Gipson (cited in Ellis 1984 p. 31) found that priming with visually presented and conventionally spelled words (eg. FOX, YACHT, COME, BLOOD) did not prime their recognition if the same words were then presented as spoken targets but that if the prime words were visually presented but spelled phonetically (eg PHOCKS, YOTT, CUM, BLUD) then subsequent auditory presentation was primed. Time taken to process the spoken word ‘FOX’ or ‘BLOOD’ was shorter when PHOCKS or BLUD had just been seen than when FOX or BLOOD had (and the target word was presented verbally, as a spoken stimulus). Gipson concluded that the reader in case one was not activating (or perhaps not strongly activating) an entry in the auditory lexicon (because he was reading to meaning purely visually) but that in case two the reader was activating phonological lexical entries (through grapheme/phoneme conversion) in order to reach meaning. In other words his subjects, under normal circumstances, were reading visually and only called phonics into play when visual reading failed. Normal reading did not require phonological analysis.

Purely orthographic structures (visual letter patterns, as opposed to their sound-correspondences) strongly affect our perceptions of words and non-words. We are visually highly sensitive to the likelihood, or ‘legality’ of letter strings. The commoner the letter string, the more powerfully it is excited when we see it. We take longer to decide that a non-word is a non-word if it contains a ‘legal’ letter pattern than if it does not - if its letter patterns are ‘illegal’. We have more trouble deciding to reject FLUNCH, or BLANTICK than we do HCLNUF or KLBNACIT, for example. We have no difficulty rapidly and accurately seeing that the last two are utterly ‘illegal’. The more common, and so plausible, the patterns in a non-word are, the more difficult it is swiftly to reject it. BLONG, for example, contains two highly legal strings, one very common indeed, and will take longer to reject than BLIOM in which one string is very uncommon (iom as in idiom). (e.g. Rubenstein et al 1970)

Very similarly, it has been shown that a letter string in a non-word which is very like a letter string in real words makes the non-word more difficult to reject. For example, we take longer to reject TRIAN (which is close to TRAIN) than we do TRUAN (which is close to nothing much). (O’Connor & Forster 1981.) Furthermore some words have a considerable ‘neighbourhood size effect’ as many, and many common, words are close ‘neighbours’ (orthographically speaking). (Coltheart et al 1977.) For example JATE will take more time to reject because it is similar to so many, and such common, words (JADE, GATE, LATE, HATE, RATE, DATE, FATE, MATE, TATE etc) whereas RALD is more easily rejected as its ‘neighbourhood’ is sparse and uncommon (only RASP & RAMP, perhaps RAND).

The pronunciation of written language, the sounds corresponding to the letters and letter patterns, does appear to have an effect on reading. It still remains unclear precisely how much effect this phonological processing exerts, at what point it exerts it, and how it does this. Some priming experiments have investigated this. For example Rubenstein et al (1971) demonstrated the pseudohomophone effect. They showed that it takes longer to reject a non-word if it is homophonic with a real word (if it makes the same sound). If some of the targets in an experiment are non-words deliberately chosen to sound like common real words (pseudohomophones) their rejection will take longer. If, for example, non-word targets included SPUNE, KREME or PHICKS then extra time would be taken deciding they are not real words compared to the time taken to make the same decision when the non-word does not sound like a real word. Clearly, letter strings were processed, phonologically, and reached meaning by this route. This secondary processing obviously interfered with the direct, visual reading, and then dismissal, of the non-word. When and how, exactly, did this interference occur, however? We know that, with real and easily recognised words we read entirely visually. Have we done this with, for example, KREME or PHICKS, finding no matching entry in the visual lexicon, only to have the ‘useful second pass’ of phonological reading arrive with the news that there is an entry in its lexicon which appears to match, and having therefore to do a re-check?

The pseudomember effect shows that the sounds of letter strings do affect reading to meaning. Suppose subjects are asked to make a simple yes/no decision on target words – eg are they members of a particular group (eg foods, or fruits) or not? The prime might, for example, say FOOD? or FRUIT? and the question is whether the upcoming target word is a member of that group or not. Some prime/target pairs and target words are unremarkable (eg FOOD? --- BREAD calls for a yes, FRUIT? --- APPLE likewise), but some targets are homophones of affirmative answers (eg FRUIT? --- PAIR or FOOD? --- MEET). These two are pseudomembers of the group. You will have guessed by now that subjects are consistently slower, and less accurate, when faced with homophonic non-members (pseudomembers) than with similarly spelled but non-homophonic non-members (eg FOOD? --- MELT, or FRUIT? --- GROPE). Once again it is clear that the phonological representation of these letter strings has been made, and taken through to meaning, automatically. The pronunciation of the letter string has clearly interfered with our rejection of it as a non-member of a group but it is not obvious how, or at what point in the chain, this interference happens. (Van Orden 1987)