New APPS

By Gordon Hull

I’ve been looking (part 1, part 2) at a couple of articles by Lydia Liu (here and here) demonstrating a Wittgensteinian influence on the development of large language models.  Specifically, Wittgenstein’s emphasis on the meaning of words as determined by their contexts and placement relative to other rules gets picked up by Margaret Masterman’s lab at Cambridge and then becomes integrated into the vector semantics models that underlie current LLMs.  Along the way, Liu argues that the Masterman approach to language, which learns a lot from Chinese ideographs, in this sense goes farther than the Derridean critique of logocentrism.  Here I want to transition to Derrida’s critique, not to criticize Liu’s account, but to see an additional point in Wittgenstein, one that Derrida takes further.

To recall, Liu notes that one effect of the Wittgenstein-Masterman approach is to overturn the logocentrism in Western writing, but that Masterman is doing something different from Derrida, who remains in the space of alphabetic writing:

“For Masterman, to overcome Western logocentrism means opening up the ideographic imagination beyond what is possible by the measure of alphabetical writing. This is important, as it follows that the scientist’s and philosopher’s reliance on conceptual categories derived from alphabetical writing in their commitment to logical precision and systematization as well as their deconstruction must likewise be subjected to post-Wittgensteinian critique” (Witt., 437)

As she adds a few pages later, “I am fully convinced that Masterman is the first modern philosopher to push the critique of Western metaphysics beyond what is possible by the measure of alphabetical writing, and, unlike deconstruction, her translingual philosophical innovation refuses to stay within the bounds of self-critique” (Witt., 444).

This seems basically right, though the case of Derrida and “words” is complex.  Let me suggest another Wittgenstein passage as a starting point for this additional trajectory.  Wittgenstein remarks fairly early in the Philosophical Investigations:

“If I am inclined to suppose that a mouse has come into being by spontaneous generation out of grey rags and dust, I shall do well to examine those rags very closely to see how a mouse may have hidden in them, how it may have got there and so on.  But if I am convinced that a mouse cannot come into being from these things, then this investigation will perhaps be superfluous (PI 52).

I think Derrida is on the same page, but there’s a political edge to Derrida’s analysis that I think is absent in the Wittgenstein.  At least, the political analysis comes out more readily in Derrida. In “Plato’s Pharmacy” (1968) Derrida explores the sense in which writing is a pharmakon (drug, or potion) that escapes Plato’s best efforts to domesticate it by rendering it subordinate to speech.  In one of the myths recounted in the Phaedrus, the Egyptian god Theuth presents writing as a “potion [pharmakon] for memory and for wisdom,” only to be sharply criticized by Thamus, as I note below.  As becomes (relatively) clear, for Derrida writing is the space which inevitably intervenes when truth as presence disappears.  In the essay, Derrida is out to show that Plato, especially in the Phaedrus was not “simply condemning the writer’s activity.”  Rather, “nothing here is of a single piece and the Phaedrus also, in its own writing, plays at saving writing – which also means causing it to be lost – as the best, the noblest game” (Dissemination, 67 [references to this edition]).

The specific issue is our relation to truth.  As Derrida shows by reading a number of Platonic texts (not just the Phaedrus), Platonic – and thus metaphysical and philosophical – truth is meant to be a full, unmediated presence, but this sort of full, unmediated presence is in fact unattainable to us.  In other words, to possess or live in truth would require access to an eidos, but we live as temporal and embodied beings, and at the moment we try to articulate what that eidos is, we’ve slipped away from it.  The best we can do is try to approximate it in language.  Language, however, creates what amounts to a political problem: how do we differentiate between discourse that aims at the truth – that correctly tries to copy, as best as can be done, the truth as governed by eidos – from discourse that does otherwise (say, the sophists)?  The architecture of the Platonic distinction is familiar from the allegory of the Cave, where the distinction is between those who see untruthful shadows and those who see things illuminated by the good.  Derrida puts it this way:

“Socrates thus adopts … the subtle difference between two forms and two moments of repetition: a repetition of truth (aletheia) which presents and exposes the eidos; and a repetition of death and oblivion (lethe) which veils and skews because it does not present the eidos but re-presents a presentation, repeats a repetition” (135).

Compared to living, voiced logos, writing gets it wrong, as writing “would be pure repetition, dead repetition that might always be repeating nothing or be unable spontaneously to repeat itself, which also means unable to repeat anything but itself: a hollow, cast-off repetition” (135).  Writing is “thus from the start stripped of all its own attributes or path-breaking powers.  Its path-breaking force is cut not by repetition but by the ills of repetition” (ibid).

The condemnation of writing we see in the Phaedrus turns out to be a bit of policing, designed to advance the thesis that “live memory repeats the presence of the eidos” (111), whereas writing:

“Appears to Plato (and after him to all of philosophy, which is as such constituted in this gesture) as that process of redoubling in which we are fatally (en)trained: the supplement of a supplement, the signifier, the representative of a representative …. The structure and history of phonetic writing have of course played a decisive role in the determination of writing as the doubling of a sign, the sign of a sign. The signifier of a phonic signifier. While the phonic signifier would remain in animate proximity, in the living presence of mneme or psuche, the graphic signifier, which reproduces it or imitates it, goes one degree further away, falls outside of life, entrains life out of itself and puts it to sleep in the type of its double.  Whence the pharmakon’s two misdeeds: it dulls the memory, and if it is of any assistance at all, it is not for the mneme but for hypomnesis.” (110)

This passage is incredibly dense, so several comments are in order.

(1) The basic hierarchy is from eidos to voice to graphic writing, and the Platonic gesture is that voice is (capable of being) an appropriate repetition of eidos, of approximating and enacting a desire to properly repeat the eidos or to re-member it (think of the boy in the Meno).  The deconstructive critique amounts to pointing out that both voice and writing are repetitions of the missing eidos, and so there is no a priori way for Plato to enforce the hierarchy between them.  Rather, Platonism consists in a series of elaborate conjuring tricks and mythologies designed to shore up the hierarchy and to make it appear non-arbitrary.  As Derrida says, “the opposition between the true and the untrue is entirely comprehended, inscribed, within this structure or this generalized writing. The true and the untrue are both species of repetition” (168).

(2) One recognizes immediately the Platonic critique from the Phaedrus.  Following the discussion of Theuth, who (recall) introduces writing as a “pharmakon for memory and for wisdom” (274e), Thamus criticizes him on the grounds that writing enables what we’d now call cognitive offloading.  This is bad!  Plato has Thamus say that writing “will introduce forgetfulness into the soul of those who learn it …. you have not discovered a potion [pharmakon] for remembering, but for reminding” (275a).  Thus Derrida’s explanatory pairing: psuche-mneme (soul-memory: good) and pharmakon-hypomnesis (writing-reminder: bad).

The other Platonic complaint is that even the reminders of writing are no good, because they are reminders of themselves, and not what is to be understood or recalled. That is, “you’d think that they [written words] were speaking as if they had some understanding, but if you question anything that has been said because you want to learn more, it continues to signify just that very same thing forever.  When it has once been written down, every discourse rolls about everywhere” (275d-e), indiscriminately attaching itself to everyone, whether they should hear it or not.

(3) There is another way of reading this paragraph, one that perhaps speaks to Liu’s comments about Derrida and Wittgenstein.  Because Derrida wants to make the point that not only is there is no such thing as original access to the eidos, but also that the entire apparatus of distinguishing between voice and writing depends on writing (in other words, there is no hierarchy from voice to writing except as a retrospective mythology), he emphasizes voice as “phonetic writing.”  But one could easily enough create a Platonic hierarchy of writing systems, based on their proximity to voice.  In such a hierarchy, ideographic Chinese as Liu presents it would be the ultimate pharmakon, lacking even the “phonetic” aspect of alphabetic writing.

Such a hierarchy is present precisely in Hegel, who thinks Chinese writing is primitive and constraining (though not, as I’ll explore next time, the absolute worst as I think it would be for Plato).  Here is Hegel, from the Philosophy of History:

“They [the Chinese] have, as is well known, beside a Spoken Language, a Written Language; which does not express, as our does, individual sounds — does not present the spoken words to the eye, but represents the ideas themselves by signs. This appears at first sight a great advantage, and has gained the suffrages of many great men — among others, of Leibnitz. In reality, it is anything but such. For if we consider in the first place, the effect of such a mode of writing on the Spoken Language, we shall find this among the Chinese very imperfect, on account of that separation. For our Spoken Language is matured to distinctness chiefly through the necessity of finding signs for each single sound, which latter, by reading, we learn to express distinctly. The Chinese, to whom such a means of orthoepic development is wanting, do not mature the modifications of sounds in their language to distinct articulations capable of being represented by letters and syllables. Their Spoken Language consists of an inconsiderable number of monosyllabic words, which are used with more than one signification. The sole methods of denoting distinctions of meaning are the connection, the accent, and the pronunciation — quicker or slower, softer or louder. The ears of the Chinese have become very sensible to such distinctions. Thus I find that the word Po has eleven different meanings according to the tone: denoting “glass” — “to boil” — “to winnow wheat” — “to cleave asunder” — “to water” — “to prepare” — “an old woman” — “a slave” — “a liberal man” — “a wise person” — “a little.” — As to their Written Language, I will specify only the obstacles which it presents to the advance of the sciences. Our Written Language is very simple for a learner, as we analyze our Spoken Language into about twenty-five articulations, by which analysis, speech is rendered definite, the multitude of possible sounds is limited, and obscure intermediate sounds are banished: we have to learn only these signs and their combinations. Instead of twenty-five signs of this sort, the Chinese have many thousands to learn. The number necessary for use is reckoned at 9,353, or even 10,516, if we add those recently introduced; and the number of characters generally, for ideas and their combinations as they are presented in books, amounts to from 80,000 to 90,000.” (152-3; parts of this passage are discussed by Derrida: see below).

Like the rest of the “character of the Chinese people in its various aspects” (156), Chinese writing is insufficiently spiritual.  The words are too word-like, too graphic, and not close enough to sounds.

I’ve subjected you to such a long quote from Hegel to make a couple of points about Derrida’s critique, which is what I’ll start with next time.