Crispy Chicken,

Reading your recent letter on context and social games, I was struck by its line that communication is “the coordination of hidden representations in the brain.” It helped me finally (retroactively) understand what you mean when you write in 1.2, “I must point at the common grounding points so we know on which axes we can rotate our perspectives in order to keep them interlocked.” A speaker may open up a garden of forking paths, but it is on the interlocutor to pick a path and walk it. (“But surely I’m not the only one who, in their fervor to respond to a previous point, slips into single-mindedly searching for a segue into the thought that I would like to share?”1) This is perhaps best understood as a decentralized strategy or routine for establishing common ground and priority without needing explicit hashing out. This discovered commons is a strategy of mutual utility maximization (the outcome or product of the interaction). This may instantiate as optimizing for the interestingness of a conversation (e.g. in cocktail chit-chat) or determining organizational priorities in the workplace (e.g. at the beginning of a meeting, two founders determining triage or “agenda” of the coming conversation).

Now, if we replace the concepts of meaning-as-semantics with meaning-as-coordination—as I take you to advocate—what are we left with regards to one of our initial questions, that of “literal” meanings? We began to discuss this on 9/11, but not sufficiently, I think. What does it mean to “coordinate hidden representations in the brain” that does not entail a kind of semantic meaning? Where does the heart of your issue lie with a “function as ends, semantics for means” frame (that you say is somewhat well-represented already in the field, i.e. what is left to be discovered?). What do we do with this “literal” level, if we place meaning not in semantics but in the complex matrix of goals and effects a speaker may wish to bring about, the transformation of environment and futures which he attempts with words? I know you’re hesitant to lock us into these words at all, but I guess I wanna hear the status on them before we switch to new ones. The discovery part of Chestertonian fences.

Next, you write that my chosen examples of meaning, such as the “understanding your boss’s instructions” case, are special cases—those in which it is advantageous to know what the boss intends, what he is “really”2 asking for, either in order to fulfill it or else tactically provide the appearance of doing so (or choosing deliberately not to, etc…). But I’d like to ask: how many cases are there in which the additional information—speaker intent—is not advantageous? These, rather than the reverse, seem like the special case; they are what Sarah calls cases of “strategic ignorance,” in which genuinely not knowing can provide plausible deniability. (This is very similar to Bob Trivers’s ideas about self-deception as a strategy for deceiving others convincingly.) In almost every other situation, it seems that more information cannot hurt. Consider also the idea from joint attention science (see also Tommasello) that the ability to gauge interlocutor intent (e.g. by noticing where attention is directed, and the synergy between attention and desire/intentionality) develops earlier, and is perhaps a prerequisite to, linguistic communication.

But perhaps you would answer that either intent is not primary in guiding our understanding and response, or that only so much information can be processed, and intent is often lower priority even in situations where it could not (given sufficient bandwidth) hurt. This is somewhat how I take your examples of Marsha’s education system reforms, or the waiter at the restaurant. Grokking the semantic content of an interlocutor’s utterance is not always important to properly (i.e. in a sustainably self-advancing way) handling an interaction. Rather, it is one instrument for optimizing action, a useful but far from necessary instrument towards “getting what we want out of life.” Fair enough—then meaning is not equivalent to speaker intent, which is perhaps a rebuttal against the likes of Knapp and Michaels, who make intent an absolute equivalent with intentionality.3

But I think the real heart of your point is that understanding the intended semantic content (what is traditionally conceived of literary and linguistic “meaning”) is rarely an ends in itself.4 Rather, the semantic content is a way of getting at an understanding the interlocutor’s larger intent and values and strengths in order to then coordinate. (Steven Gross, Mind & Language 2010: “primate intelligence, conceived in competition, was revamped by cooperation.”) In this picture, “understanding” speech is “the planning stage mostly directed at considering the other person’s actions.” Public information acts as proxy for private information, which is to say, active intent—not semantic intent, but telos, functional intent.

This shift squares away as well with my voiced frustrations about how hard it is to hammer out semantic intent, since the sense-making is produced fluently and largely subconsciously. Asking an author what ze “meant” by a sentence or line is infamously unproductive—but more macro goals (which often can be deduced “from the text itself,” shout-out to the New Critics, if perhaps requiring, in their conjunction, existing conventions to leverage5)

Notice the elegant connection to signaling theory: the evaluating or interpreting agent uses external clues to gauge the “internal” and “invisible”—an internality which also happens to be the common denominator (best predictor—!) of the interpreted agent’s behavior. Mutual gauging of these qualities (values, aims, attitudes—I’m sounding much like Knapp & Michales!). A lot of this internal algorithm is hidden even to the host agent itself (see Trivers on the evolutionary advantages of self-deception). But that doesn’t mean it’s necessarily invisible to outsiders—just that it’s revealed, through actions and signals. This high-level concern with intent is perhaps why we are so lost when we do not know why we are being told something. “Why are you telling me this? What do you want from me? What should I say?” Slippages that reveal the choreography of desires, expectations, and actions that underly communication.

Similar to this establishing of commons, a sense of roles (e.g. status and power dynamics) enables coordination (decisional deferences, debts, appropriate divisions of labor e.g. who will go out of the way, whose desires take priority, whose well-being is prioritized). A good portion of talk is, I think, actively negotiating these roles—and, in general, actively establishing the “what is” premises in order to coordinate futures. An accounting, the mutual confirmation of relevant premises, that ground coordination. Is this what you mean, or am I putting words into your mouth?


I’ll talk briefly about the “dressing for a social occasion” example, to build another possible path in our garden. You can choose to take it or not; it’ll always be here for back-reference.6

What I like about the dressing up example, as you say, is that it is primarily about communicating through curation. It is not about spontaneous generation but the selection of existing options. If we want to post a conscious and unconscious mind, the conscious mind is “served up” possible utterances, and (when at all self-conscious) will “screen” the options which are served up, choosing to pass on certain possibilities (“paths”) and deciding to commit to others, perhaps with edits, additions, subtractions, etc. What’s more, the unconscious mind may be generating possible utterances, much like a neural network, w/r/t the training data of inputs, it carries a bias toward recent experience. This, I think, is part of why and how people can quickly “take on” a group voice and begin communicating in the language and tone of those around them.

(It’s also worth pointing out that the extent of screening has to be talked about as a spectrum of options, rather than as there being a default human behavior. In states or societies or settings with high penalties for wrongthink, revealed by wrongtalk, these is more self-censoring obviously, and in those arenas without much penalty, there is much less self-censoring. This is why first encounters, or business encounters, are more self-conscious than those with longtime friends. We can only describe this spectrum or “axis” of behavioral choice, rather than “characterizing” speech as “spontaneous and free” or else “heavily constrained”—that would be a mistake, though one I’m sure at least a few thinkers have made. Merleau-Ponty writes, “The orator does not think before speaking, nor even while speaking; his speech is his thought.”7 True enough for some. The French love their sweeping universals, their territorial over-grabs.)


So there is “macro-intent” and “micro-intent,” the “matrix” of subordinate goals and functions, but like you say, few linguists would dispute this, preferring to say that there is an intended function that is accomplished by the tool that is a literal semantic meaning. The other stance is that there is intention in the final sense of a goal, and intention in the sense of strategy: the implementation which reaches that goal. There is intention in the sense of the particular meaning, the particular “sense,” that a string of words evoke, in order to create those final effects. The intention of word-by-word curation is rarely known even to the speaker; we cannot presume it. But the larger purpose of zir speech, and the occasional strategically crucial decisions on which the purpose rest (wordchoice, delivery…) is capable of interpretation and knowing, even if it is sometimes obscured. You raised this issue in our last conversation, but I want to really get to the heart of what your objection to this “semantic means, functional ends” frame is.

In general it seems valuable to ground these ideas in the actual dynamics and blow-by-blows of interaction. How else can we understand whether they’re right or wrong? Intuitive resonance is a good heuristic and a bad analysis.

So I agree, it’s likely best that we continue bringing up case studies, and using those to build theory from, rather than getting abstract too fast. Here’s a quick case study of gossip, which you brought up in one of your letters. It’s not real speech, but a generic amalgam produced from my training outside middle school lockers (and watching Clueless and Mean Girls):

Evan: Oh my god, you won’t believe it. Lizzie and Grant have been hooking up at break.

Lucy: No fucking way! Get out of here. Shut up.

Here’s my read: Evan has not just presented information. She has sold it as valuable, as interesting (which is to say, high KL divergence, which is quite literally the definition of informational value to a predictive agent). And Lucy has acknowledged that value through the ironic “get out of here, shut up” which (ironically) means its opposite: keep going, tell me more. She’s agreed to pay; an economic transaction has just taken place. By requesting more information, she has made clear that Evan is giving her something of value.

If Lucy is higher status than Evan, she may not need to reciprocate; her (public) attention (displayed preference conferring status) on its own is a worthy exchange for the intel. But if they are peers, there is some real sense in which Lucy now “owes” Evan. This helps account for why individuals sometimes act at (honestly or dishonestly) being bored by information given to them: if it isn’t new, or valuable to them, then they don’t “owe” their interlocutor, their interlocutor hasn’t done them a favor or contributed value. Nor do they wish to receive more of this type (or level) of intel again—the gift is “valueless” to her, and gratitude should not be expected. This reflects the classical “conversation as information exchange” idea that sometimes gets tossed around in linguistics, with the important caveats that 1) there are many types of value that are optimized for, and information exchange is merely one of those types, and 2) the power dynamics play a large role.

If you’re game I’d like to explore more Kulback-Leibler divergence in future letters, and in general how “interestingness” corresponds with informational value, is high information in the sense of updating the Bayesian-approximate schema as much as possible. Murray Davis’s “That’s Interesting!” is good skimming if you have time, but a Simon DeDeo lecture should do the trick to. Think standard laboratory stuff about how gaze drifts toward high-surprisal changes in the visual field.

Lastly, I think it’s important we take seriously feelings as analysts—not in the touchy-feely way but in way that takes serious what folks in complexity studies + ML know well—that as math gets complicated and emergent, it stops looking like math. You write, “We have very little means for saying why certain things were implied, so we simply say we felt them to be.” Very well, but/and the feeling is a signal of important dynamics—the dopaminergic trackers of the reward functions, of shifting forecasts driven by new information, of the active inference that underlies a social interaction.

I’ll wrap up for now. I haven’t said much, in fact, almost nothing at all, and yet somehow seven pages are up. Last thought for now: I like the carving of “pigeon projection,” but I’m unsure about the name. Something about the pigeonholing and projection metaphors seem to really clash. The basic dynamic is a kind of parsimony, right? Easier to preserve the schema than completely update it. Parsimonious pigeonholing? Schema-preservation? Until next time.



  1. That there is a balance between listening to one’s interlocutor and listening to one’s own voice, that parts of conversations are tuned out, that some sentences are merely desperate attempts to stay afloat, efforts to minimize possible errors and embarrassing blunders, to come off as basically socially competent. 

  2. Though we cannot forget John Wentworth’s recent words (“Words & Implications,” LW) which illustrate the extent of inferential work performed in fielding everyday requests: “In the information/knowledge economy, a key part of most jobs is realizing that what someone literally asks for is only very loosely correlated with what they actually want. […] Down the hall, a manager asks an analyst for the click-through rate on the checkout screen. What the manager actually wants to know is whether lowering prices would lead to more sales. Whether that click-through rate is a good proxy for customers’ price sensitivity is the sort of question the analyst needs to answer, which means the analyst needs to figure out that that’s the real question in the first place.” 

  3. 1982’s “Against Theory” picks up where E. D. Hirsch left off with his argument that literary meaning “just is” intended meaning—that all hermeneutics “must stress a reconstruction of the author’s aims and attitudes in order to evolve guides and norms for construing the meaning of his text.” An interesting factoring of intent—as “aims and attitudes.” Side note: to be fair to the New Critics, their stance may have been as much a pragmatic attempt to delineate a tractable and knowable of scholarly analysis more than it ever was a conceptual-analytic or ontological argument. That’s the charitable view anyway—the other take is “rigor cargocult.” 

  4. Legal text may be an exception, but only because the concept of some formal semantic meaning, and the hypothetical interpretation by a judge, somewhat overlap. 

  5. I think it’s fair to say—but challenge if you disagree—that we exist in a landscape of meanings which we can leverage but cannot create from scratch or alter. For instance, a fashion statement will be interpreted by strangers in a crowd as it signifies to them based on past experience in the existing signification schema, and _not _as the wearer wishes it to.) Of course, the way one “wears” the look—which is to say, “mixes” parts into amalgams, it is possible to create emergent feelings, or change the isolated meanings of an item, but the parts have an “objective” life of their own outside intent; they are “pre-fab.” This is the case of all “signs,” including linguistic ones. Since the connection between sign and referent is historical more than “innate” or formal, the referent cannot be deduced from the sign itself. 

  6. So much of understanding is retroactive, yeah? Things that come after narrow down considered options—constrain interpretations, make sense of what came before. 

  7. Elsewhere: “One does not know what one is saying, one only knows after one has said it.”