The sounds have been heard and recognised as speech - as language-as-sounds. We still have no meaning, at this point, however - we have not yet understood the speech. The phonemically coded bits and pieces of language-as-sounds must be sent further, to area 5 on figure 2.1, which is the language association area. This is where language-as-meaning is stored; it is the site of the semantic lexicon. In it the phonemically coded units of language-as-sounds access and activate their counterparts representing language-as-meaning, as semantic units. At this point the language has been translated from language-as-sound in phonemic code into language-as-meaning - into semantic code. When the semantically coded representation of the language we heard has been thus activated we can say that the speech has been understood, we have reached meaning from noise.
(Philosophy intrudes at this point; have we really reached any meaning without further associating the bald dictionary meanings of the speech with the real world? Unless I somehow associate the words I have just heard (‘don’t forget your dental appointment!’) with my life they still remain meaningless, philosophically speaking. Let us leave it right there for the time being, however, and push on with cognition.)
Another small diversion: You may already be becoming a serious cognitive psychologist and wondering how speech (and, for that matter, handwriting) can be decoded and understood so accurately and easily if they are such degraded, distorted, messy and mangled signals? How do we deal with such blurred and patchy material? How do we ‘hear’, perfectly clearly, ‘the cat sat on the mat’ when what was actually said was ‘ca’ sa’ [mumble] ma’? How does anyone manage to read handwriting, especially mine? This is a practically and theoretically important question and one elegant and plausible answer is the pattern associator network. A pattern associator network is a small network of interconnected neurons, a neural net, enabling one mental representation to be translated correctly into another - phonemically coded language into semantically coded language, for example, or the picture of a rose into a memory of its scent. A pattern associator network is a plausible neuroanatomical explanation of how this may be done even when the first representation is incomplete or seriously degraded. (See notes to this chapter and Rumelhart and McClelland 1986 vol. 1 pp. 33-37.)
The neuropsychology (for such it is) of reading to meaning is much the same, in principle. Language is presented to us as text; as a collection of visual symbols or graphemes. You are looking at a mass of graphemes right now, and making meaning from them. (A grapheme is the smallest meaningful visual representation of language. It is usually a letter, of course, but might also be a sign, or even a logo (+ or - for example, also £, $, %, &, @, ♫, ♂, ♀ etc.). The eyes pass the information, the detail of all those squiggles on the page, to the visual cortex, actually at the very back of the brain, to the visual association area which is at area 3 on figure 2.1. Here the initial analysis into mental representations of symbolic language - language-as-symbol - begins. Text is recognised as linguistic symbols and this information is then passed to the symbol association area which is at area 4 on figure 2.1. Here, the symbols are associated and recognised as units of language-as-symbol. This incoming language is being represented purely as symbols, as graphemes, and is, perforce, in graphemic code. At this moment it is only a collection of graphemes represented in graphemic code. It is graphemically coded language coming in from the eyes and presently held in a mental lexicon we can call the visual input lexicon.
(As yet we do not know whether this representation is usually of letters, letter patterns, whole words or even whole phrases. Indeed, it is likely that the answer is different in different circumstances. Except when text is difficult or words are very new we probably ‘read’ much larger units than letters - whole words or at least morphemes. You read the words ‘except when text is difficult’ much easier, and perhaps differently, than you did the names ‘Rumelhart and McClelland’. This opens an interesting and often very warm debate, of which more later. Cognitive psychology has real answers.)