And the word superiority effect supports this idea. This effect is shown in one variation on the venerable technique of priming, a stalwart of experimental cognitive psychology exploring the act of reading. Priming experiments, of one kind or another, have enabled us to demonstrate and explore mental lexicons and language pathways and to understand much of what we do understand of language and literacy management. (And see notes to chapter three) For example, we know that our minds hold morphemes as free morphemes and bound morphemes separately. (In the words ‘laughing’, ‘singer’ and ‘wilted’, for example, ‘Laugh’, ‘sing’ and ‘wilt’ are free morphemes while ‘ing’, ‘er’ and ‘ed’ are bound morphemes.) We hold these as separate items in the semantic lexicon (where we hold language as meaning) and, if we want to produce, say, ‘singer’, we produce ‘sing’ and ‘er’ from there separately and join them together (concatenate them) en route to speech. We suspect this from analysis of the sorts of error we commonly make when speaking, but also from priming experiments (and see Taft 1991).
What is a typical priming experiment? If you were a subject of such an experiment you would find yourself sitting in front of a computer. You would be shown something (typically a word, or a letter, or a letter string) on the screen. This would be the prime stimulus. You just look at it while it’s on screen, without doing anything. The screen then goes blank for a short time and then another item (typically a word, or a letter, or a letter string again) appears on the screen. This item is a target stimulus and you have been asked to make a decision about every target. Your task is to indicate, as fast and accurately as you can, your answer to a question about it - for example, whether it is a real word or not. You do this, in this example, using two pre-programmed keys on the keyboard, one for ‘yes’ and one for ‘no’. You may see anything up to about 100 prime-target pairs all told and may be one of many subjects undergoing the experiment. The speed and accuracy at which you perform are measured for each prime-target pair, by the computer. This gives a good idea how easy, or otherwise, it was to make the decision you had to make. Clever manipulation of prime-target pairs can give powerful insights into what must have been taking place inside your black box to produce such results.
As an example, let us now enjoy a word superiority priming experiment. You sit before a computer, fingers poised over the ‘yes’ and ‘no’ buttons on the keyboard. A prime stimulus appears on the screen. It is either a word, a non-word or a single letter. (It might be WORD or WORK, it might also be DOWR or KRWO. It may, though, be a single letter, D perhaps, or W). The screen goes blank, then up comes the target stimulus. It is a single letter. (It might be D or perhaps M.) You have been asked to indicate, as fast as possible, whether the letter appeared in the previous prime stimulus (as a letter or part of a word). You don’t know this at the time, but the experimenter is actually only interested in what happens when the letter does in fact appear in both target and prime. Now for the fascinating bit: On the occasions when the target letter was in prime and target, and when the prime was a real word, you will spot this faster and more reliably than when the target letter was in the prime but this prime was not a real word. Astonishingly, you will also do this faster even when the prime consisted only of the target letter! You spot, and react to, that D in the prime stimulus faster and more reliably when it is in a WORD than when it appears as a single D on the screen; when the prime-target pair was a real WORD then D you did better than when it was simply D followed by another D! In other words it is apparently easier to recognise a letter in the context of a word than the same letter when it is presented in isolation. Experimenters have tentatively concluded from this that we do, indeed, seem to read (common) words as whole words rather than letters (although, by the same token, we have obviously recognised the letters as well as the word as we spotted the D in the WORD). Rumelhart and McClelland (1986) suggested that a word is easier to recognise as we normally seek and identify whole words rather than letters, if they are easy enough, which accounts for the speed with which we do it, but that exciting the recognition of a word also at least somewhat excites the recognition of the letters of which it is composed (by onward processing?), which accounts for our recognising the letters when asked to do so. Perhaps our representation of the letter, when it is in a word, is being excited two ways as a result. The cascade feature analysis search works up and down, as we have just seen. We may search for words by feature analysis but our search also cascades back to letter level, perhaps through mandatory spreading activation. (And see Ellis and Beattie 1986, Rayner and Pollatsek 1986, Rumelhart and McClelland 1986, Taft 1991, Underwood and Batt 1996.)