Text, Subtext, Signal

Every communication provides three kinds of information:

• The text is the message itself. (What was said?)

• The subtext is the received meaning of the message. (Why or how was it said?1)

• The signal is the message considered as an action, or as a data point in a larger trend. (When/where was it said? Who said it? Was it said at all?)

Of these three, signal is generally the most overlooked—and, broadly speaking, the most important. Let’s look at a few examples to understand why.

First, consider a legendary episode in recent history: Nancy Pelosi’s response on Morning Joe when asked about President Biden’s 2024 reelection campaign:

The text of Pelosi’s response is on its face supportive of Joe Biden.

The subtext of the message is more mixed, but also on its face also leans supportive (of “whatever he decides to do”).

To grasp what’s really going on, you need to look at the signal of Pelosi’s message, which can only be understood by zooming out. The broader context here is: 1) Pelosi is among the most powerful Democrats in the country; 2) She has been explicitly asked whether she supports Biden’s reelection campaign; and 3) Biden has been repeatedly making statements like this:

The message (text, subtext, and signal) of Biden’s Tweet is not subtle: He intends to run. In this context, Pelosi’s refusal to accept Biden’s decision as settled sends a strong signal that she wants him to drop out—regardless of the text and subtext of her statement.

Second, consider a hypothetical recommendation letter that begins like this:

I have known Alice for approximately three years at Acme Co. Although we do not work closely together, she always brings a warm, welcoming presence to the office.

Again, the text of this message is positive (Alice is “warm” and “welcoming”), and the subtext is on its face positive as well, if a bit qualified.

But the signal, from the perspective of a hiring manager, is not good at all. Given the freedom to choose any recommender, the best Alice could come up with was someone who seems to not really know her, and who leads with vague praise for her “presence.” (The simplest explanation for this is that Alice was unable to obtain a truly strong reference; alternatively, she could have poor judgment in who to ask for a recommendation, or have gotten unlucky for any number of plausible-but-not-very-likely reasons.) Whatever the true cause, the signal is a yellow flag.2

As a third and final example, compare these two emails:

The obvious contrast between these two messages is in the subtext: The first email reads cheery and deferential while the second one is direct to the point of coldness.3

In this example, though, notice how the real signal depends much more on “who” and “when” factors than can be deduced from subtext alone. If either email is sent to you by the head of HR after several reminders have gone out, the tones may vary between the two (passive aggressive vs. transparently annoyed), but the message is essentially the same: Your tardiness is a problem. The messenger’s typical communication style has additional bearing on the signal. (When someone who normally sends the email on the left sends the email on the right, you know they’re really annoyed. Other people ~always send emails like the right one, so the signal from them is less strong.) Or consider how, if the two emails are sent sequentially, the signal changes depending on the order. (Left then right tells a story of growing frustration. Right then left tells a more subtle story—perhaps an implicit apology for being too brusque.)

The point here is that you can’t really know what signal is being sent without accurately parsing the subtext; and to accurately parse the subtext, you need sufficient context, repeated exposure to the messenger’s communications, or other kinds of data.

Of course, there may be conceptual blur between text, subtext, and signal at times. But whatever you call it, the signal frame offers several unique attributes not available within the other two modes of analysis. Specifically:

1. In a lot of situations, signal is ~all that matters. Many communications serve a social or administrative function such that it’s much more important that you said something than what you said or how you said it. If you want to make amends, a sincere “I’m sorry” is usually much more important than the precise explanation of why. If your friend loses a loved one, it’s a condolence text that shows you care much more than the perfect expression of sympathy. And if you want to hire or fire someone, it’s the material change in their employment status which sends the loudest message. (Of course words may heighten or soften the news!)

2. There can be strong signal in silence/absence. If you publicly decline to endorse, you’re assumed to be in opposition. If you never respond to texts or calls, you’re assumed to be less-than-fully-invested in a relationship (absent another explanation). And if you’re only focused on (sub)text, you’ll fail to see the text that was never sent at all.

3. Signal can often be understood through paraphrase or synopsis. In fact, for signal, it’s often clarifying to ignore the exact wording and just focus on the question, “At its core, what did the message accomplish?” (Nancy Pelosi’s intentions become much more obvious when considered in broad strokes.) By contrast, subtext and of course text are sensitive to precise phrasing and syntax.

4. A “mixed” signal is usually just a conflict between signal and (sub)text; trust the signal. E.g., If he says he’s not into you but won’t stop texting, take the rejection with a grain of salt. If someone appears on a widely streamed podcast, assume the host supports them/finds their ideas worth discussing. (Overcoming this signal requires strong evidence of opposition—not just a couple challenging questions.4)

5. Signal is more often unconscious than (sub)text is. Strong communicators will make an effort to monitor and manage the signals they are sending, but these can be harder to control than text or subtext. As moods and relationship dynamics shift, our communications ebb and flow organically, and these changes are detectable—again, often unconsciously—by our interlocutors.5 (These changes could be, e.g., fluctuations in the statistical regularity of emojis, or in “I” statements, or any other feature of language.) As a collective, society also signals interest in certain topics through the frequency and intensity of its attention. These trends are also “unconscious,” in the sense that they are an emergent property of culture not controlled by any one individual.

6. The signal’s-eye view is Bayesian. This means you can detect signal by asking: What is the likelihood I would have received this message conditional on X being true vs. Y being true? (So e.g., for Pelosi: “What’s the likelihood she would have made this statement if she was in fact supportive of Biden running?” And for Alice’s recommendation letter: “What’s the likelihood I would have gotten this letter of recommendation if Alice was in fact an outstanding candidate?”) Even when the message is not the strongest possible signal that could be sent (i.e., for the above, an explicit negative endorsement), it can still provide a clear update in one direction or another.

7. Text : subtext : signal :: f(x) : f'(x) : f''(x). To unpack this analogy a bit: You can learn a lot more about where a car will end up over time by looking at its acceleration than at its current speed; and you can learn a lot more by looking at its speed than at its current location. By the same token, just knowing what a message says (text) is less useful than knowing what it means (subtext), which is in turn less useful than knowing how meaning accrues and changes over time (the signal). With each level of abstraction comes more analytical power.6

Whenever I write a post related to communication, it’s a bit daunting to find the perfect language to drive home my key point. Fortunately, the key point in this instance is that placement and timing speak much louder than content! When it comes to signal, the last word is the strongest.7