👋 Hi, welcome to Signal Lab where the team at Innate Language Processing digs deep into language & behavior, the limits of technologies, and the connection between what people say and do. 100% human authored, never with AI. For venture investors and founders.

Have you played the viral YC Partner Simulator game? You can watch 60-second Y Combinator application videos, then guess whether the startup got in to YC or not. Maybe you can pick winners just as well as Partners at the most successful startup accelerator in the world….

Of course, to do well on the simulator, it helps to know what YC looks for. For that, you could refer to ycinsight.com and its 43 databases to guide you. Or, read the 2,000 word essay “How to Apply to Y Combinator” by YC founder Paul Graham.

You don’t even have to read the whole thing. In the first sentence, Graham makes the point: “Probably the biggest thing people don’t understand about the process is the importance of expressing yourself clearly.”

Clarity is more than just word choice; it’s also expressed in syntax — choosing simpler structures over more complex ones. For instance, active voice or simple sentences (i.e., no dependent clauses), and all the little grammar decisions that follow.

If “the biggest thing” really is “expressing yourself clearly”, then could linguistic clarity alone differentiate YC acceptances from rejections? Agnostic to data like growth rate, burn rate, runway (YC classics) — even to vertical? Could you predict YC admission looking only at how clearly the pitch was communicated?

And if so, what would that mean?

PG’s blog post was written in 2009. At that time, YC reviewed 1000 applications over a 10 day time frame, i.e., 100 applications a day. Today, YC receives more than 40,000 applications each for four batches a year.

To be clear, promoting the use of computational linguistic screeners is not where this post is heading (though manual pitch review at scale is not without its own issues). We simply picked up where a TechCrunch reporter left off. She improved her accuracy by playing the game without attention to what the company was pitching, only focusing on how quickly the company could convey what they did. So we devised a test that did not consider anything but the syntactic and morphological structures used in the pitch.

If you completely ignore who the founders are and what the companies do, can you still figure out if they’ll be admitted to YC?


The Mini-Test

We built a test to examine whether applicants speaking a certain way have a greater rate of admission, without considering any other factors. We ran a model optimized for linguistic clarity (see below for some of the factors).

We used it to assess 18 pitches.

3 of those pitches were accepted to YC. Our model picked 2 of the 3. Superside (currently 850+ team) and Reducto. (600M valuation). We missed Bulletin (acquired).

15 of the companies were rejected from YC. We correctly identified 12 of them, without logging any details about the founders or the company itself. Interesting, one of our 3 “false positives” turned out to be a successful acquisition (a not-so-false positive)…. So our false positive count should really be 2.

18 pitches. 2 of 3 acceptances identified. 12/13 of 15 rejections identified.

Pretty good for ignoring what all of the companies actually do?

This level of accuracy in prediction across samples spanning a decade suggests there really is a “YC Sociolect”. What is a sociolect? It’s a specific way of speaking used by a particular social, economic, or professional group that people use to gain access to that group or to fit in. And YC has one.

So what are the passwords?

Many YC applicants make the mistake of trying to sound smart. Smart is not the same as clear. There are two main ways clarity is achieved in YC pitches, both of which contribute to that almost-casual tone that saturates Demo Day.

Adopting a sociolect increases chances of group admission (i.e., in this case, acceptance to YC).

So when Paul Graham says, “You have to be exceptionally clear and concise” he means….

  1. Action-orientation.

    1. Change nouns into verbs. It’s called ‘verbification’ (no joke) like decision —> decide or motion —> move. This is a common style in YC, dense in strong verbs.

    2. Delete adjectives. This is called ‘pruning’; makes the tone more executive and clean.

    3. State the problem first in terms of individuals and actions. This should be done in a Subject-Verb-[Object] ‘simple sentence’ structure.

    4. Short sentences. One idea in each. Direct.

  2. Don’t do marketing in any recognizable way, but still do it.

    1. Use numbers as adjectives, ‘numeral determiners’. The specificity is more credible and convincing.

    2. Use epistemic markers like “We realized people want X not Y” to neatly demonstrate problem, solution, and progress.

    3. Don’t make bald claims, show the work that gives them standing. “We called 300 customers.”

    4. Be contrarian because it’s a good hook and triggers FOMO. “Everyone thinks X, but Y.”

    5. Hedge limitations, but definitely mention them. “It’s only been six weeks, but we have 500 customers.”

If you can become fluent in “YC”, you just might get into YC without even trying that hard.

See you next month.

Reply

Avatar

or to participate