Pedro Zylbersztajn
Independent scholar
Email: pedrozylber@gmail.com
Web: https://www.pedrozylber.com
Reference this essay: Zylbersztajn, Pedro. “[Im]possible Alphabets, or What Does It Mean to Mean in the age of AI?” In Language Games, edited by Lanfranco Aceti, Sheena Calvert, and Hannah Lammin. Cambridge, MA: LEA / MIT Press, 2021.
Published Online: April 15, 2021
Published in Print: To Be Announced
ISBN: To Be Announced
ISSN: 1071-4391
Repository: To Be Announced
Abstract
Artificial scripts have been developed for millennia. These are (usually politically minded) conscious efforts to constitute or crystalize a set of meanings through the establishment and operation of written code. Cryptographers use statistical analysis to unveil hidden meanings not only in purposefully occluded messages but also in long lost natural and constructed scripts. In one case, that of the infamous Voynich Manuscript, centuries of specialists were not yet able to determine the contents of the writing, but have been able to predict that there is enough statistical consistency in the text for it to be considered properly linguistic, and not only gibberish. This suggests one can identify language without identifying meaning. Syntax supersedes semantics in defining ‘linguistic-ness.’ However, one needs to assume that at the very least the inventor(s) and writer(s) of the script would guard its semantic bearings, which would keep the script grounded in actual language—that is, the meaning is hidden, but present, and that is the only way through which syntax can actually be articulated. Building on the discussions on formal linguistics presented in Andrea Moro’s Impossible Languages, on Pierre Bourdieu’s sociological analysis of language use and on Trevor Paglen’s research on invisible images, this paper speculatively addresses the (im)possibility of the development of a completely artificial language and script that would not be comprehensible by any human, even those invested in its creation, but would have syntactic fluency in a way that statistical processes, e.g. machines and AI, could interpret as language.
Keywords: Constructed scripts, meaning, AI, language use, asemiosis
A fascinating act of transduction is language. But we worry. We worry about the imaginary, supplemental alphabets starting with letter twenty-seven.
– Alexander Galloway & Eugene Thacker, The Exploit
This Might Be (Im)possible
This text emerged from the process of development of an artwork that has never reached full existence. During the course of five years of experiments, none of which have come to fruition, I have started to consider not only the original themes of the work but also its very difficulty, or rather, potential impossibility (at least from a human perspective) as one of its central concerns. The initial proposition I set to myself was to create a writing system which would ‘work’ syntactically, but carry no semantic value, testing the sometimes blurred boundaries between written form and language, form and meaning, use and grammar. While experimenting with different approaches, it became apparent that computing and what were then newly popularized neural network technologies could play a significant part in the implementation of this premise, which in turn led the project to address new dynamics of language comprehension in the age of AI.
As one could expect from its inception, this text is anchored in a certain descriptive tension of language in and through writing. As we will see, these are two relatively independent things, which are quite often conflated, in and outside of these paragraphs. While most of the arguments draw from studies of any past and current configuration and display of language—be it written, voiced, numbered or otherwise—the emphasis on text emerges initially from a personal and artistic interest in reading practices, but more profoundly, for the sake of this exposition, on computation’s and AI’s reliance on written modes for the processing of linguistic utterances. The following article is drawn from research notes, loose thoughts and educated conjectures. This account does not intend to be descriptive or prescriptive of the uncompleted work—it is but one of the possible readings of the proposed artwork, and is more speculation than assertion.
But Does It Mean Anything?
Despite what common-sense associations might propose, the relationship between language and writing is far from one of co-dependency. As it is largely postulated, spoken language precedes all known forms of writing by millennia, and, as it is largely known, there are currently myriads of spoken and signed languages which do not rely on or even relate to any form of scripting. Likewise, writing systems only tend to be bound to linguistic use, and are also free to exist on their own, in what has been increasingly named in artistic circles as asemic writing. These are calligraphic or typographic markings which operate under the same visual formal constraints as writing which intends to stand for meaningful symbolism, while carrying no prescribed semantic value. [1]
It would be unreasonable, however, not to acknowledge the symbiotic (or sympoietic) relation of language and writing. It might be fair to say one always suggests the other—that being particularly true of the latter regarding the former. And precisely because of this independent affinity, history has seen many different ways in which they have converged, more prominently through the multiple emergences of so-called natural scripts, but also through the invention of so-called constructed ones. While these are unfortunate terms for those of us who avoid facile oppositions, they bear some significance. Natural scripts are those which have developed in a socially organic way within a community of users, evolving with a speaking culture to map the signs of a host language, and eventually being remapped and in some cases ‘retrofitted’ to attend to (or many times to tame and control) the languages of other communities. The Latin alphabet, used to convey this very text, is one such prominent example. Alternatively, constructed scripts are conscious efforts to constitute or crystalize a set of meanings through the establishment and operation of a written linguistic code. These are deliberately designed by a specific individual or group with specific uses in mind. While this use might be something like the portrayal of a fictional language on screen, most often it is the actualization of the script through regular use in an active community of speakers of a given language. In this case, such designs tend to be politically minded: the Glagolithic alphabet, for instance, is told to have been devised by St. Cyril and St. Methodius as an instrument to facilitate the Christianization of Slavic peoples; [2] the Hangul alphabet was constructed in the 15th century under personal responsibility of King Sejong the Great to substitute the Classical Chinese script and thus promote literacy in the Kingdom of Korea; [3] the N’Ko alphabet was created by Guinean writer Solomana Kanté in 1949 in order to combat harmful stereotypes of West Africans as a culture-less people and strengthening the cultural identity of Manding language speakers, in the wake of post WWII de-colonial sentiment in Africa. [4]
Written objects, of course, are also a kind of time-capsule for languages past. By knowing the language which a lost script encodes, one might break the script, and alternatively, by knowing the script, one might decode a lost language. If faced with an unknown writing system in an unknown language, one might approach it as a crypto-text with no key. By employing cryptanalytic strategies, such as looking for regularities, performing frequency analysis and pairing multiple sources, researchers have been able to achieve understanding of many previously obscured utterances and in some cases entire linguistic systems, unveiling meaning in long lost natural and constructed scripts. That is the case, for instance, with Linear B, a proto-Hellenic syllabary found in the island of Crete, [5] or the Maya glyphs. [6] Some scripts, however, remain more elusive. One such case is Linear A, found in the same area as Linear B, but with sufficient apparent difference to not only being classified as another writing system but one which has not been susceptible to the same code-breaking strategies employed onto its linguistic sibling, thus avoiding over 100 years of attempts at reading it, after its modern unearthing in 1886. [7] Similarly, the Rongo-Rongo script of the Easter Island Rapa Nui, the Southwestern Script from the Iberian Peninsula and the Indus script of the Harappan civilization in Central-South Asia have thwarted the scientific community’s long-lasting efforts in providing access to the information they store and the languages they (partially) conserve. [8]
An example that deserves some additional attention for this article’s purpose, however, does not seem to contain the key to a larger historical understanding, which is why it is all the more puzzling and all the more important for this present analysis. The infamous Voynich Manuscript, an early 15th century codex which contains around 240 pages of undeciphered text and nearly-intelligible illustrations, has been under constant academic scrutiny since at least 1912, when it was rediscovered by the book-dealer who lends it its name, Wilfrid Voynich. Previous to that, it is documented to have belonged to a series of renowned European clerical scholars and scientists, with its oldest assumed owner being Rudolph II, Holy Roman Emperor and King of Bohemia, who acquired it circa 1600. [9] It seems, from the documentation, that throughout its known history, the book has been the subject of inquiry and successive attempts at decipherment, as its script had been deemed unreadable even from its first appearances in historical sources. [10] This means it has withheld the mystery of its contents for over 400 years, despite increased visibility and deployment of resources in the last one hundred.
The script that drives the texts is lonely in its existence: no other similar piece has ever been found. It seems to dwell only on these particular vellum pages, an artifact that refers only to itself and that does not bear particular connections to known or even partially known socio-historical contexts, let alone lost languages and societies. Its carbon-dating, its documental history and even its most prominent aesthetic features seem to localize it in late Medieval Europe, a period and geography which is largely historically accounted for in the West. These characteristics seem to make the script hard to place in its legitimacy: while something like Linear A can be found in multiple archaeological evidences, the group of which being linked to a purported community of users with historiographic endorsement, the Voynich manuscript cannot. This means that there is an ever-present questioning of the mere suggestion that it is a meaningful script at all, and not just a hoax or some piece of glossolalia in the form of gibberish, purely asemic markings. Researchers have to go to extra lengths to first establish a reasonable assumption of the existence of an underlying linguistic sense to the text, to then try to find it. While still a disputed subject, many reports point to conclusive evidence that there is indeed a solid basis to consider that the undergirding content of the script is some sort of (if still very unclear which) language.
The methodologies vary, but are mostly around the previously described cryptoanalytical toolset: in search for regularities, inspectors of the text have found statistical patterns that correspond to that of natural language. The frequency and disposition of the characters seem to be too coherently determined for a randomly written text and the word entropy of the text sits at a similar bit rate to that of Latin languages. [11] Others have suggested that the word structure and some syntactic features that appear to be present in the text seem to be statistically compatible with that of East Asian languages, adding to the confusion but furthering the argument for the book’s linguistic legitimacy. On top of that, computational models which intended to create a topology of words within the text, clustering and networking terms on the basis of their occurrence and concentration deduced that “it differs from a random sequence of words, being compatible with natural languages,” [12] and that on 90% of the analyzed systems, the results present similarities to known corpuses such as the New Testament. [13] Whereas there is still a plausible assumption that the text might be encoded with a cipher, it does not seem to fit any of the known encryption schemes existing in the time of its making, and graphological analysis of the ductus suggests a smooth flow of ink, different to what would be expected of the less fluent experience of writing encrypted, illegible text. [14]
In any case, all of these analyses, even the ones who purport to disavow the text as gibberish, have one epistemological attitude in common at their bases: they all suggest that one can identify language without identifying meaning. Syntax supersedes semantics in defining ‘linguistic-ness.’ However, one still needs to assume that at the very least the inventor(s) and writer(s) of the script guard its semantic bearings, which would keep it grounded in ‘actual’ language—that is, the meaning is hidden, but present, and that is the only way through which syntax can actually be articulated. Would it ever be possible, though, to construct a script which under all the analyses typically performed to an occluded or otherwise inaccessible message ‘accuses’ language, but which no one, its creator included, can decipher?
One can approach this question from two main perspectives, which are not necessarily mutually exclusive: a humanistic one, in which we identify a human creator and a human circulation of the script, and a machinic one, in which we identify a computational agent as the pivot for the construction and consumption of such writing system. The first perspective might elicit questions on the nature of human language, and discussions on the actual feasibility of this task might resonate with what the Italian linguist and neuroscientist Andrea Moro elaborates as “the boundaries of Babel,” [15] the limits of possibility of linguistic existence. The second one, which raises less questions on the realm of technical expediency, seems particularly relevant to inquire into the nature of language in the age of AI, in times when, much like the machine-to-machine images discussed by Trevor Paglen, machine-to-machine text which never even surfaces to the visibility of humans is becoming more and more the dominant flow in our informational landscape.
It May Not Mean Anything
To elaborate on the first perspective, it is worth considering Moro’s book Impossible Languages. This self-proclaimed “short book” [16] summarizes and localizes much of the linguist’s work in the past decades, and functions to some extent as a reader of formal linguistics as it interfaces with neuroscience and neuro-imaging technologies. The author produces a significant account of the current state of theoretical research on the human faculty of language. His main analytical framework is the search for what would be an impossible language, considering that through defining these impossibilities, one would be able to discern what is then the group of imperatives that constitute the minimal definition of a possible language. In the author’s words, “what is a language made of that there can be an impossible one?” [17] Some properties, which he determines as trivial, play a role in making a language not only possible but usable: “[it] must be rich enough to carry all the meanings we can think of; it must be capable of allowing logical inferences and coherent pragmatic interpretation; it must lend itself to communication; and ultimately, children must be able to spontaneously acquire it within the first years of their life.” [18] Human language, however, is not bound by a certain kind of law, in the same way as a possible animal would be bound by the laws of physics. [19] How would one go about defining, then, the aforementioned “boundaries of Babel,” [20] the limits of our cornucopia of linguistic possibilities? The answer, for this book, is deeply grounded in Chomskyan theories of universal and generative grammar, formal linguistics, and a study of the biological faculties that are involved with language processing in the brain.
The analysis of language, both as a general abstraction and as its specific implementations, is emphatically syntactical, within Moro’s framework. From a formalist perspective, human language is:
a set of discrete primitive elements (to put it broadly, a dictionary of words, each constituted by an arbitrary association of sounds and meanings) and a set of rules of combination (its ‘syntax,’ to use a word coming from Ancient Greek meaning ‘composition’) that use these primitive elements to generate a potentially infinite array of structures, which are then interpreted at the relevant interface. [21]
It differs, however, from the idea that any combinatorial rule system would form a sufficient basis for linguistic processes, an information-theoretical view that would assume any linear group of statistical regularities with the aim of communication to have the properties of human language. [22] The author takes the lead from Eric Lennenberg, an MIT psychologist who in 1967 wrote Biological Foundations of Language, a foundational treatise on the subject, to dismiss any suggestion of arbitrariness or cultural convention in language structure (which Lennenberg ascribes to “Wittgenstein’s followers” [23]) and assert the importance of the study of the formal faculties of language as biologically determined.
If syntax is the core of our faculty of language, then order and its source must determine a liminal point in what is possible and impossible in language. And while this faculty is biologically determined, Moro claims that “the origin to the source of order in human language can be approached from within language […] without necessarily referring to other cognitive domains.” [24] He establishes a dialectical opposition that structures how Western thinkers have historically organized reasoning around the constitution of order. Tracing this dialectic back to Hellenistic philologists in a debate that unfolded between the Alexadrian and the Pergamon schools, he defines two extreme ‘etiological’ principles for this subject: analogy and anomaly. In the former, “order is seen as emerging from an unformed, infinite magma through the spontaneous development of symmetric analogical relations.” [25] The latter, however, considers that “order would be formed by unforeseen and unforeseeable fractures in an immense lattice of symmetric regularities, where everything would otherwise be inert since it would everywhere be the same.” [26] And while there is an acknowledgement of analogical processes in the development of morphological structures by children in language acquisition phases, the book is emphatic in denying that analogy holds true for the source of syntactical order. Based on the current state of Chomsky’s generative grammar, Moro is quite direct in his support for the anomaly model. This model holds true, he says, as we understand syntax as a system that uses minimal lexical variations within a recursive combinatorial rule to produce an infinitely complex structure. [27] His reading of the matter seems to remain unclear, however, on whether that would automatically suggest that the recursive combinatorial rule system as it is would represent a kind of linguistic homeostasis; and if it comes indeed from a pre-determined biological feature, how that initial order would have come about in the first place.
To analyze an opposing position to Moro’s in regard to human language we can turn to Pierre Bourdieu. In his compilation of essays Language and Symbolic Power, originally published in French as Ce Que Parler Veut Dire, the sociologist (and self-confessed follower of Wittgenstein [28]) puts forward a case for an analysis of language that not only involves but centers around the social context of its generation and use. Grounding it into his ‘theory of practice,’ Bourdieu insists on noticing that language is not only used as a means of communicating verbal messages but also to pursue and advance power in social relations.
Especially in the first part of the book, named The Economy of Linguistic Exchanges, he develops a strong critique and argues against structural and formalist approaches to linguistics, such as the ones employed by Saussure and, later, Chomsky. Disputing Saussure’s
inaugural act through which he separates the ‘external’ elements of linguistic from the ‘internal’ elements, and, by reserving the title of linguistics to the latter, excludes from it all the investigations which establish a relationship between language and anthropology, the political history of those who speak it, or even the geography of the domain in which it is spoken, because all of these things add nothing to a knowledge of language taken in itself, [29]
Bourdieu searches to incorporate such ‘externalities,’ those being the fact that no linguistic interaction would be free of carrying within itself the social structure to which it pertains, by reproducing it and/or reinforcing it. Thus, the concepts of langue and parole, brought by Saussure, or competence and performance, within a Chomskyan vocabulary, are concerned with an abstract and idealized speaker-listener, involved in a homogenous speaker-community and hence removed from any actual use that could elucidate anything about the production of meaning. While it is true that language is a generative process that produces unlimited amounts of grammatically sound utterances, and that this seems to be a uniquely human capacity, the actual competence that is required from humans within their use of language is not that of unlimited utterances, but that of precise and appropriate utterances à propos, within the constraint of a particular social arrangement.
This happens through a ‘linguistic market’ that informs how the enunciator will consider the enunciated according to its interlocutors and their position in a power scale, and a ‘linguistic habitus,’ a set of dispositions socially constructed within a field of action that inclines agents within such field to operate in certain ways. Such dispositions are the mainly unconscious result of a gradual process of inculcation, in which a group of practices, attitudes and impressions are normalized as a standard. The habitus is in a circular motion with its environment, as it is shaped by it and also shapes it constantly. It is also embodied in its practitioners, in what Bourdieu calls a bodily hexis. The hexis is the physical incorporation and manifestation of the habitus, in the way they behave in that social condition. This, in the case of language, is represented by accents, certain uses of the mouth, vocal posture, etc. All of these dispositions also become markers of authority, as they necessarily embed in them the whole social structure of their production. That means that stratified environments will value certain dispositions that compose a specific habitus, like that of an aristocracy, and devalue others, like that of the working-class. Linguistic exchanges are then symbolic exchanges, administered in a market that circulate different kinds of symbolic capital. Holders of such capital will usually exert their power by reinforcing a specific habitus by making it official, thus allowing for a larger accumulation of it. One of the instruments for the officialization of a certain set of uses of a language is the institutionalization of a single grammar. Through that operation, not only specific meanings but also the very structure of enunciation are transformed by their social context. [30]
While Bourdieu does not manifest interest in using his thought as a supplement to formal linguistics, but as a radically different proposition, there are interesting points in which one theory can be placed in a possible dialogue with the other. In the case of juxtaposing him with Moro, specifically, there seems to be an even more productive middle ground. The linguist produces a significant amount of evidence surrounding the limits of possibility of language in syntax, but never gives much thought to the role semantics play in the boundaries of linguistic-ness. By doing so, he neglects an enormous part of the faculty of human language, one that might be said to be the greatest reason for this faculty: its use. Indeed, the word “semantics” appears for the first and only time on page 52 of Moro’s book. Even if we are to believe that a deep grammar is common to all of us, biologically pre-configured in our brains, people make do with the structure they are given, to use a vocabulary extracted from Michel de Certeau’s theory of appropriation. [31] The search for an impossible language has to relate to certain impossible social configurations of language use as well. It is possible to assume a language that is partially based on biological capacities but also constructed upon arbitrariness and social conventions. What might be an impossible language is one that is either one or the other.
Through this lens, one can say that both these representatives of major and opposing currents in human language understanding reach some kind of agreement in that, be it from a use-centered analysis or from a formalist syntactical analysis, regularity does not equate to linguistic-ness. While it is always there, reflected in how language operates, one does not suffice to stand in permanently for the other. The action of a person constructing the speculated absolutely occluded writing system would be purely mathematical and not linguistic. It would therefore exceed the boundaries of Babel, and not enlarge it. At best, what one could expect from the encounter of humans with our fictional constructed script is a state of rich asemiosis, an infinite suggestion of reading—as that which one can experience by facing the text of the Voynich Manuscript.
But It Might Just Work
Things are not quite the same when considering the second, machinic, perspective. To go back to what has been stated through Moro’s voice, while it is not the case that any linear group of regularities based on a combinatorial rule system would have the effect of human language, it is precisely this information theoretical approach that drives the most employed artificially intelligent systems of today. Neural networks are, at their core, number crunching technologies, which are able to emulate and extend specific aspects of human intelligence through large-scale statistical processing. This means that the way linguistic data is handled by such systems is essentially different to how it is done by humans. While in a formalist approach language is defined by syntax which is in its turn undergirded by meaning, and in a sociological approach the opposite equation holds true, in the AI language processing scheme, meaning does not figure at all, being instead replaced by the idea of information. And information is not what it refers to or what is signifies, it is simply what fits the redundancy to ambiguity ratio of a given encoding/decoding protocol. [32] In that sense, any and all linguistic content that flows through artificially intelligent systems is treated just the same—as plain statistical data.
Google Translate’s 2016 update to what they named the Google Neural Machine Translation system (GNMT) provides a good example of how current developments in machine learning have been blurring the lines between text processing and comprehension. The implementation of GNMT enabled the company to develop experiments in zero-shot translation, that is, “translation by language pairs never seen explicitly by the system.” [33] In effect, what this means is that the shift from rule-based machine translation to deep-learning-based machine translation allowed reasonable success in having words that the system had seen be translated from English to Korean and from English to Japanese to be translated directly from Korean to Japanese without using English as an intermediary and without having been explicitly taught to do so. This raises a quasi-philosophical question which the project researchers state as such: “Is the system learning a common representation in which sentences with the same meaning are represented in similar ways regardless of language—i.e. an ‘interlingua’?” [34] By analyzing a 3-dimensional representation of the internal network data and the word groupings in the generated relational vector space, they go on to say that “the network must be encoding something about the semantics of the sentence rather than simply memorizing phrase-to-phrase translations,” which to them implies the “existence of an interlingua in the network.” [35] While this achievement has provoked a few sensational headlines on specialized media (such as Google’s AI translation tool seems to have invented its own secret internal language [36] or Google Translate AI invents its own language to translate with [37]), these claims should not be read from the perspective of a progress of artificial intelligence systems towards the natural functioning of language, but as a fundamental shift in the ontology of language as it disconnects from its bearings on human experience. Representation and meaning are no longer referent to experience: the word ‘tangerine’ disengages from any notion of the physical object, it’s fleshiness, flavor and cultural significance, to be entirely represented as a point in a vector space, uniquely positioned in relation to other data points such as ‘orange,’ ‘fruit,’ ‘citric’ and every other related or unrelated word known by a given system, through their statistical proximity alone.
That, however, is not a fringe phenomenon, but how the bulk of text artifacts are being processed at any given moment. Our knowledge infrastructures have been reshaped, in the last five to eight years, to incorporate an increasing amount of machine intelligence. Most digital media conglomerates rely on massive machine-led text processing technologies, and from Facebook’s news feed ordering and ad suggestions to Google’s page ranking indexes and translation services, content is being mediated in this way. The way in which computational and artificially intelligent systems handle language, many times absolutely under the threshold of human perception, has become the most frequent mode of contemporary language use. These out-of-sight machinic textual processes represent the lion’s share of how language is being shaped; much like in the case of what Trevor Paglen calls invisible images, which are made and circulated from machine to machine, only rarely reaching human visibility. [38] In both cases, the scale has tipped at least in volume and, as one could try to argue, therefore in relevance. Paglen’s research, even while focusing on visual culture, may provide relevant points to consider the transformation entailed by this new prevailing mode in textual culture. The main maxim we can generalize from the artist’s take on digital images to our take on texts is that, while the majority of the inquiry around them assumes “that the relationship between human viewers and images is the most important moment to analyze,” [39] it is precisely the fact that this class of images is “fundamentally machine-readable” [40] and that “the image doesn’t need to be turned into human-readable form in order for a machine to do something with it” [41] which makes them particularly relevant. As he puts it, that “allows for the automation of vision on an enormous scale and, along with it, the exercise of power on dramatically larger and smaller scales than have ever been possible.” [42] All the more important to our speculation, “the machine-machine landscape is not one of representations so much as activations and operations.” [43] Texts, as much as images, in their machine-machine readings, do not serve a semiotic purpose but a logistical one, and “if we want to understand the invisible world of machine-machine visual [textual] culture, we need to unlearn how to see [read] like humans.” [44] Just as “the theoretical concepts we use to analyze visual culture are profoundly misleading when applied to the machinic landscape, producing distortions, vast blind spots, and wild misinterpretations,” [45] so are our theories of reading, of language and of meaning insufficient in an era of growing AI.
It is in this context that the creation and circulation of the proposed occluded script becomes not only attainable but meaningful in its meaninglessness. This script or alphabet [46] would to all measures behave precisely like any other language-encoding script within the confines of machine-machine textuality. In all that it would avoid the human requirements for language, it would fit the computational ones. For what it lacks in meaning, it provides in regularity. And even without the context a human would need to make internal relations in the text, machines would be more than able to produce a vector-based network of proximity drawing exclusively from statistics, as usual. The boundaries of Babel—a reference to the tower erected in an act of human cooperation in defiance of humanity’s powerful overseer—are not the same for our machinic counterparts. Just as the aforementioned historical examples of constructed scripts have been loaded with political significance, this alphabet has an intrinsically political vocation. As Paglen points out,
machine-machine systems are extraordinary intimate instruments of power that operate through an aesthetics and ideology of objectivity, but the categories they employ are designed to reify the forms of power that those systems are set up to serve. As such, the machine-machine landscape forms a kind of hyper-ideology that is especially pernicious precisely because it makes claims to objectivity and equality. [47]
As a wedge between human signification and machine operation, this alphabet can create a parallelization of agency. The exercise of power is emptied if it can only refer to its enunciator. It disengages these systems from the mediation and control of human communication and meaning-making schemas, for the duration of its occupation of the machine-machine channels, and it frees machine-machine relations from human demands for faux (or forced) semiosis. And finally, to extrapolate from the GNMT’s interlingua example, one could even risk saying this alphabet would be fully realized in generating communication through representation. Just not from anything a human would expect to classify as language.
Author Biography
Pedro Pedro Zylbersztajn is a Brazilian artist. His practice resides in the crossing space between different polysemic (intra)actions, such as reading, drawing, writing, editing, programming and sounding. In particular, his work is inclined to discuss language, rhetoric, translation, the social-artistic-semantic ramifications of technology and the implications of these themes in dynamics of authority. He has participated in a number of group shows, art publication fairs, panels and publications in Brazil and internationally. Pedro had his first solo exhibition at Casamata (Rio de Janeiro) in 2016, and an individual performance at Americas Society Visual Arts (New York) in 2018. Most recently, he has been thinking of coded messages, of indexes and of branches. He holds a Master’s degree from the MIT Program in Art, Culture and Technology and is currently a Postgraduate Fellow at the Art by Translation research and exhibition program (2019-2020).
Notes and References
[1] Francesco Aprile and Cristiano Caggiula, “About Asemic Writing,” SCRIPTjr.nl, March 25, 2018, http://scriptjr.nl/about-asemic-writing/3767
[2] Jean W. Sedlar, East Central Europe in the Middle Ages, 1000-1500 (Seattle: University of Washington Press, 1994), 114.
[3] “The Background of the Invention of Hangeul,” National Institute of Korean Language, last modified January 2018, https://www.korean.go.kr/eng_hangeul/setting/002.html
[4] Dianne White Oyler, The History of N’ko and its Role in Mande Transnational Identity: Words as Weapons (Cherry Hill: Africana Homestead Legacy Publishers, 2005), 1.
[5] Maurice Pope, “The Decipherment of Linear B,” in A Companion to Linear B: Mycenaean Texts and their World, ed. Yves Duhoux and Anna Morpurgo Davies (Louvain-la-Neuve, Belgium: Peeters, 2008), 1-24.
[6] Michael D. Coe, Breaking the Maya Code (London: Thames & Hudson, 1992).
[7] Jeffrey Christopher Stanley, “To Read Words not Images: Computer-Aided Analysis of the Handwriting in the Codex Seraphinianus,” (MSc. thesis, North Carolina State University, 2010), 17.
[8] Ibid., 13-16.
[9] René Zandbergen, “Voynich MS—Long tour: Known history of the manuscript,” Voynich.nu, last modified October 10, 2018, http://www.voynich.nu/history.html.
[10] René Zandbergen, “Voynich MS—17th Century letters related to the MS,” Voynich.nu, last modified April 23, 2019, http://www.voynich.nu/letters.html.
[11] Gabriel Landini, “Evidence of linguistic structure in the Voynich manuscript using spectral analysis,” in Cryptologia 25, no. 4 (October 2001): 275–295. doi:10.1080/0161-110191889932.
[12] Diego R. Amancio, et al., “Probing the Statistical Properties of Unknown Texts: Application to the Voynich Manuscript,” in PLOS ONE 8, no. 7 (2013): 7. https://doi.org/10.1371/journal.pone.0067310.
[13] Ibid.
[14] Lisa Fagin Davies, “Manuscript Road Trip: The World’s Most Mysterious Manuscript,” Manuscript Road Trip, January 17, 2015, https://manuscriptroadtrip.wordpress.com/2015/01/17/manuscript-road-trip-the-worlds-most-mysterious-manuscript/.
[15] Andrea Moro, Impossible Languages (Cambridge: The MIT Press, 2016), 46.
[16] Ibid., “Acknowledgements”, ix.
[17] Ibid., 1.
[18] Ibid., 2
[19] Ibid.
[20] Ibid., 46.
[21] Ibid., 1.
[22] Ibid., 4.
[23] Ibid., 45.
[24] Ibid., 77.
[25] Ibid.
[26] Ibid.
[27] Ibid., 83-84.
[28] Pierre Bourdieu, “Fieldwork in Philosophy,” interview by Axel Honneth, Hermann Kocyba and Bernd Scwibs, Paris, 1985, in Pierre Bourdieu, In Other Words: Essays towards a Reflexive Sociology (Stanford: Stanford University Press, 1990), 9.
[29] Pierre Bourdieu, Language and Symbolic Power (Cambridge: Harvard University Press, 1991), 33.
[30] Pierre Bourdieu, Language and Symbolic Power.
[31] Michel de Certeau, The Practice of Everyday Life (Berkeley: University of California Press, 1984), 29.
[32] Claude Shannon and Warren Weaver, The Mathematical Theory of Communication (Urbana: University of Illinois Press, 1949), 31.
[33] Mike Schuster, Melvin Johnson and Nikhil Thorat, “Zero-Shot Translation with Google’s Multilingual Neural Machine Translation System,” Google AI Blog, November 22, 2016, https://ai.googleblog.com/2016/11/zero-shot-translation-with-googles.html.
[34] Ibid.
[35] Ibid.
[36] Devin Coldewey, “Google’s AI translation tool seems to have invented its own secret internal language,” TechCrunch, November 22, 2016, https://techcrunch.com/2016/11/22/googles-ai-translation-tool-seems-to-have-invented-its-own-secret-internal-language/.
[37] Sam Wong, “Google Translate AI invents its own language to translate with,” New Scientist, November 30, 2016, https://www.newscientist.com/article/2114748-google-translate-ai-invents-its-own-language-to-translate-with/.
[38] Trevor Paglen, “Invisible Images (Your Pictures are Looking at You),” The New Inquiry, December 8, 2016, https://thenewinquiry.com/invisible-images-your-pictures-are-looking-at-you/.
[39] Ibid.
[40] Ibid.
[41] Ibid.
[42] Ibid.
[43] Ibid.
[44]Ibid.
[45]Ibid.
[46] Even though alphabets form a very specific category of writing system which is not necessarily the best one for this particular endeavor, I am using this term to refer to our described proposition when in direct contact with machines as a way to avoid confusion with another frequent usage of the term script in relation to computation, which is somewhat synonymous with high-level programming languages.
[47] Trevor Paglen, “Invisible Images (Your Pictures are Looking at You),” The New Inquiry, December 8, 2016, https://thenewinquiry.com/invisible-images-your-pictures-are-looking-at-you/.
Bibliography
Amancio, Diego R., Altmann, Eduardo G., Rybski, Diego, Oliveira, Osvaldo N. Jr and Costa, Luciano da F. “Probing the Statistical Properties of Unknown Texts: Application to the Voynich Manuscript.” PLOS ONE 8, no. 7 (2013).
Aprile, Francesco and Cristiano Caggiula. “About Asemic Writing.” SCRIPTjr.nl, March 25, 2018.
Bourdieu, Pierre. In Other Words: Essays towards a Reflexive Sociology. Stanford: Stanford University Press, 1990.
Bourdieu, Pierre. Language and Symbolic Power. Cambridge: Harvard University Press, 1991.
Certeau, Michel de. The Practice of Everyday Life. Berkeley: University of California Press, 1984.
Coe, Michael D. Breaking the Maya Code. London: Thames & Hudson, 1992.
Coldewey, Devin. “Google’s AI translation tool seems to have invented its own secret internal language.” TechCrunch, November 22, 2016.
Davies, Lisa Fagin. “Manuscript Road Trip: The World’s Most Mysterious Manuscript.” Manuscript Road Trip, January 17, 2015.
Duhoux, Yves and Anna Morpurgo Davies, eds. A Companion to Linear B: Mycenaean Texts and their World. Louvain-la-Neuve, Belgium: Peeters, 2008.
Jean W. Sedlar. East Central Europe in the Middle Ages, 1000-1500. Seattle: University of Washington Press, 1994.
Landini, Gabriel. “Evidence of linguistic structure in the Voynich manuscript using spectral analysis.” Cryptologia 25, no. 4 (October 2001), 275–295.
Moro, Andrea. Impossible Languages. Cambridge: The MIT Press, 2016.
National Institute of Korean Language. “The Background of the Invention of Hangeul.” Last modified January 2018.
Oyler Dianne White. The History of N’ko and its Role in Mande Transnational Identity: Words as Weapons. Cherry Hill: Africana Homestead Legacy Publishers, 2005.
Paglen, Trevor. “Invisible Images (Your Pictures are Looking at You).” The New Inquiry, December 8, 2016.
Shannon, Claude and Warren Weaver. The Mathematical Theory of Communication. Urbana: University of Illinois Press, 1949.
Stanley, Jeffrey Christopher. “To Read Words not Images: Computer-Aided Analysis of the Handwriting in the Codex Seraphinianus,” MSc. thesis, North Carolina State University, 2010.
Wong, Sam. “Google Translate AI invents its own language to translate with.” New Scientist, November 30, 2016.
Zandbergen, René. “Voynich MS — Long tour: Known history of the manuscript.” Voynich.nu. Last modified October 10, 2018.
Zandbergen, René. “Voynich MS — 17th Century letters related to the MS.” Voynich.nu. Last modified April 23, 2018.