Non-parametric Algorithm to Isolate Chunks in Response Sequences

Abstract

Chunking consists in grouping items of a sequence into small clusters, named chunks, with the assumed goal of lessening working memory load. Despite extensive research, the current methods used to detect chunks, and to identify different chunking strategies, remain discordant and difficult to implement. Here, we propose a simple and reliable method to identify chunks in a sequence and to determine their stability across blocks. This algorithm is based on a ranking method and its major novelty is that it provides concomitantly both the features of individual chunk in a given sequence, and an overall index that quantifies the chunking pattern consistency across sequences. The analysis of simulated data confirmed the validity of our method in different conditions of noise, chunk lengths and chunk numbers; moreover, we found that this algorithm was particularly efficient in the noise range observed in real data, provided that at least 4 sequence repetitions were included in each experimental block. Furthermore, we applied this algorithm to actual reaction time series gathered from 3 published experiments and were able to confirm the findings obtained in the original reports. In conclusion, this novel algorithm is easy to implement, is robust to outliers and provides concurrent and reliable estimation of chunk position and chunking dynamics, making it useful to study both sequence-specific and general chunking effects. The algorithm is available at: https://github.com/artipago/Non-parametric-algorithm-to-isolate-chunks-in-response-sequences.

Differences in chunking behavior between young and older adults diminish with extended practice

“…Previous research has proposed multiple analyses to analyze, sometimes rather specific, aspects of chunking. The methods used include t tests (Bo et al, 2009;Ruitenberg et al, 2014), k-means clustering (Song & Cohen 2014), dynamic network analyses (Wymbs, Bassett, Mucha, Porter, & Grafton, 2012), hidden Markov models (Acuna et al, 2014) and non-parametric rank-order algorithms (Alamia, Solopchuk, Olivier, & Zenon, 2016). An elaborate discussion of each of these methods is beyond the scope of the current work.…”

Motor sequence learning in older adults

“…Previous research has proposed multiple analyses to analyze, sometimes rather specific, aspects of chunking. The methods used include t-tests , k-means clustering (Song & Cohen, 2014), dynamic network analyses (Wymbs et al, 2012), hidden Markov models (Acuna et al, 2014) and non-parametric rank-order algorithms (Alamia, Solopchuk, Olivier, & Zenon, 2016). An elaborate discussion of each of these methods is beyond the scope of the current work.…”

“…Many approaches have been explored and many different aspects of chunking have been measured. Researchers have experimented with different analyses including simple t-tests , k-means clustering (Song & Cohen, 2014), dynamic network analysis (Wymbs et al, 2012) and even a hidden Markov model (Acuna et al, 2014) and a non-parametric rank-order algorithm (Alamia et al, 2016). It is important to realize that that while all these methods refer to 'chunking', they are developed to measure specific aspects of chunking.…”