A word may have many parts, but at root it involves a language-specific association between two pieces of information:
 

A phonological form, e.g. [kalb]

<=>

A meaning, e.g.

It also seems well-established from psychological experiments, speech error data and evidence from the speech of people with brain injuries that words are organised in the lexicon in such a way that similar phonological forms are stored together, as are similar meanings. In other words, the lexicon must involve a rather tangled network of associations between phonological forms and meanings.

Word-sounds without meanings: nonsense words etc. Word-meanings without sound: tip-of-the-tongue states? Also, things without names: a sheaf of 500 pages of paper is called a ream, but there is no single word for a pile of leaves.

2. Two types of phonological representation

In fact, we saw evidence in the previous lecture that people have two different types of phonological representation: the auditory impressions of heard and imagined words, and the motor plans for speaking. Brain damage on the "speaking" side can give rise to Broca's aphasia, in which speaking is impaired, but comprehension is largely spared. Brain damage on the "hearing" side can give rise to Wernicke aphasias; e.g. jargonaphasia, in which the motor aspect of speaking is relatively unimpaired, but the form of spoken words seems to be damaged: "maccordium" for "accordion", "optokis" for "octopus", etc.

Consequently, in imaging studies we see that when a person listens to words, there is activation in primary auditory cortex, in auditory association cortex, as well as other parts of the brain that depend on what the word means. In listening to words and simply repeating them aloud, primary motor cortex and the supplementary motor area are also then employed. (A similar pattern has also been observed in an auditory lexical decision task, i.e. the subject judges whether an auditorily presented "word" is a real word or nonsense, and says "yes" or "no" accordingly.) In a verbal fluency task (thinking up words from a particular category), there is more frontal activity, and a reduced level of STG activation. Frontal activations (in or in front of Broca's area) are quite normally found in studies involving searching for words.

3. Various kinds of semantic representation

Martin et al. (1995) used PET to compare the activations associated with naming an object (the baseline task) with stating i) its colour and ii) an action associated with use of the object. Note that the linguistic task is pretty much the same each time, but the kinds of meanings (names of things vs. colours vs. actions) are different. Some areas were activated in both conditions, but the contrast between colour words and action words was as follows:

Colour words: left and right fusiform gyri (at base of and to the rear of the temporal lobes, near to the occipital lobes, where visual processing takes place);

Thinking of action words (compared to colour words) activated the middle temporal gyrus, as well as Broca's area and the right lateral cerebellum. The latter two areas are known from other studies to be involved in motor tasks.

In a study of the sites of left-hemisphere lesions in patients with specific kinds of naming impairments, Damasio et al. (1996) found that damage to different parts of the temporal lobes is related to different naming impairments:
 

Lesion site	Naming impairment
Temporal pole	Names of familiar people
Inferior temporal lobe	Names of animals
Posterior inferior temporal lobe and  anterior lateral occipital region	Names of tools

H. Damasio et al. consider these regions to be involved in mediating (i.e. associating) "conceptual knowledge" - meanings - with phonological forms of words, and consequently to be also involved in retrieval of names.

Later, Martin et al. (1996) use PET to compare the activations associated with naming tools vs. animals. Naming pictures of animals and tools was associated with activation of the lower part of both temporal lobes - as in the H. Damasio study - and Broca's area - as in verbal fluency tasks. In addition, naming animals activated the left medial occipital lobe, a region involved in the earliest stages of visual processing. In other words, subjects appeared to be evoking the appearance of the animal in naming it. (Note that they were given visual input in both the animal- and tool-naming tasks.) In contrast, naming tools activated a left premotor area also activated by imagined hand movements, and an area in the left middle temporal gyrus also activated by the generation of action words.

4. Nouns and verbs? Or things and actions?

Damasio and Tranel found a double dissociation between noun and verb retrieval:
 

Patient identifier	Lesion site	Word retrieval impairment
AN-1033	Left anterior temporal lobe	Nouns
Boswell	Left temporal pole, anterior temporal cortex + deeper structures	Nouns
KJ-1360	Left premotor cortex	Verbs

Note that KJ-1360 can describe actions, by creating neologistic denominal verbs ("going ... scissoring", "sailboating" etc.) Likewise, Boswell and AN-1033 have some nouns (e.g. horse, bird), and can identify things. Consequently, D & T consider this to be a deficit of verb and noun retrieval, rather than a deficit at describing actions and things (i.e. a non-linguistic cognitive deficit).

5. An emerging theory

The consequences of this work seem to be:

1. The sound of words and their meanings are not stored as an abstract code, but as long-term experiential memories of what words sound like, how to say them, what things look like, how to use or do them.
2. The link between sounds and meanings is mediated by distributed, associative memory. Damage to associative memory can impair access to lexical items, even if phonology and semantics as such are not as badly impaired (?).
3. The heard sounds of words and the motor plans of how to say them are both stored. The auditory form is probably the main one, because it is activated even in speech production tasks.

Damasio, A. R. and D. Tranel (1993) Nouns and verbs are retrieved with differently distributed neural systems. Proceedings of the National Academy of Sciences of the USA 90, 4957-4960.

Damasio, H., T. J. Grabowski, D. Tranel, R. D. Hichwa and A. R. Damasio (1996) A neural basis for lexical retrieval. Nature 380, 499-505.

Martin, A., J. V. Haxby, F. M. Lalonde, C. L. Wiggs and L. G. Ungerleider (1995) Discrete cortical regions associated with knowledge of color and knowledge of action. Science 270, 102-105.

Martin, A., C. L. Wiggs, L. G. Ungerleider and J. V. Haxby (1996) Neural correlates of category-specific knowledge. Nature 379, 649-652.