=Paper= {{Paper |id=Vol-2048/invited01 |storemode=property |title=None |pdfUrl=https://ceur-ws.org/Vol-2048/invited01.pdf |volume=Vol-2048 }} ==None== https://ceur-ws.org/Vol-2048/invited01.pdf 
                     Form-Function Dyads and Computational Rhetoric
                                                   Randy Allen Harris

          English Language and Literature, University of Waterloo, Waterloo, Ontario, Canada.

                                                            Abstract

   What if I stepped upon the stage and offered you a line of small linguistic units, compact dyads
   of form and function. “Compact linguistic dyads,” I tell you, “in which the form in a very real
   sense is the function? Buy the forms. Get the functions for free!” Would you cry “Snake oil!”
   and turn back to your familiar work, or would you lean back in your chair, tap your pencil, and
   wonder about the price?

   The bane of computational argument analysis—indeed, of almost any computational analysis of
   language that cares about meaning and function—is the so-called autonomy of form, a presumed
   arbitrariness of linguistic structure, precipitating out of the long history of structuralism, from
   Saussure through generative grammar, and which still infuses such developments as cognitive
   linguistics and construction grammar. There is form and there is meaning, the thesis states, and
   the only way to link them is through some abstruse and elaborate system of rules, poorly
   understood. Computers are good at finding form: figure out the pattern, build the algorithm, turn
   your machinery loose on a text or a corpus, and it dutifully returns with every scrap and snippet
   that satisfies the pattern. Then what? Under the autonomy thesis, then you to start pretty much
   from scratch to figure out what they mean, what they are doing in the text, with a meagre set of
   abstruse rules. Under a form-function dyad model, the baseline is better. Your algorithms do not
   come back with arbitrary patterns; they come back with an understanding of those scraps and
   snippets.

   This is not snake oil. It is not a magic bullet either, and it comes with a cost. Language is too
   dense and culture bound to succumb to simple cure-alls, and every model requires investment.
   But there are such dyads, predating structuralism by millennia. Studied by grammarians,
   rhetoricians, literary scholars since antiquity, but assigned to the dustbin of ‘mere decoration’ by
   modern linguistics, well before the advent of computers and cognitive science, they are known
   rhetorical figures: brief linguistic patterns manifesting unique iconicities that convey basic
   rhetorical or argumentative functions and that constrain meaning in significant ways. “All for
   one and one for all,” for instance, is not an arbitrary pair of structures of the form, [N [P
   [N]NP]PP]NP. It is an antimetabole, a reverse-repetition figure manifesting iconicities of symmetry,
   balance, and equivalence that convey a reciprocal and mutually reinforcing relationship between
   the individual and the group.

   So, step right up, venture your capital, and take home some rhetorical figures. They are an
   untapped and immensely promising resource for argument mining; heck, to every computational
   approach to language that cares about meaning and function.




18th Workshop on Computational Models of Natural Argument                                                 1
Floris Bex, Floriana Grasso, Nancy Green (eds)
16th July 2017, London, UK