Do phonemes truly have meaning? Could it really be that random seeming sounds like “chah” or “faf” result in responses like fast, bright, small, angular and sophisticated? Could other sounds like “gung” or “bod” make people think of words like slow, dark, large, rounded and rugged?

This article infers that phonemes do have meaning, and what you name something can affect other’s perceptions of it. We found this very interesting, and wanted to test it for ourselves when we were naming our company.

What did we do?

We don’t just build tools for running studies, we run numerous studies ourselves as an integral part of our work.

When we were naming our company we came across the claims mentioned above wanted to find the fastest, simplest way to turn the predictions of the article into an online study. That way we could quickly confirm or deny its conclusions and make a better decision based on the claims.

How did we do this?

The study involved generating new fake words from random combinations of phonemes (drawn from two different phoneme groups) and asking participants to identify the appropriate word association.

For example, when asking what the obscure word means, the participants had two choices, coming from one of the following pairs:

  • small, large
  • angular, rounded
  • fast, slow
  • bright, dark
  • sophisticated, rugged
  • short-term, long-term
  • male, female

Here is an example question:

What do you think the obscure word *fash* means?

  • bright
  • dark

We used those results to determine if certain sounds elicit similar responses from our participants. The entire process of (1) designing the study, (2) building it for deployment, (3) recruiting participants, and (4) analyzing the data took merely 3 hours and 15 minutes of work, spread across only two days.

Study Timeline

Day 1 ≈ 1h30m of work

  • 10:13pm: Encountered claims about phonemes in article, decided to test them
  • 10:30pm: Started designing study to test claims
  • 10:53pm: Rapidly started building deployable study using our GuidedTrack system
  • 11:28pm: Completed building study using GuidedTrack
  • 11:29pm: Began participant recruitment using Positly
  • 11:31pm: 1st study participant finishes study
  • 11:38pm: 17th study participant done, data carefully checked to confirm no mistakes

Day 2 ≈ 1h45m of work

  • 8:31am: Final participant completes study (303rd study participant)
  • 9:08am: Downloaded full study data for statistical analysis
  • 9:40am: Wrote up the study results (estimated time)
  • 10:24am: Data is anonymized and code is set to “public mode” for public sharing
  • 10:29am: Study write-up posted to social media for general comments. Post includes all data (anonymized), link to experience the study from the participant perspective, and full study code. Online discussion ensues, with 8 people making comments about study details and/or its relation to other studies
  • 10:50am: Sent study results via email to the original author of claims to get author’s perspective

What did we learn?

Much to our surprise, the study confirmed most of what the article claimed as true!

We found that six out of seven of the author’s claimed phoneme group associations were in the predicted direction. The only phoneme group that did not behave as predicted was the “maleneses” category.

So, why did “maleness” go the opposite way predicted by the article?

Our working theory is that the first group of words has “f” (as in female), “h” (as in her), and “s” (as in she). We think that words in the first group starting with f, h or s caused people to predict the word was female.

Compared with the the second group has where we suspect participants selected male when it started with “m” (as in male) and “b” (as in boy).

It’s important to note that the effect sizes found in this study are fairly small. Only about 60% of answers going in the expected direction, which means about 40% of the time people did NOT have the expected association. Whether this is a strong enough effect to care for any particular application is an open question (it depends on your particular application).

Still, though, the results were highly statistically significant (meaning that the results are unlikely to be the result of sampling error). While this doesn’t rule out coding mistakes or flaws in the study design, we saw a similar outcome. As we mentioned, one category went in the opposite direction than was expected (maleness).

Why is this important?

This example shows how easy it can be to quickly inform your decision-making by conducting quick studies. When you are acting on research that may not necessarily replicate in your context this is especially useful.

This can be a game-changer for your research because using Positly, you can receive feedback on your theories or test potential outcomes in real-time. This gives you the ability to update your thinking or adjust your research based on your findings.

If you want to dive deeper into this research – or do your own replication – check out the supporting links below.

Supporting Links