WordKit: a Python Package for Orthographic and Phonological Featurization

WordKit: a Python Package for Orthographic and Phonological Featurization St´ephan Tulkens, Dominiek Sandra, Walter Daelemans CLiPS, University of Antwerp Prinsstraat 13, 2000, Antwerpen firstname.lastname@uantwerpen.be Abstract The modeling of psycholinguistic phenomena, such as word reading, with machine learning techniques requires the featurization of word stimuli into appropriate orthographic and phonological representations. Critically, the choice of features impacts the performance of machine learning algorithms, and can have important ramifications for the conclusions drawn from a model. As such, featurizing words with a variety of feature sets, without having to resort to using different tools is beneficial in terms of development cost. In this work, we present wordkit, a python package which allows users to switch between feature sets and featurizers with a uniform API, allowing for rapid prototyping. To the best of our knowledge, this is the first package which integrates a variety of orthographic and phonological featurizers in a single package. The package is fully compatible with scikit-learn, and hence can be integrated into a variety of machine learning pipelines. Furthermore, the package is modular and extensible, allowing for the future integration of a large variety of feature sets and featurizers. The package and documentation can be found at github.com/stephantul/wordkit Keywords: Psycholinguistics, Phonology, Featurization 1. Introduction Psycholinguistic models of word reading have been exten- sively investigated using computational models. Examples of such modeling efforts include the Interactive Activation (IA) model (McClelland and Rumelhart, 1981), the Triangle family of models (Seidenberg and McClelland, 1989; Harm and Seidenberg, 2004) the MROM model (Jacobs et al., 1998), the Dual Route Cascaded (DRC) model (Coltheart et al., 2001), and, more recently, the Devlex (Farkas and Li, 2002; Li and Farkaˇs, 2002) and the BLINCS models (Shook and Marian, 2013). All the implementations of these models, however different their core assumptions may be, have in common that they assume that words can be represented as numbers. In spite of this reliance on feature-based representations, the debate surrounding these models has mostly focused on the way these models process their featurized word representa- tions, not on the features themselves. The way words are featurized is usually treated as a factor which only deserves peripheral mention. This has ramifications for the way re- search progresses: first, it hampers empirical comparisons between models; while there is some consensus on which words are valid to use in simulations of psycholinguistic experiments, there is no consensus on how to featurize them. This can lead to situations in which models are deemed to be comparable because they use the same words as input, while they are actually not comparable because they fea- turize them in different ways (Plaut et al., 1996). Second, because features are not regarded as part of their respective models, there is a distinct lack of freely available featurizers, which hampers comparability and reproducibility. In this work, we attempt to remedy these issues by present- ing wordkit, a package which implements a variety of orthographic and phonological featurizers and correspond- ing feature sets. wordkit aims to be a one-stop, …

Content truncated to 500 words