Sound and Sense in Poetry: New Light from Algorithms (abridged preprint)

Abstract

This article examines the relationship between sound and meaning in poetry, with a focus on Shakespeare’s sonnets. It discusses a recent study that applies computational methods to analyze phonetic and semantic patterns in poetic texts.
The author reviews historical debates about sound-meaning correspondences in language and introduces the concept of “phonosemantic antiphrasis” in poetry.
The paper compares traditional phonosemantic analysis with newer algorithmic approaches, using examples from Shakespeare’s works. It considers the potential application of these methods to other poets, such as Dante.
The article also touches on broader implications for linguistic theory, cognitive poetics, and the philosophy of language.
It concludes by reflecting on the interplay between computational analysis and human interpretation in understanding poetic language.

Rodolfo Delmonte, from the University of Venice, has recently documented (Computing the Sound–Sense Harmony: A Case Study of William Shakespeare’s Sonnets and Francis Webb’s Most Popular Poems, “Information”, 2023, 14, 576) how, in poetic text, sounds that should evoke pleasant sensations and joyful moods are sometimes applied, with either ironic or dramatic effect, to ideas and content of the opposite sign, creating a sort of “phonosemantic antiphrasis.”
Many poetry readers have certainly noticed this phenomenon, perhaps only vaguely, more at the level of intuition and sensation than precise consciousness, without having the mathematical-statistical tools to document it.
In particular, open vowels would tend to evoke brightness, peace, serenity; closed vowels surprise, seriousness, rigor, gravity. Voiceless stop consonants harshness and severity; voiced consonants pleasure, softness, lightness.
The Sparsar algorithm (System for Poetry Automatic Rhythm and Style AnalyzeR) identifies with absolute precision the exact phonetic value of letters (while, as the author emphasizes, Artificial Intelligence currently proves fallible in this as in many other aspects of literary analysis).
In parallel, Latent Semantic Analysis and Sentiment Analysis (methods developed mainly for the web and marketing, but also applicable to literary analysis) objectively illustrate the dominant meaning cores and emotional traits.
This new methodological approach merges the two areas, finally rigorously identifying the connection between sound and sense.
An old study (B. F. Skinner, The alliteration in Shakespeare’s sonnets: a study in literary behavior, “The Psychological Record”, 1939, 3) showed, data in hand, that the presence of alliterations in Shakespeare’s sonnets was no higher than what one might expect from a purely random process.
However, this conclusion only highlighted all the limitations of a purely quantitative approach. Sound effects will or will not be motivated (therefore not random, although not necessarily calculated and deliberate in a premeditated way) if one takes into account the connection between sound and sense, and focuses on both in the most objective way possible.
Certainly, the phonosemantic theory (i.e., the intrinsic expressive value of certain vowels and consonants) contrasts with the now peacefully accepted assumption of the arbitrariness of the linguistic sign.
This assumption was already reflected in the often-quoted phrase by Mallarmé (though attentive as anyone to the evocative value of sounds), who (in a text, if not the text par excellence, fundamental to modern poetics, Crise de vers) emphasized the paradox by which the French language expresses night, nuit, with a luminous sound, day, jour, with a dark sound (“À côté d’ombre, opaque, ténèbres se fonce peu; quelle déception, devant la perversité conférant à jour comme à nuit, contradictoirement, des timbres obscur ici, là clair”).
It was Gérard Genette, in a monument of erudition and irony, Mymologiques. Voyages en Cratylie, from 1976, who retraced the efforts, fervor, but also the deviations, fantasies, and fallacies of those who set out to document a natural and necessary link between the sound and meaning of words.
Perhaps standing out among all these is Antoine Fabre d’Olivet, for his rigorous knowledge of Hebrew roots, his ability to trace almost scientifically the elements of language, and the praise that the great linguist Edward Sapir gave him as a “mystic and metaphysician.”
And anyone who returns today to browse, even out of pure scholarly curiosity, d’Olivet’s La Langue Hébraique restituée, from 1815, will see how, in it, A is associated with strength, stability, generative principle; Vav, resolved in OUW, evokes the “inconceivable mystery,” the knot of being and nothingness; Hel expansion and upward motion; Hun (closed vowel and voiceless consonant) darkness and corruption; M motherhood, matrix, matter. It was in the roots, as conceived by d’Olivet, that Sapir saw an anticipation of the modern idea of “phoneme.”
For English, one can mention an old and daring study by Dwight Bolinger, courageously titled The sign is not arbitrary (“Thesaurus”, V, 1949). A study that, however, reflects the main limitation of almost all those conducted so far around phonosymbolism: having limited themselves to the mimetic, onomatopoeic, iconic, in short, merely exterior and material aspect of the relationship between sound and sense; while especially the analysis of poetic text would seem to suggest the possibility of a sort of “metaphysical onomatopoeia,” or “existential phonosymbolism,” in which sounds and their complex juxtapositions and interweaving reflect immaterial and interior emotional resonances, belonging to the domain, if not of the philosophy of existence, at least of psychology or the phenomenology of perception.
Finding “phonosemantic universals,” that is, expressive values of individual phonemes that can have correspondence in different and distant languages, seems like a mirage.
Yet already Greek and Latin treatises emphasized the contrast between the trachytes, the hardness, of explosive consonants, especially gutturals, and the leiotes, the levitas, of voiced ones, especially if liquid, as well as the characteristic of open vowels (the Platonic megala grammata) to suggest greatness, vastness, calm, solemnity, albeit always with reference to underlying semantic nuclei.
Innovative in this regard are the studies of Cesare Marco Calcante, summarized in Eufonia e onomatopea: interpretazioni dell’iconismo nell’antichità classica, Como 2005: who, as Grammont did for French, emphasizes how it is always the sense that elicits, that is, brings out, concretizes, makes pass from pure virtuality to full realization, the evocative power of sounds.
Augustine, in De dialectica, follows this contrast (on one hand words like stridor and clangor, harsh to the ear like what they designate; on the other hand words like lene, synesthetically light to the ear as if transmitting a tactile sensation) reaching however a subtlety of analysis that almost anticipates the reflections of Lacan and Kristeva (in La Révolution du langage poétique) around the first constituents of the Signifier as clues to the Autre, as partial revelations of a pre-rational and primordial psychic substrate.
«We must continue to search until we reach the point where reality sings by analogy in accord with the word, as when we say ‘tinkling of the air,’ ‘blare of trumpets,’ ‘clanking of chains.’ For these words sound just like the things themselves that they signify. But since there are things that make no noise, in these the analogy of touch applies, as if the sweetness or harshness of the letters, touching the sense in a delicate or rough way, had generated names for them similarly to how it touches the hearing: thus the word ‘sweet’ itself sounds sweet.»
An algorithm like SPARSAR could finally bring light to this primordial sound magma that has so far eluded (except for flashes, glimmers, intuitions) a complete investigation.
There seem to be, as mentioned, phonosemantic constants at least within the Indo-European family.
The extension of this network of correspondences to other language families could, in the future, add a piece to the controversial Nostratic Theory (which seems to have now found confirmation from genetics, according to a model that takes into account, starting from a mass of linguistic data, the degree of probability that a new word is born or that a given phoneme changes or is replaced: Pagel, M., Darwinian perspectives on the evolution of human languages, «Psychonomic Bulletin & Review», 2017, 24, 1, pp. 151–157).
Linguistics will have to dialogue, if not with theology (which would risk regressing it to a pre-scientific stage), at least with the philosophy of language, especially in its points of contact with the history of religions: because in the study of phonosemantics, the subjective perception that different cultures (even if at the level of mere pseudo-etymology) have had of the expressive value of sounds (and their matrix phonemes) is at least as relevant as the objective and scientifically ascertained linguistic data.
It is significant that Kūkai (great Japanese mystic of the eighth century) found (based on a Buddhist cosmology) also in Chinese and Japanese phonosemantic values analogous to those of Sanskrit, an Indo-European language. In the syllable OM, contraction of AUM, A is the state of wakefulness, light, supreme energy; U, more opaque, twilight light, revealing dream; M deep sleep: gradation of light, revelation, soft abandonment. RA is fire, KYO is emptiness. Thus the voiced rotary is vital force, the explosive guttural negation and privation.
But, for now, more realistically, it would be interesting to verify if the values identified by the algorithm coincide with those already found, with traditional methods, in previous studies on Shakespearean phonosymbolism.
Among the sonnets that Delmonte’s study focuses on is the eighth (Music to hear, why hear’st thou music sadly?): an ironic, mischievous and bittersweet allusion to the melancholic solitude, narcissistic and autoerotic, in which the interlocutor seems, by his own will, enclosed.
It is difficult to deny that the charm of the composition lies, indeed, for a large part, in the “phonosemantic antiphrasis,” in the bittersweet tint that arises from the mixture between the closed and dark sounds, prevalent in the text, and the amorous sweetnesses evoked and denied, suggested by scattered glimmers («Resembling sire and child and happy mother»: verse, this, whose lightness and brightness are all in the harmony of voiced consonants and open vowels).
Michael Shapiro, in one of the few studies dedicated to Shakespearean phonosymbolism (Sound and Meaning in Shakespeare’s Sonnets, «Language», 74, 1998, 1), based on the distinction between vowels followed by a single consonant and those occluded, instead, by a group of obstruent consonants, with a consequent expressive effect of suffocation and narrowness (of ankos and pnigos the Ancients would have said), dwells on another of those that could be renamed “sonnets of Narcissus.”
It’s enough to linger on verses like these: «Unthrifty loveliness, why dost thou spend / Upon thyself thy beauty’s legacy? / (…) Then, beauteous niggard, why dost thou abuse / The bounteous largess given thee to give», to see how the pleasantness of open vowels contrasts, in a jarring clash, with the harshness of groups of occlusive consonants, giving the idea of a beauty strangled and nullified by its own solitude (a “strangled life” Montale would have said).
Two different approaches (that of man and that of the algorithm) seem to converge towards a not dissimilar outcome.
It would be interesting to apply (with particular reference to the phonic iterations linked to rhymes) a similar method to Dante, paying particular attention to rhyme-words.
For example, the double voiced affricate consonant (zz) seems to emblematically represent what could be defined as “phonosemantic polysemy,” that is, the varied and heterogeneous range of nuances, feelings and descriptive and evocative traits that, by way of so to speak enharmonic, the same phonic climate can assume in different contexts: from crude realism to the grotesque (“Tra’ti avante, Alichino, e Calcabrina,/ cominciò elli a dire, e tu, Cagnazzo; / e Barbariccia guidi la decina. / Libicocco vegn’oltre e Draghignazzo, / Ciriatto sannuto e Graffiacane / e Farfarello e Rubicante pazzo”), from light irony to sfumato and chiaroscuro lingering (“E quale, annunziatrice de li albori, / l’aura di maggio movesi e olezza, / tutta impregnata da l’erba e da’ fiori; / tal mi senti’ un vento dar per mezza / la fronte, e ben senti’ mover la piuma, / che fé sentir d’ambrosïa l’orezza”), from parodic reduction (“Sennuccio, la tua poca personuzza / onde di’ che deriva il desiuzzo / il qual ti fa portare il cappucciuzzo / così polito in su l’assettatuzza”: moreover of doubtful authenticity, and therefore suitable for probing the attributional potentialities of this method), to the sigh of amorous anguish, torn by consonantal hardness (“Così vedess’io lui fender per mezzo / lo core a la crudele che ‘l mio squatra; / poi non mi sarebb’atra / la morte, ov’io per sua bellezza corro: / ché tanto dà nel sol quanto nel rezzo / questa scherana micidiale e latra”).
Perhaps a significant step has been taken towards a “stylometry” that goes beyond the letter and seeks, so to speak, to measure the spirit.
And, likewise, “cognitive poetics” could also overcome the schematism that leads to seeing in the structures of poetic discourse, and in the modalities of its cerebral genesis, a simple reflection of neuronal mechanisms and structures.
The relationship between sound and sense unfolds in a sort of fluctuating space, an oscillating interstice open to an indeterministic multiplicity of chords, counterpoints and solutions: in those “fields of space-time influence” in which, according to John Eccles (Hypotheses relating to the brain-mind problem, “Nature”, 168, 1951) “the mind connects to the brain.”
If (M. S. Gazzaniga, La coscienza è un istinto, Cortina, Milan 2019) DNA is to proteins as signifiers are to signifieds, then in verbal language the relationship between sound and sense may be subject to the same quantum indeterminacy that characterizes, at the genetic level, the “catalyst enzymes” that mediate the relationship between one and the other.
In this aspect of linguistics, as in physics, the role of the Observer will be fundamental.
It is significant and surprising that concepts and terms not far from those of Augustine return in Wheeler (Genesis and observership, “Foundational Problems in the Special Sciences”, 10, 1977).
One will not find, descending into the depths of Nature and Matter (as, in our case, in the most hidden and convoluted folds of language and its origin), a foundational and ultimate level, infinitely deep, at which research stops.
Nor, on the other hand, will those structures proceed “layer after layer, ad infinitum, to a bottomless night.” One will perhaps have to return “to the Observer himself, in a closed system of circular interdependencies.”
Nature and Life are, after all, like a text, a flow and a network of information. This is perhaps the most authentic meaning of the World as a Book, according to the allegory that goes, emblematically, from Galileo to Mallarmé.
Algorithms will certainly limit the arbitrariness of interpretative subjectivity. But the linguist, like the hermeneuticist, will have to return, assiduously and cyclically, to interrogate (although made more aware and cautious by the relative certainty of acquired data) his own consciousness in its intimate relationship with the Word. (Matteo Veronesi)