Effects of complexity and unpredictability on reading acquisition: Behavioural and computational studies[Bearbeiten]
European orthographies differ on the extent to which print-to-speech correspondences are complex and/or predictable (Schmalz, Marinus, Coltheart, & Castles, 2015). Complex orthographies are characterised by many multi-letter rules, such as sch → /ʃ/ in German. Predictability refers to the sufficiency of the print-to-speech correspondences for achieving high accuracy in reading (e.g., the English words “yacht” or “colonel”, are unpredictable, because whole-word knowledge is required to give a correct pronunciation).
In a behavioural and two computational studies, I aim to assess to what extent complexity and unpredictability pose independent challenges for the cognitive system of a learning reader. In a behavioural learning experiment, adult participants will learn a made-up orthography, where we will manipulate the complexity and unpredictability of print-to-speech correspondences. In a first computational study, we will implement a connectionist model to simulate the results of the experiment. In a second computational study, we will extend the connectionist model to real European orthographies, to derive specific predictions about how complexity and unpredictability should affect reading acquisition in children.
Previous research has already implemented connectionist models for learning to read (e.g., Plaut, 1999), and these are widely used to test predictions about reading behaviour (e.g., Graves et al., 2014; Jared, 2002). However, none of the implemented versions are openly available. The current project will make a contribution to Open Science, because we aim to make the model (including the source code) freely available.
Beschreibung des Vorhabens: Problemstellung, Methoden, Herangehensweisen[Bearbeiten]
The ease with which children learn depends on the “depth” of their orthography (Seymour et al., 2003). Deep orthographies have many inconsistent grapheme-phoneme correspondences (GPCs). English is deeper than German: For example, in English, a has different pronunciations in the words “ball”, “park”, “bank”.
We distinguished between two sources of orthographic depth: complexity and unpredictability (Schmalz et al., 2015). For example, in Italian and in English, the grapheme g can be pronounced as /g/ or /dʒ/. In Italian, its pronunciation can be described by two GPCs: g → /g/ (“gusto”), and g[e,i] ¬→ /dʒ/ (“giorno”). These rules are complex, because more than one letter is involved in determining the pronunciation, and predictable, because applying them will always lead to the correct pronunciation. In English, g may have different pronunciations in near-identical contexts (e.g., gift → /gɪft/ and gist →/dʒɪst/). Thus, GPC knowledge is not sufficient for a correct pronunciation, and knowledge about the whole word is required.
Complexity and unpredictability are dissociable on a linguistic level (Schmalz et al., 2015), but it is an open empirical question how they affect reading acquisition. Answering this question will partly explain why learning to read is more difficult in some languages. In the long run, this will allow for optimisation of teaching practices and effective treatments of developmental dyslexia across languages.
I will conduct three studies:
1) In an experimental study, adult participants will learn an artificial orthography. Participants learn pseudowords (e.g., /faɪp/), written in unfamiliar symbols (e.g., extinct Hungarian runes). We can include GPCs which are complex (e.g,. [/k/]S1 → /a:/, [/t/]S1 → /o:/). Another symbol (S2) can have two pronunciations which are not predictable from the context. We can then explore the learning rate of complex and unpredictable GPCs, and the system’s behaviour in response to unpredictability. Pilot data and a detailed report here: https://osf.io/z8d72/.
2) In a computational study, I aim to simulate the results of the first study with a connectionist model. Connectionist models learn the print-to-speech correspondences through exposure to a set of words (Plaut, 1999). The main aim will be to simulate the effects of complexity on the learning rate, and the behaviour in response to unpredictable GPCs.
3) In a second computational study, the model will be extended to real alphabetic orthographies. This will allow us to make specific predictions about how complexity and unpredictability will affect learning to read in a real-world setting.
- Name: Xenia Schmalz
- Kontakt: firstname.lastname@example.org