theinexactsciences.github.io

1.1: The Meaning of Meaning

Wrench,

I want to take seriously an idea that both of us seem to believe possible: that at this point, much of our theoretical verbiage for describing linguistic transactions encumbers, rather than enables, breakthroughs in understanding. Problematic terms in my mind include “meaning,” “connotation,” “denotation,” “literal,” “figurative,” “explicit,” and “implicit,” etc. There seem to be real differences in these poles, and they have been undeniably historically useful, but they seem often to rest on false distinctions or conflate meaningful differences, e.g. much of our “literal” language is merely metaphor made “stale” or conventional, and much of “explicit” meaning is conveyed through the same processes of context-based inferences that implicature supposedly relies on. Perhaps the best we can do is identify key parts of linguistic transactions, identify which are typically called “literal meaning” or “implicature” etc—and then, from this overview of the whole, suggest some alternate way of carving things up—of lumping and splitting.

In trying to start again from the beginning, in the hope of stumbling upon some conceptual clarity, it seems best to ground our understanding in the very simplest, atomic known properties of language: that it is associative, conventional, and conventionally associative. Signification—a pointing or evocation—is possible on the basis of understood connections—understood associations among words, among real-world phenomena, and also between words and phenomena (the inter & the intra). Different schools of thought have approached meaning as resting in the intension (i.e. in its definition through other words) or in its extension (i.e. the real world set of phenomena it refers to)—or else, similarly, have grounded meaning in the structurally differential relation between words (as in structuralism) or the combinatorial construction of semantics as grounded in correspondence (as in positivism). Perhaps there is an approach which meaningfully de-prioritizes any single grounding of “meaning,” insofar as a correspondent system is also coherent (since the world it corresponds to is coherent).

In addition, we can posit that linguistic production is always an action, while an interpretation is an active perception, i.e. a perception structured in the service of informing a response (i.e. re-action) of one’s own. This response may only be silence, but silence is still a response.

This response-oriented perception I will call “reading,” in the sense that a socialite “reads” a room, or a salesman a customer. It is perception structured in the service of reactive winnowing and triage, and recent advances in cognitive science give us good reason to believe that all perception is action-oriented; in other words, we can never get away from interpretation or “reading,” filtering out what is relevant to our aims and actions from what is not. I sit down to a meal and “read” my plate, just as I also “read” the behavior of the waiter, and “read” the situation with a dinner companion to decide whether to begin eating without them.

In other words, what I “read” from a scene, or “text,” is what the “meaning” of that text is “to me”: it is the relevant set of entailments with respect to my project-at-hand. If that project-at-hand is merely to receive a piece of advice or information being imparted to me, I may attempt to best understand the intention of the speaker, to reverse-engineer the concepts and relationships that their speech attempts to describe. But speech is frequently not so ideal (see Cappelen/Dever, Bad Language 2018)—the picture sometimes painted of conversation as a cooperative exchange of knowledge is far from the full story. In a debate setting, I may deliberately avoid genuine attempts to understand an interlocutor’s intent, hanging them on the noose of technicality. Or rather, it is not so much that I am deliberately avoiding as much as it is that their intent is largely irrelevant to my project-at-hand, which may be to win over the audience. Thus my model of what the audience sees in my opponent’s speech is more important than the opponent’s intent.

Now, I think the best thing is to take these basic dynamics—association, convention, action-oriented perception, and goal-oriented production—and look to other, non-linguistic situations where these dynamics are in place, and to build up from them to understand what additional features or properties are true of language narrowly. The three I want to look at first are:

  1. Tracking an animal in the woods.
  2. Driving down a city street.
  3. Dressing up for a social event.

All of these are instances in which an individual reads the environment, deducing some “meaning” in the sense of an actionable set of entailments. But the cases slowly build up with increasing particulars of human language, introducing man-made convention of meaning (as in traffic signals) and the social landscape of inferences (as in fashion). I won’t get to the third, fashion, in this letter, long as it is already, but instead, after the example of driving down a street, will discuss Knapp & Michaels’s “Against Theory” (1982), which argues that the “meaning” of a text “is”—fully and completely—its authors’ intention (the two do not overlap but are identical). This view doesn’t make sense within the context of how we normally use the word “meaning,” as we’ll soon see.

1.

What do animal tracks “mean”? There are two ways to answer this. Knapp and Michaels would contend that animal tracks don’t “mean” anything at all, since there is no communicative intent at-hand. But popular parlance suggests that, indeed, animal tracks “mean” quite a bit—that any entailments, relevant to a human’s project-at-hand, are its meaning. If I am out looking for my lost pet, the tracks mean Sparkey may yet be alive. To a tracker hunting game, with all his local hermeneutic expertise, they will “mean” quite a bit more. To put it succinctly, meanings are scoped to an activity, hence the idea of a “read” as action-oriented.

To understand how these tracks “mean,” we have to understand the environmental gestalt, or “scene,” in which the tracker exists. He must pick out sensory patterns from this scene which are relevant to his goal. We’ll assume for the sake of example that the tracker is hunting a deer. He will attend to the visual and spatial gestalt looking for signs of disturbance that indicate—that mean—an animal has passed through.

The tracker draws an entailment inferentially. Inference is performed statistically. We can take the shape of the animal tracks: there is a given set of possible deer hooves that are present in almost all deer, short of an extreme genetic abnormality. Its range of possibles, in size and shape, in turn gives way to a range of possibles for the impressions left by hooves in the mud—the tracks. It is of course _possible _that a beaver’s passage is somehow presenting mud impressions resembling deer tracks, but it is highly statistically unlikely. Thus the tracker can make an inference that the tracks “mean” a deer likely has passed through a given area.

The size of an impression in the mud correlates strongly with the size of the deer’s hooves. The size of hooves correlates strongly with size and age. Thus, to an experienced tracker, the tracks “mean” that a deer of a certain approximate size and age has recently passed through. Depending on whether this is the kind of deer the tracker wishes to hunt, the tracks “mean” (entail) that he will or will not choose to pursue them.

The same is true of the orientation of tracks, the direction of the hooves. Virtually all deer hooves face the same forward direction, with the rounded end behind them and the middle “toed” split facing forward; moreover, virtually all deer movement is forward; backing up is rare. Thus the direction of the movement can be statistically inferred. The tracker “reads” the scene by recourse to known _statistical associations _between disparate things. Here, there is a visible trace in the mud of an event that already happened (and is thus “invisible” to the tracker)—this visible trace, by association, “testifies” to the unseen. The tracks “mean” insofar as they probabilistically entail something pertinent to the organizing context at hand.

Before we move on to our second case study, driving, it will help to consider how this scenario would change if, instead of tracking an animal, one was tracking a human being, as in the case of a slave catcher. Morally perverse as such a hypothetical may be, it’s incredibly useful in isolating and pointing to those elements of “reading,” meaning, and inference that are altered by interacting with another person—if only because a person has higher symbolic reasoning and theory of mind capacities than a deer or fox does. These capacities prove crucial because they allow the tracked entity to model _how _his behavior will leave behind a trace, in other words, how it will be “testified” to, and the inferences which his tracker will draw on and make decisions from. Insofar as the tracker’s behavior is relevant to the escaping slave’s project of evading capture, certain inferences and actions by the tracker are preferable, for the tracked party, to other ones. So: where “reading” a scene means an action-oriented perception of entailments, the tracked person might create—i.e. “write”—a scene so as to create, in the tracker, a reading. The syllogism is straightforward: the function of a reading is to inform action; the function of a writing is to inform a reading; and thus the indirect function of a writing is to inform action (at least in the context of tracking—whether this is true of actual literary production and reception will come much later; we’re sticking to analogues of language for now).

What allows the tracked person to write a scene? Two things: an understanding of statistical correlations, e.g. between the orientation of tracks and the direction of movement, and an understanding that the tracker is cognitively similar (i.e. a fellow prediction or inference machine), and can and will leverage these associations to make inferences. From here, in order to achieve a goal—to evade the tracker—the tracked may attempt to produce false inferences in the interpreting tracker. The tracked may double back, retracing steps in reverse, or snap branches off in the wrong direction. Understanding which “indicators” metonymically “testify” to some larger, invisible reality (here temporally, rather than spatially, invisible, since the behavior of interest to the tracker—the presence and movement of the tracked—occurred at a different time) the tracked is able to create a testimony that differs from reality, to produce an inference that is false—in a word, to deceive.

(At the same time, of course, the tracker is aware that this is an _adversarial _relationship, that such fabrications might occur, which will greatly alter his own strategy of reading from that employed in animal tracking. Understanding the motivational structures, i.e. the “game,” of the reader and writer, will help understand reading and writing—and this motivational structure is never so simple as “conveying information” and “receiving information,” as purist literary models sometimes suppose.)

2.

Now we’ll take driving down a street. A driver, like a tracker, is situated within a complex “scene”—a vast information space which engulfs him. He must pick out those bits of information that are pertinent to accomplishing desired tasks, here getting from point A to point B efficiently without accidents, tickets, anti-social effects, etc.

Suppose now the driver comes to a stoplight. For the driver’s purposes, there are two main aspects orienting his action: whether the intersection can be crossed safely and legally (depending on the driver, the latter may alternatively mean “not getting ticketed”). A convention has been established—by a combination of law, public participation, and history—between the colors of the traffic light and whether it is safe and legal to cross. This association is “man-made” in the sense that is _conventional _(some would—mistakenly—call this “arbitrary,” but it is anything but). There is no inherent physical correlation like there is between the size of animal tracks and the size of the animal. But all the same, the convention structures behavior and the final effect is the same: a statistical correlation between color and a capacity for action (being able or unable to cross), which is learned through experience.

There is an additional dimension to the signal, beyond conventionality. There is now a social aspect: those waiting behind you will _expect _you to go on green, and those waiting perpendicular to you at the intersection will _expect _you to stop on red. These expectations further pressure a kind of response—but they also evidence a second statistical correlation: the expectation emerges because drivers _typically _go on green, and stop on red.

Knapp & Michaels: “Against Theory”

Here is what is uncontroversial about literary transmission: the author has some communicative or significative “intent”; that intent is encoded into a text in accordance with the author’s significative schema, which includes a model of an audience’s schema; that text is then interpreted into an “understanding.”

(So much of “intent” involves the complex management of parts according to implicit values, such that I have hesitations about the uncritical use of the term: for instance, here in the text at hand, my ostensible macro-intent includes 1) kicking off a longform discourse as previously discussed, 2) mulling over questions we’ve both found compelling the past few weeks, 3) making as much progress on the asked questions as possible, 4) creating a document of our thoughts that’s relatively accessible to a public, 5) making whatever cases I attempt to make as compellingly and coherently as possible, 6) presenting myself well, etc etc. Thus the macro intentions of the piece give way, on the basis of my own accounting of the piece’s direction, need, and active questions, to the present digression. Thus some macro-intent is shaped by its contact with ground. All sentences in this digression are in the service of explaining as best as I can my misgivings with the concept of “intent”—but do I have a specific intended meaning, a semantic parsing or inferential entailment which I wish the reader to draw from this specific phrase or that? How conscious is that intent? Or else, are the parts more designed as structural elements to “round off” and “complete” my point? Surely there is more meaning to this digression than merely having completely made my point that “intent” is a problematic term. Your thoughts?)

“Against Theory” starts off promisingly: “Theory attempts to solve—or to celebrate the impossibility of solving—a set of familiar problems: the function of authorial intention, the status of literary language, the role of interpretive assumptions, and so on. […] In our view, the mistake on which all critical theory rests has been to imagine that these problems are real.” Was I too optimistic praying for a compatibilist thesis?

In addition, the pair make a claim that theory is fundamentally “the attempt to govern interpretations of particular texts by appealing to an account of interpretation in general”—reminding me of Gigerenzer’s concept of a “surrogate idol” in the sciences, this being a highly sought, so-called universal method of statistical inference which can be uniformly, automatically applied to any problem. Instead, Gigerenzer persuasively argues, there is a kind of art involved requiring a context-sensitive application of statistics; a context insensitive solution is incoherent. This seems in keeping with the idea that “meaning” is dependent on the activity and subject at hand: meaning to whom, where the _whom _is a socially and environmentally situated entity with desires, agendas, behavioral patterns, etc. That building a holistic understanding of a writing may require a different approach each time.

Unfortunately, the pair quickly, and without irony, present a universal theory of their own: meaning _is _intended meaning, full-stop. There is no distinction or difference, they are one and the same.

But of course, if I were to try to tell you that bread was _brown bread, full stop, the claim would be self-defeating. In my need to attach a qualifier to distinguish a subset of the whole (where brown bread is a subset of bread, or intended meaning a subset of meaning), I have actually, unintentionally (here is the irony!) conceded that the subset is _not the whole, that it is only a subset. Thus, in a class narrow-and-conquer method (a la Taleb’s “true rationality is what works,” or Callard’s “true courage is prudent”) do Knapp and Michaels unintentionally make the argument opposite the one they intend to. They’ve attempted to sneak in a conceptual re-engineering under the guise of a neutral analysis!

“From the standpoint of an argument against critical theory,” they continue, “the only important question about intention is whether there can in fact be intentional meanings. If our argument against theory is to succeed, the answer to this question must be no.” In this frame, meaning is only _created upon the production of a scene, i.e. in the writing phase. It is _never _produced in the reading phase. Thus the reading tracker, who stumbles upon animal tracks _unknowingly left behind by an animal incapable of communicative intent—can he properly say that the tracks “mean” something to him? Clearly there is real informational entailment, real inferences relevant to the activity at hand which can be drawn from the tracks. Clearly also, the use of “to mean,” as used in the tracking example, is fairly standard; it does not strike us as strange to say that animal tracks “mean” that an animal has passed through, or that especially large prints “mean” that the animal was also likely large. What more is this than a verbal dispute, in which one side—Michaels and Knapp—have isolated a subset of the way that English speakers use the term “meaning,” and declared it the full and “true” meaning of the text? And on what grounds are they any more authoritative, in this essentially lexicographic claim, than any other side in the debate? Nor do they ever appeal to practical reasons to prefer their carving to, say, the reader-response theorist’s carving, who says that meaning “just is” what a given reader understands, that the intent of the writer plays no part in its meaning. (Another narrow-and-conquer view here, which takes a part of the meaning process and calls it the whole of the phenomenon… useful as a corrective to intentionalism, but not a full picture on its own.) Instead, they treat it as self-evident: meaning _is _intended meaning, end of story, and if you disagree you’re confused. Well, I think the confusion lies elsewhere, as is clear to see.

Let us look also to some popular uses of “mean”: That wasn’t what I meant to say, the speaker corrects himself. Such a sentence is incoherent if the intent of an utterance is identical to the meaning of the utterance; it is definitionally impossible to mean anything other than what one ends up saying. If one says a string of gibberish due to a speech impediment, then the meaning of that gibberish is “just,” and no more, what the speaker intended, and has absolutely nothing to do with the actual words that came out. The position is ludicrous! This is one problem with narrow-and-conquer approaches: by “giving up” ground that the term still occupies in discursive space, the theoretician becomes unable to explain a whole host of real-world usages.

Knapps and Michael would further have it that the text constructions of GPT-3, or any other non-human language creation, is definitionally meaningless—indeed, they deny that they are even words to begin with. Clearly, those text constructions are treated by readers as signifying, as entailing; there is something “more” which emerges out of the words. Is it merely an illusion? The phenomenology is identical, regardless of the origin; it seems no more an illusion than the reception of any other kind of speech.

Here, we’re caught in a question as to the meaning of meaning. A few premises: one, that meaning is a process; it is used, usefully, to describe different stages and states in this process. If we pick a single point in this process, a single state in its flux, and pick this out as “the meaning,” we have already reduced a full, living thing to something dead and partial.

But K&M do get something right, though I’m most unsure of its particulars. A large portion of our interpretation is _grounded in trying to figure out what the writer intended e.g. what did the boss _mean, what did the president mean, etc. This may be truest in cases where the speaker’s internal model, which his “writing” lossily encodes or conveys, is the real subject of interest: what the boss or president means matters because it is an ostensible reflection first, of what they desire of us, and second, of their own future actions, around which we wish to adapt. A few questions arise which I want to explore (unless you think there are other, more relevant ones, or that these questions collapse in the light of a new frame): what other purposes we have, how to consider “performativity” vs. communication, how to understand and refer to the interpretive approximations of intentions that e.g. an employee has of a boss’s utterance, and finally, what to do with the information linguistic products unintentionally “testify” to (as well as the approaches which seek out these unwitting testimonies—hermeneutics of suspicion, structural claims, etc). Much of my thought is confused here, I think, because of the word “intent”—there are many levels of intent, and the way that a macro-intent touches down on the ground and plays out isn’t straightforward. Many linguistic products merely attempt to create an affective response in the receiver—is there an “intended meaning” in these utterances?

Until next time, SR