Computational models of reading and acquired dyslexia Flashcards
acquired dyslexia
alexia: partial or complete loss of the ability to read subsequent to a brain lesion
developmental dyslexia
lasting impairment in the acquisition of the ability to read
normal reader definition
From a cognitive point of view, reading is information-processing: transforming print to speech, and/or at the same time, print to meaning.
- Require different sets of skills
goals of science of reading
to uncover the components of this information-processing system and the pathways of communication within this system (Coltheart, 2005).
broad theoretical consensus
Despite diverging views concerning implementation (cf. Seidenberg, 2005), generally agreed that there are at least two different procedures accomplishing the transformation from print-to-speech.
- Dual mechanism
in the beginning
“We read in two ways; the new or unknown word is scanned letter after letter, but a common or familiar word is taken in at a glance, without bothering about the individual letters: its visual shape functions like an ideogram” (de Saussure, 1922).
- Divide two sets of skills
- Brain has two kinds of memory
However, it is not until the 1970s that such a conception began to make its way
“There are presumably two alternative ways in which this coding can be assigned. First, the pronunciation could be computed by application of a set of grapheme-phoneme rules, or letter-sound correspondence rules. […] Alternatively, the pronunciation may be determined by searching long-term memory for stored information about how to pronounce familiar letter sequences, obtaining the necessary information by a direct dictionary look-up, instead of rule application. Obviously, this procedure would work only for familiar words” (Forster & Chambers, 1973).
2 routes of reading - ‘dual-route model’
lexical
non-lexical
lexical route
involves looking up words in long-term memory, so as to retrieve knowledge about their meaning and pronunciation.
- Stored for every word as long as you have a memory of this word
relies on visual word recognition: once the letter string is recognized as a known word, information in memory relative to what the word means and how it should be pronounced can be retrieved for free (i.e., these do no need to be
allows direct retrieval of pronunciation (and spelling and meaning) of known words, irrespective of whether their soundform is predictable or not (e.g. CAVE = /keIv/, but HAVE = /hEv/).
for obvious reasons, it does not work for novel words, as these aren’t part of the lexicon (e.g. PLICK = /?/) derived/computed on the fly).
Because the lexical route relies on whole-word recognition, for it to work, the reader needs to know the word in the first place, so that it can be accessed and recognized by the visual system.
As this mechanism does not rely on the sound of individual letters, it can be used for irregular words, as well as regular words, but not for neologisms or unknown words (which do not have yet an entry in the reader’s mental dictionary).
nonlexical route
makes no reference to the mental lexicon (or mental dictionary), but instead involves making use of rules relating segments of orthography (how it is written) to segments of phonology (how it sounds like).
- Implicitly sampling
- How you translate a letter into a sound
- Can read any new word, just have to apply the knowledge
deriving the sound forms of the words by concatenating the individual sounds of their letters or groups of letters (i.e., the graphemes)
As such, recognizing the words is not needed to be able to pronounce it. In contrast, accessing its meaning requires to recognize the word, either visually or auditorily (i.e., once its sound-form has been built).
relies on a set of grapheme-to-phoneme correspondence (GPC) rules to derive pronunciation in an indirect fashion, e.g. RIGHT = R-IGH-T = /r-aI-t/
works with regular words (e.g. CAVE = /k-eI-v/)
but not irregular ones (e.g. HAVE = /h-eI-v/)
- Rely on vocab knowledge
- Have to use lexical route
allows an approximation of a novel word pronunciation (e.g. PLICK = /p-l-I-k/).
phoneme
smallest sound difference that reflects a difference in meaning. For instance, the sounds /s/ and /v/ are different phonemes in English for sure, because CASE and CAVE have a different meaning.
grapheme
written segment corresponding to a phoneme, and it can use one letter (making it simple; e.g., ‘c’ > /k/) or more than one letter (making it complex; e.g., ‘ch’ > /k/).
Dual-Route Cascaded (DRC) model (Coltheart et al., 1993, 2001)
see notes
The model does not decide a priori which route reads what: no traffic agent at the letter identification level!
The output is produced by the model as a whole, to get the most of it.
The two routes are not racing with one another, except when reading under time pressure.
DRC lexical
Accessing a representation in the model’s orthographic input lexicon.
From there, activating the word’s node in the phonological lexicon.
This in turn evokes the word’s phonemes at the phoneme level
The lexical route allows you to also activate a word’s meaning, from either its orthographic form or its retrieved sound form.
Hence, even nonwords (e.g. SARE), although they cannot be read correctly, yet generate some lexical and meaning activation (e.g. CARE, SORE, SANE).
- By definition, nonwords (or made-up words) like SARE do not have an entry in long-term memory.
- Nonetheless, due to their potential visual similarity, they may evoke existing words (CARE, SORE, SANE), which could lead to false recognition and a sense of familiarity.
- Also, their proximity to an exception (irregular) word may influence the way they are pronounced, against what correspondence rules would “say”.
- For example, FINT is more likely to be read as PINT (exception) than as MINT (regular pronunciation), because FINT is more visually similar to PINT than it is to MINT.
the lexical route (blue) involves recognizing the word on the page in order to access the mental lexicon/dictionary.
The latter allows the reader to consult their long-term knowledge regarding the sound form of the word (whole-word phonology) and the meaning of the word (semantics).
This route with a Y junction after letter identification is called the direct route, because it acts based on stored “whole-word” information accessed automatically rather than derived or built on the fly.
DRC nonlexical
Applies grapho-phonological conversion to convert letters into sounds
It does so in a serial fashion.
By the application of GPC rules, in this route irregular (exception) words are ‘regularized’ (e.g. HAVE = /heIv/).
does not involve the mental lexicon (i.e., it does not rely on recognizing letter strings as known visual words).
Instead, it involves a translation of individual letters or groups of letters into their respective sounds (using long-term knowledge of these print-to-speech correspondences), followed by the concatenation (or assembling) of the individual sounds, so as to build a plausible pronunciation for the word.
Once this is obtained (with activation occurring in the phoneme system), the meaning of the word can be accessed based on its whole sound-form.
the demands on the lexical v nonlexical route
vary as a function of the transparency of the language.
Need to know where to segment letter stream
Transparent - read out every sound/letter individually – doesn’t work like this in English – high % of errors – lots of exceptions that need to be stored
Hard to know which ones exceptions are
Pseudoword reading depends on transparency of language - allows assessment of knowledge about print-to-sound correspondences. As English (and to some extent, French and Portuguese) are not so clear-cut about how letters or groups of letters should be read, children learning to read in these languages take more time to master the set of correspondences. This is not the case in so-called ‘transparent’ languages, such as Dutch, Spanish or Italian, where individual letters are always read by the same sound. In these languages, children master print-to-sound correspondences much more quickly.
