Adversarial Poetry – Subversive embeddings in artificial neural word space

In the 21st century we have seen that as soon as social subjects and social movements aim for political visibility, they become inscribed in opaque layers of new technologies, whether they intend to do so or not. My doctoral project aims to show that this does not necessarily mean submitting to their laws. Also, that it is possible – if one knows which rules to break, and how to break them, without bringing down the overall set of rules – to destabilize the socio-technical space of action into which one enters in order to become politically visible. How one appropriates this space with poietic rule-breaking and thus create new participatory spaces will be briefly outlined below.

With the advent of information technologies and computer-assisted communication, the analysis of social relations and networks has also experienced a revival, so to speak. Especially in times of social unrest, these networks and spaces of action are increasingly observed, measured, and controlled, not only for the sake of functioning or for their own business interests, but also to measure the reactions of the public and to estimate the duration and severity of the associated protests. Therefore computational modeling of text analysis processes has become a major role. In disciplines such as the life sciences, economics, psychology, or the digital humanities, they became a central technique for deriving reliable insights from data about social phenomena, obtained through observation. These predominantly computational linguistic insights flow unobtrusively into apps for the private sector, government agencies, intelligence agencies, police departments, and the military. Quite often they are used for soft biometric procedures. At the turn of the millennium, it were precisely these procedures that successfully merged with cognitive technologies such as deep learning to extract organizational units and parts of social subjects, identify them, and transform them into formal, computer linguistic levels of description (morphology, syntax, semantics, aspects of pragmatics, etc.). In this intermediate form, they serve as operands for a variety of models of Natural Language Processing (NLP).
NLP can be seen as a mixed science consisting proportionately of Computer Linguistics, Computer Science, and Artificial Intelligence: the science of algorithmic processing of language, the science of processing data, and the science of artificial, intelligent behavior.
Repressive instruments of domination, based on Language modelling preventively limit the political power of emancipatory movements, not only in the virtual world, but also on the streets, up to the point of the absolute political inability to move. It is the power of the lyrical in the numerous attempts of emancipatory movements to manifest new cultural norms through language that pose the greatest challenge to the computational processing of natural language. For those new plays with language, the slangs, such as that of the Black Lives Matter movement f.ex., through which a non-binary, anti-sexist and anti-racist vocabulary can be established within the movement, the meaning vectors (word embeddings) of pre-trained NLP language models are not matched. Language modelling techniques always require a formal framework, a collection of a priori knowledge about the language, the social context being analyzed. In this new vocabulary, these are not yet set and cannot be easily trained to the machines.
To exploit this scope within language models, in my doctoral project I am developing a subversive strategy called „Adversarial Poetry“. For this I elaborate a concept in how combinations of poetry and hacking can be used precisly to scramble neural word embeddings in language models. By deliberately changing the codes, the respective text genre, form, or even individual characters and signs in the composition of political writings, of tweets, forum posts, emails, etc., Adversarial Poetry thus inscribes poietic (primarily syntactic) rule-breaking into the respective socio-technical language and action spaces. In spaces where the linguistic formation of spaces of meaning to be filled by the reader can no longer be found on the phenomenological level (Heilbach).
The aim of Adversarial Poetry is, to offer emancipatory social movements a literary construction kit to protect themselves from repression and to create participatory open spaces within and outside these socio-technical spaces of action. It can be seen as a strategy to serve as an interspace between aesthetic practice and technological and political critique. It unfolds a new space of possibility for individual and collective positioning vis-à-vis constitutive inequalities and the patterns of domination inscribed in modern liberal democracies and their instrumentalities. Spaces of emancipation, empowerment and visibility have to be fought for anew today, this is beyond question. In particular, cultural spaces as a place of expression, where those to whom political spaces remain largely closed have an opportunity to articulate their say.
Nowhere, than in social media networks, is it currently more apparent that the computational processing of natural language and of interpersonal relations is rather unsuitable for the promotion of politically liberal action and as a tool to support the creation of liberal and fair societies. On the contrary, they systematically promote dominant languages with their rigid terminologies and precision in language. The social developments that we can read from the social networks in recent years show us that these networks are not only passive carriers of signs, but also active producers.

Well, let‘s code new participatory spaces together!