors (i.e., of possible meanings they evoke [19]. Therefore, open metaphors found in 19th-century literary texts). In terms questions remain regarding how literary metaphors are of metaphor structure, we focused on metaphors in the processed. It must be also underlined that the literary form of ‘A is B’ (e.g. “Stars are dancers”) and ‘A of B’ metaphors used in previous studies were written tens or (e.g., “Clouds of melancholy”), as they clearly display hundreds of years ago. Yet, the efect of this diachronic di- the two metaphorical elements (topic and vehicle) and mension on their processing costs, as well as its interplay allow to avoid possible confounding factors (length of with textual genre in which metaphors are embedded, expression, intervening words, etc.). Twenty- four (24) remains an open question. ‘A is B’ metaphors were taken from the dataset in Katz et al. [16] and 115 metaphors in the form ‘A of B’ were retrieved from a collection of literary texts of the 19th century. These latter were identified by PoS-tagging a corpus of literary texts from the 19th century (see below) with spaCy [32], and then extracting only the ‘NOUN of NOUN’ constructions. The resulting list was then further reduced by manually searching for words belonging to known sources of metaphors, such as atmospheric events (e.g., ‘rain’) or physical locations (e.g., ‘river’) [33], following the methodology in Bambini et al. (2014) [18].
In addition to its diachronic application, the use of vector space models can help characterize metaphors thanks to the ability of these models to approximate human performance in psycholinguistic tasks. Measures derived from vector space models were shown to be able to approximate how humans process word meaning [20, 21, 22] and, more specifically to correlate with how humans perceive metaphorical expressions in terms of metaphoricity, dificulty, and other psycholinguistic dimensions [23, 24, 25]. In particular, semantic similarity, operationalized in vector space models as cosine similarity (CS) between topic and vehicle, has long been considered relevant for metaphor studies [26] and, more recently, for automatic metaphor identification [27].
In a previous study on Italian [28], we developed a
novel method, employing the Temporal Word
Embeddings with a Compass (TWEC) model [
To train aligned temporal vector space models, we fol- extracted from the temporal vector space models (CS,
lowed the procedure by Di Carlo et al. [
The context matrix of the compass is then maintained
ifxed to train on each corpus a time- and genre-specific The linear mixed model considers CS as dependent
target matrix from which we derive the temporal word variable and the interaction between epoch and genre
embeddings. The four sets of embeddings obtained for and word-level variables as predictors. In all models
the four corpora will represent the meaning of words Items (metaphors) were added as random variables. The
in each time slice for the two genres. To validate our resulting formula was:
models, following previous studies [
the vector space models, we correlated the human scores
of relatedness and the semantic similarity derived from
2.3. Measures of interest and analyses our word embedding for each pair of words in the MEN
dataset [37] (Table 1). These results show strong
correFor each metaphor, we collected four measures of interest, lations, comparable to the results obtained by Hamilton
at the metaphor- and word-level. et al. (2016) [
CS is here considered as a proxy value of dificulty of the metaphors.
19th Literary 19th Nonliterary 21st Literary 21st Nonliterary
.55 .58 .61 .59
of dificulty from the dataset by Katz et al. [ 16]. Results showed a moderate correlation (r(26) = .49, p < .05): metaphors with higher semantic similarity between topic and vehicle were rated with lower values of dificulty by participants, coherently with previous studies.
mantic similarity between the topics and the vehicles of literary metaphors is driven by the interaction between the Epoch, Genre and single-word variables. The results of our predictors of interest are reported below.
Concerning the ‘A of B’ metaphors’ mixed model, results showed a main efect of genre ( = 0.81, t = 2.44, p = .01) and a significant three-way interaction between epoch, genre and vector coherence, both of the topic ( = 0.34, t = 2.018, p = .04) and of the vehicle ( = -1.715, t = -4.954, p < .001). These results indicate that the cosine similarity of literary metaphors’ terms did not change over time, but it changed as a function of textual genres, resulting in greater dificulty (lower cosine similarity) in nonliterary texts than in literary (Figure 1). As shown by the three-way interaction between Epoch and Genre and the single-word variables in Figure 2, the efect of VC acted diferently in the two time points and in the two genres. VC of the vehicle did not afect CS in literary and non- literary texts in the past; conversely, more stable vehicles significantly lowered CS in present literary texts and in- creased CS in present nonliterary texts. A similar trend can be observed for VC of the topic, where its stability did not afect CS in the past, regardless of the literary genres. Conversely, stability of the topic contributed to significantly increase CS in present literary texts, but less so in nonliterary texts.
For ‘A is B’, the model revealed a significant three-way interaction between epoch, genre, and the frequency of the vehicle ( = 0.06, t = 2.077, p = .04), but no main efects. The efect of WF of the vehicle showed diferent patterns in the two time points and in the two genres (Figure 3): while WF of the vehicle did not afect CS in literary texts both in the past and in the present, more frequent vehicles significantly increased CS in past nonliterary texts and lowered CS in present nonliterary texts.
poral dynamics of a set of English literary metaphors to
understand whether their processing costs changed over
time. We also explored if this change was afected by the
genre of the texts, as well as by the semantic properties
of the constituting elements of the metaphors (topic and
vehicle). By leveraging on the diachronic applications of
distributional semantics and extending a method already
applied to the study of Italian literary metaphors [28],
we created a series of time-locked semantic
representations of 139 English metaphors, from which we derived
a measure of the cosine similarity between their terms
(CS), taken as a proxy of their dificulty, together with
semantic neighborhood density (SND), stability over time
(VC), and, from four diachronic corpora, frequency (WF)
of their topics and vehicles.
when found in literary contexts, compared to when
encountered in nonliterary texts. Hence, the dificulty of
these metaphors is sensitive to the style of the text in
which metaphors are found: when read in a text that has
a literary style and aesthetic intent, the metaphor is less
striking than the same metaphor in a nonliterary text.
Moreover, we found a strong efect of the stability of the
meaning of the vehicle in interaction with epoch and
genre. This suggests that ‘A of B’ metaphors with more
unstable vehicles are perceived as less dificult than ‘A of
B’ metaphors with vehicles whose meanings remained
stable over time. We interpreted this result in light of
Traugott’s [
Results showed no efect of epoch for either ‘A is highlighted an efect of the frequency of the vehicle in
B’ or ‘A of B’ literary metaphors. Thus, no noticeable interaction with epoch and genre. In nonliterary texts,
change in CS over time was revealed, suggesting that the perceived dificulty of ‘A is B’ metaphors difered as
these metaphors come with similar processing costs for a function of the WF of their vehicle, to the point that
contemporary readers and for readers of the epoch in metaphors showed opposite patterns in the past and in
which the metaphors were created. The absence of an the present: in the past, the less frequent the vehicle,
efect of epoch can be better understood by consider- the more metaphorical the whole metaphorical
expresing the historical evolution of the English language, and sion; in the present, the less frequent the vehicle, the less
specifically its early standardization. As stated by Wyld metaphorical the metaphor. The pattern found in the
[
While literary metaphors are not processed diferently as a particularly low frequency word in the 19th century, based on the epoch, the influence of textual genre is was indeed communicating something new about the noticeable. This factor emerged both as a main efect topic wind. As such, the metaphors might have been perand in diferent interaction patterns with single-word ceived as more dificult and “more metaphorical”, leading variables, varying according to the type of metaphor. to the creation of a new concept. The very same metaphor is nowadays perceived diferently, because the frequency
For ‘A of B’ metaphors, results revealed that the dif- of the vehicle has changed: wrestler has become more ifculty of these metaphors changed as a function of the frequent, and the whole expression has lost some of its genre. In particular, they are perceived as less dificult metaphoricity for the 21st-century readers.
Overall, our results suggest that for the English language, metaphor processing costs are not afected by the temporal distance between the creation of metaphors, This work received support from the European Rewhich occurred in the 19th century, and their processing search Council under the EU’s Horizon Europe program, by today’s readers. Instead, the key factor modulating ERC Consolidator Grant “PROcessing MEtaphors: Neumetaphor processing costs seems to be the textual genre rochronometry, Acquisition and Decay, PROMENADE” in which they appear. This modulation, however, occur [101045733]. The content of this article is the sole responto a diferent extent depending on the syntactic structure sibility of the authors. The European Commission or its of the metaphors and in interaction with single word mea- services cannot be held responsible for any use that may sures. Indeed, we observe that in defining what drives be made of the information it contains. the dificulty of metaphors, diferent patterns emerged for the ‘A of B’ and ‘A is B’ structures. While for the for- References mer, in addition to the main efect of genre, we found the efect of vector coherence in interaction with epoch and genre, for the latter the diachronic evolution of metaphor processing costs is related to the interaction of word frequency with epoch and genre.
While these diferences might reflect genuine efects of
the syntactic structure and how it impacts metaphorical
predication [
In conclusion, this proof-of-concept study proposed an adaptation from Italian to English of a method employing temporal word embeddings to study the evolution of metaphors. Thanks to this approach, we could elucidate that the processing costs of English literary metaphors is stable over time (diferently from Italian) but is dynamically afected by stylistic features of texts and by single-word measures. The proposed method seems to be sensitive to the specificities of the language under investigation, supporting its crosslinguistic applicability.
of literature, thus enhancing the literary heritage with innovative methods that can provide insights for scholars from a wide range of disciplines. We are aware, however, that the corpora used are not representative of the entire spectrum of varieties of English, but of educated, Western English. Hence, our results may not coincide with the general evolution of the language but provide a partial view of it.
Temporal vector space models and metaphor datasets used in the study are available at https://osf.io/j8bd7/ ?view_only=4cd623d5622b4ed0bd1624c42af0f40$. the 37th Annual Meeting of the Cognitive Science synesthetic metaphors, Metaphor and the Social Society, 2015, p. 2703–2708. World 13 (2023) 59–80. doi:10.1075/msw.00029. [14] D. Wilson, R. Carston, A unitary approach to lex- win.
ical pragmatics: Relevance, inference and ad hoc [26] A. N. Katz, A. Paivio, M. Marschark, Poetic comconcepts, in: N. Burton-Roberts (Ed.), Pragmatics, parisons: Psychological dimensions of metaphoric Palgrave Macmillan, Basingstoke, 2007, p. 230–259. processing, Journal of Psycholinguistic Research [15] M. Fludernik, D. C. Freeman, M. H. Freeman, 14 (1985) 365–383. doi:https://doi.org/10.
Metaphor and beyond: An introduction, Poetics 1007/BF01067881.
today 20 (1999) 383–396. [27] E. Shutova, Design and evaluation of metaphor [16] A. N. Katz, A. Paivio, M. Marschark, J. M. processing systems, Computational LinguisClark, Norms for 204 literary and 260 nonliter- tics 41 (2015) 579–623. doi:https://doi.org/10. ary metaphors on 10 psychological dimensions, 1162/COLI_a_00233.
Metaphor and Symbol 3 (1988) 191–214. doi:10. [28] V. Mangiaterra, C. Barattieri di San Pietro, V. Bam1207/s15327868ms0304_1. bini, How literary metaphors change over two [17] E. Semino, G. Steen, Metaphor in literature, in: centuries (in prep).
J. R. W. Gibbs (Ed.), The Cambridge handbook of [29] A. Englhardt, J. Willkomm, M. Schäler, K. Böhm, metaphor and thought, Cambridge University Press, Improving semantic change analysis by combining Cambridge, 2008, pp. 232–246. word embeddings and word frequencies, Interna[18] V. Bambini, D. Resta, M. Grimaldi, A dataset of tional Journal on Digital Libraries 21 (2020) 247–264. metaphors from the italian literature: Exploring doi:10.1007/s00799-019-00271-6. psycholinguistic variables and the role of context, [30] Q. Feltgen, B. Fagard, J.-P. Nadal, Frequency patPloS one 9 (2014) e105634. doi:10.1371/journal. terns of semantic change: corpus-based evidence of pone.0105634. a near-critical dynamics in language change, Royal [19] V. Bambini, P. Canal, D. Resta, M. Grimaldi, Time Society open science 4 (2017) 170830. doi:10.1098/ course and neurophysiological underpinnings of rsos.170830. metaphor in literary context, Discourse Processes [31] H. Al-Azary, A. N. Katz, On choosing the vehicles 56 (2019) 77–97. doi:10.1080/0163853X.2017. of metaphors 2.0: the interactive efects of seman1401876. tic neighborhood density and body-object interac[20] S. Bhatia, R. Richie, W. Zou, Distributed seman- tion on metaphor production, Frontiers in Psycholtic representations for modeling human judgment, ogy 14 (2023) 1216561. doi:10.3389/fpsyg.2023. Current Opinion in Behavioral Sciences 29 (2019) 1216561.
31–36. doi:10.1016/j.cobeha.2019.01.020. [32] M. Honnibal, I. Montani, spacy 2: Natural [21] F. Günther, C. Dudschig, B. Kaup, Latent semantic language understanding with bloom embeddings, analysis cosines as a cognitive similarity measure: convolutional neural networks and incremental Evidence from priming studies, Quarterly Jour- parsing (2017). doi:https://doi.org/10.5281/ nal of Experimental Psychology 69 (2016) 626–653. zenodo.3358113.
doi:10.1177/174569161986137. [33] P. Hanks, Metaphoricity is gradable, Trends [22] M. N. Jones, J. Willits, S. Dennis, M. Jones, Models in Linguistics Studies and Monographs 171 of semantic memory, in: J. Busemeyer, Z. Wang, (2006) 17. doi:https://doi.org/10.1515/ J. Townsend, A. Eidels (Eds.), Oxford handbook of 9783110199895.17. mathematical and computational psychology, Ox- [34] V. Bambini, M. Trevisan, Esploracolfis: ford University Press, New York, NY, 2015, pp. 232– Un’interfaccia web per le ricerche sul cor254. pus e lessico di frequenza dell’italiano scritto [23] S. McGregor, K. Agres, K. Rataj, M. Purver, (colfis), Quaderni del Laboratorio di Linguistica 11 G. Wiggins, Re-representing metaphor: Modeling (2012) 1–16. metaphor perception using dynamically contextual [35] L. Han, A. L. Kashyap, T. Finin, J. Mayfield, J. Weese, distributional semantics, Frontiers in psychology Umbc_ebiquity-core: Semantic textual similarity 10 (2019) 765. doi:10.3389/fpsyg.2019.00765. systems, in: Second Joint Conference on Lexical [24] N. J. Reid, H. Al-Azary, A. N. Katz, Cognitive fac- and Computational Semantics (* SEM), Volume 1: tors related to metaphor goodness in poetic and Proceedings of the Main Conference and the Shared non-literary metaphor, Metaphor and Symbol Task: Semantic Textual Similarity, 2013, pp. 44–52. 38 (2023) 130–148. doi:10.1080/10926488.2021. [36] T. Mikolov, K. Chen, G. Corrado, J. Dean, Eficient 2011285. estimation of word representations in vector space, [25] B. Winter, F. Strik-Lievers, Semantic distance pre- arXiv preprint arXiv:1301.3781 (2013). dicts metaphoricity and creativity judgments in [37] E. Bruni, G. Boleda, M. Baroni, N.-K. Tran, Distri