Consolidating iconicity research

Readers of this blog know that I believe serendipity is a key element of fundamental research. There is something neatly paradoxical about this claim. We might like ‘key elements’ to be plannable so that we can account for them on budgets and balance sheets. But here is an element that I think can make a huge difference to the quality of our scientific work yet that is pretty much the antithesis of gantt charts, KPIs and work packages. (Then again, who has ever thought that gantt charts contributed to the quality of their work?)

A lot of my collaborations start out serendipitously. Someone mentions a cool topic in the comments on my blog. A colleague mentions a weird observation at a conference. You have some interesting data, I have a research question (or vice versa). This is how I’ve come to work on vowel-colour synaesthesia, a link between ‘r’ and rough, and communicative redoings across species. Sometimes such serendipitous collaborations turn into more durable associations. That is the case with three papers on iconicity that were published this month.

Making and breaking iconicity

My involvement in iconicity research has always been a bit cautious. The topic doesn’t play a large role in my early work on ideophones, because I saw that was where prior work had sometimes gone astray (either overclaiming or underclaiming the relevance of iconicity). When I did get into it, it was with the goal of bringing more conceptual and methodological precision (as in our teasing apart of iconicity vs systematicity, our cross-linguistic study of what sound-symbolism can and cannot do, and Gwilym Lockwood’s PhD work on individual differences in sensitivity to iconicity). In a 2017 keynote at SALC in Lund, I summarised my approach as “Making and breaking iconicity“.

To the extent that I continued work on iconicity, it was in this vein. The three studies out this month have grown out of long-term collaborations and each pick up some of these threads. Here they are:

  1. Winter, Bodo, Gary Lupyan, Lynn K. Perry, Mark Dingemanse, and Marcus Perlman. “Iconicity Ratings for 14,000+ English Words.” Behavior Research Methods, April 20, 2023. https://doi.org/10.3758/s13428-023-02112-6.
  2. McLean, Bonnie, Michael Dunn, and Mark Dingemanse. “Two Measures Are Better than One: Combining Iconicity Ratings and Guessing Experiments for a More Nuanced Picture of Iconicity in the Lexicon.” Language and Cognition, April 11, 2023, 1–24. https://doi.org/10.1017/langcog.2023.9.
  3. Van Hoey, Thomas, Arthur L. Thompson, Youngah Do, and Mark Dingemanse. “Iconicity in Ideophones: Guessing, Memorizing, and Reassessing.” Cognitive Science 47, no. 4 (2023): e13268. https://doi.org/10.1111/cogs.13268.

These papers build on prior work, replicating some of our findings in larger data sets, different languages, and complementary methods. They contribute to rounding out the picture of iconicity. And yet despite their excellent topical and methodological fit, they originated on different timelines, with teams in Birmingham, Uppsala and Hong Kong that worked entirely independently from one another.

Iconicity ratings for lots of words

This paper arises out of a very pleasant collaboration with Bodo Winter and Marcus Perlman at Birmingham along with Gary Lupyan (Wisconsin) and Lynn Perry (Miami). Noting the growing interest in iconicity ratings, we operationalised a measure that we believe captures raters’ subjective feeling of iconicity quite well, and we collected ratings for 14k English words (with at least 10 raters per word). This is currently the largest resource of carefully operationalised iconicity ratings available. You can find the ratings in the OSF repository.

There are many findings in this paper — the figure above highlights some of them. The strongest correlation of the new iconicity ratings is with humor ratings, a theoretically motivated association that we investigated in prior work using a smaller set of ratings for about 1,400 words (and imputed ratings for >60,000 words). These new ratings constitute the strongest replication yet of that finding — both the positive correlation between humor (or word-level funniness) and iconicity ratings, and the negative correlation to various measures of structural markedness which we hypothesized underpinned this link.

From rating to guessing

Another new paper out this month is with Bonnie McLean and Michael Dunn at Uppsala. This is a report of a range of experimental designs for assessing lexical iconicity. The key point is, as the title says, that ‘Two measures are better than one’: we can learn more about iconicity by using information from both guessing tasks (in which people link forms to meanings or meanings to forms) and rating tasks (in which people rate the match between forms and meanings).

Bonnie wrote a guest post about this paper so I don’t need to say too much. I do want to highlight two aspects of this work. First, the paper offers a systematic tour of the possibility space of experimental approaches to lexical iconicity. If you wonder whether to use random foils or opposite ones; whether to let people guess between the meanings of words (given a form) or between forms (given a meaning); or whether who rates words (native speaker or non-speaker) makes a difference for how ratings come out — this paper has you covered.

Second, as part of doing these experiments, Bonnie also wrote a python package icotools that makes it easy to set up rating and guessing experiments of all kinds. Software work is rarely credited in academia, but it is hugely important for reproducibility and for cumulative progress. Our hope is that this package will help more people to do experimental work on lexical iconicity — and if they do, they will be able to report the results in transparent and reproducible ways.

From guessing to memorizing and reassessing

The third new paper this month comes out of my collaboration with Youngah Do at Hong Kong University, with whom I co-direct a GRF project on iconicity. Postdocs Thomas van Hoey and Arthur Thompson have carried out a number of experimental studies of cross-linguistic iconicity. The backbone of this paper is a replication effort of two of our previous studies: a cross-linguistic one where we test the guessability of ideophones from 5 languages, and one on Japanese where we test the learning (or memorability) of ideophones.

Here as above, key results do indeed replicate. One prior study tested the guessability of ideophones from Japanese, Korean, Semai, Siwu and Ewe. This new paper adds another West-African language (Igbo) and replicates the work on Japanese and Korean. And where the prior study had speakers of Dutch doing the guessing, here we have speakers of Cantonese, showing that the result generalizes (we didn’t think Dutch speakers would have a unique aptitude for guessing ideophones, but it’s always nice to actually see that something works the same for speakers of a different language halfway around the globe).

So ideophones may be somewhat easier to guess, but does that also make them easier to memorise? In this paper, we find that people are pretty good at learning ideophones and adjectives (about as good, in fact, for the forms tested here), even if you swap the meanings to opposite foils. In this respect the findings differ from my earlier work on Japanese ideophones — a partial non-replication. An important difference though is that in that prior study we used the same preselection filter for both ideophones and adjectives, while here that filter is only used for adjectives, stacking the deck against finding a difference. Can we say more about this? We introduce a fun new analysis to dig into this question that looks at “flip-flopping”, or reassessing to use a more respectable term. For form-meaning pairs with ‘wrong’ foils that they encountered, it turns out that people are more willing to revise their mapping in the right direction for ideophones than for adjectives. More about this paper on Thomas’ blog.

In closing

I started out by drawing attention to the role of serendipity in fundamental research. The coming together of these three papers is also an example of that. It almost looks like it’s planned: a bouquet of complementary methods, with findings that establish the reliability and replicability of some of the first studies that brought together ideophones and iconicity. And yet the trajectories of these three papers have been entirely independent, which makes it all the more remarkable that they’ve come out within two weeks of one another.

So, here’s to serendipity!

Leave a Reply

Your email address will not be published. Required fields are marked *