préparation d'un numéro spécial de la revue Angles sur "digital subjectivities" 2017-2018
In his famous classic “What Computers Can’t Do” (1972), Hubert Dreyfus outlines the limits of computational codification of reality. His point is that binary codes cannot execute those functions of human understanding, human behavior and most of all human skills, that today we would call embodied and embedded cognition. Still nowadays Dreyfus’ insights constitute a challenge for software-development: While Computers easily outdo humans in code processing, embodiment still constitutes a major problem—beating the best human chess players is an easy task, while making a humanoid robot run through a woods is not.
Similar problems of code-based computation are not limited to extra-linguistic forms of cognition – they also take place inside language processing, because linguistic meaning is not just a product of codes, it is also a product of the execution of human skills. To give an example: Semantically, mood markers used in musical scores—like, e.g. “adagio eroico”—seem very vague. The mood-marker can become very precise, though, if exposed to a more pragmatic conception of language; more specifically: if it is not read for the purpose of content-production, but for the purpose of ‘acting out’.
Meaning, then, does not take place as an effect of signification alone; it is rather displayed in the execution of a skill—and hence, the precision of this meaning does not stem only from signification alone, but from this enacted skill as well. So it does not come as a surprise that terms like “adagio eroico” usually are neither translated nor analyzed; their meaning is rather learned through music lessons and performance and listening. Nor does it come as a surprise that the precise meaning of these words cannot come about in a-modal or context-resistant forms of semantics, but as a precise orientation of nuanced bodily action from which it cannot be simply abstracted. Like other enactive forms of meaning, it comes about “without content” (as Daniel D Hutto would argue).
The aim of my talk is to discuss these problems of codification along the challenges some metaphors expose to computational language processing in translation machines. While older software tried to handle metaphors in terms of semantic congruence, the newer (and much more successful) programs use Big Data analysis that simply calculates the statistical probability of the occurrence of one word over another – thereby only caring for linguistic routines while omitting any question about the content. The success of this approach is very much in the line with what was stated above: The calculation of probability involves the precision of trained readers having developed a precise feel for the meaning rather than a concept of what was signified.
This fact helps raise Dreyfus’ question a bit differently: Evidently, statistical computation can execute operations content-aimed codification ‘can’t do’. But the open question is what this process can tell us about the semantics, i.e. the meaning production which, as stated above had to be bracketed to make the mentioned machines work. My talk uses metaphor to ponder about the relation between codification, statistical computation and human meaning production.
About the author
Jan Söffner holds a PhD in Italian Studies and a 'Habilitation' (second, post-doctoral dissertation) in Comparative Literature and Romance Studies. From 1999 to 2007 he was research associate at the Department of Romance Studies at the University of Cologne; and from 2008 to 2010 he worked at the research project Emotion and Motion at the Centre for Literary and Cultural Research (Zentrum für Literatur- und Kulturforschung) in Berlin, from 2011 to 2014 he was based at the Morphomata Center for Advanced Studies in Cologne (as research associate and fellow), and from 2014 to 2015 he replaced a chair in Romance and Comparative Literature at Eberhard-Karls-Universität Tübingen.
Metaphor; embodiment; enactivism.