Application of a Bayesian approach for exploring the impact of syllable frequency in handwritten picture naming

ABSTRACT

Adult studies in the fields of neurolinguistic and mental chronometry suggest that the syllable plays a functional role in handwritten word production. These studies support the hypothesis of a syllabified orthographic representation stored in the graphemic buffer. However, there remains the question of the cognitive mechanisms involved in this encoding of orthographic representations and, in particular, that of the processes related to the syllable. In the study reported here, we tested the hypothesis of an orthographic mental syllabary in long-term orthographic memory by exploring the impact of syllable frequency on handwritten latencies. Thirty participants handwrote the labels of one hundred and fifty images. Bayesian analyses indicated that the data support an absence of effect of syllable frequency. We propose an alternative hypothesis to the syllabary to account for the results in the literature. This respects the constraint of an absence of effect of syllable frequency in handwritten word production.

KEYWORDS:

• Syllable frequency
• reaction times
• handwritten picture naming
• Bayesian analyses

Acknowledgement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. The authors wish to thank Patrick Bonin and Sonia Kandel for the scientific exchanges on this work and two anonymous reviewers for very helpful comments on a previous version of this paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The dot refers to the syllabic boundary.

2 The grapheme corresponds to the graphic realisation of a phoneme and can be composed of several letters. Here we use these two conceptions interchangeably.

3 The data presented here were collected in 2009 during the first author's post-doctoral position at the University of Neuchâtel (Switzerland).

4 The full list of stimuli and their values for each variable are available on OSF (http://osf.io/gazf3).

5 As pointed out by Perret and Laganaro (2013), the familiarisation phase seems to have consequences for the reported effects and more particularly for Name Agreement effects.

6 There is a large body of literature on Bayesian analyses. Readers who want to know more about these analyses can consult Kruschke’s (2011) book and a recent special issue of Psychonomic Bulletin & Review (Vandekerckhove et al., 2018).

7 Posterior distributions were estimated from 3000 iterations. The first 1000 made it possible to fit the model. 4 MCMC allowed us to obtain 8000 posterior samples. Lastly, the NUTS sampler behaviour (Bürkner, 2017) was monitored with a delta value .80 for the regression model and .95 for the syllable frequency test.

8 A series of additional analyses were conducted for verification purposes. The reference model for the model comparison was created from the principal components obtained from an exploratory factorial analysis involving the eight experimental factors. We report the procedure, the R script and the results in Appendix C.

9 The question of the choice of priors seems to be eliminated. The Bayesian analysis implies choosing a prior distribution and its characteristics, i.e., prior distribution of probability. The choice of this distribution is important because it has a direct impact on the results of estimates of posterior distributions (e.g., Kass & Raftery, 1995). We therefore performed the analyses for syllable frequency with three different prior distributions: a uniform distribution (U[-100,100]), a normal distribution N[0,100] and a second normal distribution with a 10 times smaller standard deviation (N[0,10]). All three analyses gave similar results to those observed with the prior used for the analyses (N[0,100]).