Figure 133

The WEAVER++ computational model. Adapted from Levelt et al. (1999a).

5. Phonetic encoding: speech sounds are prepared, making used a syllabary, "a repository of gestural scores for the frequently used syllables of the language" (Levelt et al., 1999a, p. 5).

6. Articulation: the actual production of the word by the speech musculature.

In addition, there is a self-monitoring process which monitors the speaker's internal speech. It is easy to get lost in the complexities of this theory. However, it is mainly designed to show how word production proceeds from meaning (lexical concepts and lemmas) to sound (phonological words, phonetic gestural scores, and sound waves). Indeed, Levelt et al. (1999a, p. 2) referred to "the major rift" between a word's meaning and its sound, and argued that crossing this rift is of major importance in speech production. In general terms, the stages of conceptual preparation and lexical selection involve deciding which word is to be produced, and the later stages involve working out the details of its word form, phonological representation, and pronunciation.

Experimental evidence

Lexicalisation is "the process in speech production whereby we turn the thoughts underlying words into sounds: we translate a semantic representation (the meaning) of a content word into its phonological representation or form (its sound)" (Harley, 1995, p. 253). According to Levelt et al. (1999a), lexicalisation is an important process that occurs when the lemma is translated into its word form in terms of morphemes, phonological encoding, and so on.

The "tip-of-the-tongue" state provides evidence generally supportive of the views of Levelt et al. (1999a). We have all had the experience of having a concept or idea in mind, but searching in vain for the right word to describe it. This frustrating situation defines the tip-of-the-tongue state. It was first studied systematically by Brown and McNeill (1966). They stated that a participant in this state, "would appear to be in a mild torment, something like the brink of a sneeze" (Brown & McNeill, 1966, p. 325). Brown and McNeill presented their participants with dictionary definitions of rare words, and asked them to identify the words defined. Thus, for example, "a navigational instrument used in measuring angular distances, especially the altitude of the sun, moon and stars at sea" defines the word "sextant". The tip-of-the-tongue state occurs when the lemma or abstract word has has been activated, but the actual word cannot be accessed. However, this conclusion has been disputed (Caramazza & Miozzo, 1998).

It has been suggested that the tip-of-the-tongue state is more likely to occur for words sounding like other words, on the basis that these other words block or inhibit retrieval of the word being sought. In fact, Harley and Brown (1998) found exactly the opposite. Words sounding unlike nearly all other words (e.g., apron; nectar; vineyard) were much more prone to the tip-of-the-tongue state than were words sounding like several other words (e.g., litter; riddle; pawn). The unusual phonological forms of the former words may make them especially difficult to retrieve.

Levelt et al. (1999a) argued that abstract word or lemma selection is completed before phonological information about the word is accessed. This is an important part of their serial processing model. In contrast, some theorists (e.g., Dell et al., 1997) have argued that phonological processing can start before word selection is completed; in other words, the two stages are not totally independent of each other. Theoretical approaches based on this assumption are often termed cascade models.

How can we test these models? According to the cascade model, phonological processing can occur before lemma selection is completed, whereas this is impossible on the alternative model. Suppose that participants are presented with pictures having a dominant name (e.g., rocket) and a non-dominant name (e.g., missile). They are given the task of naming the picture as rapidly as possible. The stage of lemma selection typically produces the dominant name (e.g., rocket). According to the serial processing model, there should be very little phonological processing of the non-dominant name (e.g., missile). According to the cascade model, however, this is not necessarily the case.

Peterson and Savoy (1998) tested these predictions. On some trials, a word appeared in the middle of the picture (e.g., of a rocket or missile), and the participants were to name that word out loud rather than name the picture. As both models would predict, word naming was speeded up or primed when the word was phonologically related to the dominant picture name (e.g., racket). Of more theoretical importance, word naming was also speeded up when the word was phonologically related to the non-dominant picture name (e.g., muscle). As Peterson and Savoy (1998, p. 552) concluded, "We obtained clear evidence for phonological activation of both dominant and secondary picture names during early moments of picture lexicalisation. Thus, in contrast to the serial model's central claim, it appears that multiple lexical candidates do undergo phonological encoding .we reject the serial processing view and argue, instead, that the cascade model provides the best account."

How did Levelt et al. (1999a) respond to these findings? They pointed out that multiple phonological encodings have only been found for synonyms, and suggested that the findings of Peterson and Savoy

(1998) represented only a minor embarrassment for their model.

Studies on brain-damaged patients support some of the theoretical assumptions of Levelt et al. (1999a). According to their theory, the lemma contains syntactic information. Brain-damaged patients who can access the relevant lemma but who cannot carry out the subsequent stages of morphological and phonological encoding should possess syntactic information about words they cannot produce. Badecker, Miozzo, and Zanuttini (1995) studied an Italian patient with anomia (an inability to name objects). This patient found it virtually impossible to name pictures, but he was almost perfect at deciding whether the correct word was masculine or feminine (a syntactic feature).

According to Levelt et al. (1999a), morphological encoding precedes phonological encoding. As a result, some brain-damaged patients might possess morphological information about a word (e.g., whether it is a compound word) without being able to gain access to its phonological form. Semenza, Luzzati, and Mondini

(1999) discussed studies on aphasic patients who exhibited precisely this pattern of impairment.

Levelt et al. (1998) reported an ambitious magneto-encephalography (MEG) study designed to identify those areas of the brain most active during the successive processing stages involved in naming pictures. The occipital lobes were most active during visual-to-concept mapping, the occipital and parietal areas were most active during lemma selection, there was left-hemisphere activity in Wernicke's area during phonological encoding, and areas in the sensory-motor cortex were especially active during phonetic encoding. These findings provide general support for the serial processing model of Levelt et al. (1999a).


WEAVER++ has some significant advantages over the theoretical approach of Dell (1986) and Dell et al. (1997). First, it makes detailed predictions about the speed with which words are produced in different situations, whereas Dell has focused on predicting error rates. Second, WEAVER++ with its emphasis on serial processing can be regarded as simpler than Dell's approach based on highly interactive processing. Third, Levelt et al. (1999a) relied much less than Dell (e.g., 1986) on data about speech errors. As Levelt et al. (1991, p. 615) pointed out, "an exclusively error-based approach to.speech production is as ill-conceived as an exclusively illusion-based approach in vision research."

There are various limitations with WEAVER++. First, as Levelt, Roelofs, and Meyer (1999b, p. 63) admitted, "WEAVER++ has been designed to account primarily for latency data, not for speech errors.. .in further development of WEAVER++, its error mechanism deserves much more attention." For example, Levelt et al. (1999a) compared the numbers of exchange errors produced by human participants against those produced by WEAVER++. The model produced far fewer than humans.

Second, most of the research carried out by Levelt et al. has involved the production of single words. The problem here was highlighted by Roberts, Kalish, Hird, and Kirsner (1999, p. 54):

Implementation of naming and lexical decision experiments involving isolated words can only yield evidence about the way in which isolated words are produced in response to impoverished experimental conditions. If the way in which a given word is accessed and uttered is sensitive to. contextual variables, single-word test procedures will not reveal this.

Third, higher-order cognitive processes probably have more influence on speech production than is assumed in WEAVER++. As Roberts et al. (1999, p. 54) argued, "The processes controlling higher-order processes cannot simply provide conceptual information; they must intrude during the process of lexical formation to define the correct prosodic form." For example, Hird and Kirsner (1993) studied patients with damage to the right cerebral hemisphere who had essentially intact articulation abilities. However, these patients failed to use prosodic cues to indicate the importance of key words in their utterances, suggesting an impaired ability to use higher-order processes to influence speech production.


The cognitive neuropsychological approach to speech production is of importance. However, the syndromes or labels attached to patients having difficulties with speech production often have little meaning. This is perhaps especially the case with Broca's or non-fluent aphasia and Wernicke's or fluent aphasia (both of which are discussed later), which do not form coherent syndromes. In spite of these problems, it is useful for purposes of communication to refer to some of the major syndromes that have been identified.


Some patients suffer from anomia, which is an impaired ability to name objects. According to Levelt et al.'s (1999a) theory, there are two main reasons why such patients might have difficulties in naming. First, there could be a problem in lemma selection, in which case errors in naming would tend to be similar in meaning to the correct word. Second, there could be a problem in word-form selection, in which case patients would be unable to find the appropriate phonological form of the word. As we will see, the evidence supports these predictions.

A case of anomia involving a semantic impairment (deficient lemma selection) was reported by Howard and Orchard-Lisle (1984). The patient, JCU, had good object recognition and reasonable comprehension. However, she was very poor at naming the objects shown in pictures unless she was given the first phoneme or sound as a cue. If the cue was the first phoneme of a word closely related to the object shown in the picture, then JCU would often be misled into producing the wrong answer. This wrong answer she accepted as correct 76% of the time. In contrast, if she produced a name quite different in meaning to the object depicted, she rejected it 86% of the time. JCU presumably had access to some semantic information, but this was often insufficient to specify precisely what it was she was looking at.

Problems in lemma selection are especially clear in patients who have problems in naming objects belonging to some categories (e.g., living objects) but not others (e.g., non-living objects). Some of the evidence on such category-specific naming disorders was discussed in Chapter 4. It is hard to interpret the evidence, as is shown by the fact that Forde et al. (1997) identified eight different potential explanations!

Kay and Ellis (1987) studied a patient, EST, who had problems with lemma selection. His performance on a range of tasks was so good that it seemed he had no significant impairment to his semantic system, and thus no real problem with lexeme selection. However, he had a very definite anomia, as can be seen from this attempt to describe a picture (Kay & Ellis, 1987):

Er.. .two children, one girl one male. the.. .the girl, they're in a.. .and their, their mother was behind them in in, they're in the kitchen.the boy is trying to get., a part of a cooking.jar.He's standing on.the lad, the boy is standing on a.standing on a.standing on a. I'm calling it a seat.

Close inspection of EST's speech indicates that it is reasonably grammatical, and that his greatest problem lies in finding words other than those in very common usage. What are we to make of EST's anomia? Kay and Ellis (1987) argued that his condition resembles in greatly magnified form that of the rest of us when in the "tip-of-the-tongue" state. The difference is that it is mostly relatively rare words that cause us problems. In contrast, with EST, the problem is present with all but the most common words. However, word frequency correlates highly with age of acquisition, with more common words being acquired earlier in life. Hirsh and Funnell (1997) found that age of acquisition seemed to be the main determinant of anomia, with word frequency also playing a role.


It has been assumed within most theories of speech production that there are rather separate stages of working out the syntax or grammatical structure of utterances and producing the content words to fit that grammatical structure (e.g., Dell, 1986). Thus, there should be some brain-damaged patients who can find the appropriate words, but cannot order them grammatically. Such patients are said to suffer from agrammatism or non-fluent aphasia. Patients with agrammatism also tend to produce short sentences lacking function words and word endings.

Saffran, Schwartz, and Marin (1980 a,b) studied patients suffering from grammatical impairments. For example, one patient was asked to describe a picture of a woman kissing a man, and produced the following: "The kiss.. .the lady kissed.. .the lady is.. .the lady and the man and the lady... kissing." In addition, Saffran et al. found that agrammatic aphasics had great difficulty in putting the two nouns in the correct order when asked to describe pictures containing two living creatures in interaction.

The greatest problem in studying agrammatism is the existence of large individual differences in symptoms. For example, Miceli, Silveri, Romani, and Caramazza (1989) studied the speech productions of 20 patients who could be classified as agrammatic. Some patients omitted many more prepositions than definite articles from their speech, whereas other patients showed the opposite pattern.

Do the syntactic deficiencies of agrammatic aphasics extend to language comprehension? They do in some cases, but not in others. This was established clearly by Berndt, Mitchum, and Haendiges (1996) in a meta-analysis of studies of comprehension of active and passive sentences by agrammatic aphasics. The sentences were constructed so that accurate comprehension required sensitivity to grammatical structure (e.g., "The dog was chased by the cat"). In 34% of the data sets, comprehension performance on both active and passive sentences was at, or close to, chance level. In 30% of the data sets, comprehension was better than chance on both kinds of sentences. In the remaining 36% of data sets, there was good performance on active sentences but chance performance on passive sentences. These findings lead to two conclusions: (1) agrammatic aphasics do not necessarily have major problems with language comprehension; (2) "selection of patients for study on the basis of features of aphasic sentence production does not assure a homogeneous [similar] grouping of patients" (Berndt et al., 1996, p. 298).

Jargon aphasia

Agrammatic aphasics possess reasonable ability to find the words they want to say, but cannot produce grammatically correct sentences. From most theoretical perspectives, it would be expected that there might be patients who showed the opposite pattern, namely, that they spoke fairly grammatically but had great difficulty in finding the right words. In general terms, this is the case with patients suffering from jargon aphasia or fluent aphasia. This is a condition in which the word-finding problems are so great that patients often produce neologisms, which are made-up words.

Ellis, Miller, and Sin (1983) studied a jargon aphasic, RD. He provided the following description of a picture of a scout camp (the words he seemed to be searching for are given in brackets):

A b-boy is swi'ing (SWINGING) on the bank with his hand (FEET) in the stringt (STREAM). A table with orstrum (SAUCE-PAN?) and...I don't know...and a three-legged stroe (STOOL) and a strane (PAIL)-table, table.. .near the water.

RD, in common with most jargon aphasics, produced more neologisms or invented words whenthe word he wanted was not a common one. Most jargon aphasics are largely unaware of the fact that they are producing neologisms and so do not try to correct them. In the case of RD, this may well have been linked to the fact that he could not understand spoken material even though he could understanding written material. However, there are considerable individual differences among jargon aphasics. Maher, Rothi, and Heilman (1994) studied AS, a jargon aphasic who had reasonable auditory word comprehension. AS was better at detecting his own speech errors when listening to someone else speaking than when listening to his own voice. This suggested to Maher et al. (1994) that AS was in a state of denial about his own speech errors.


Cognitive neuropsychological evidence has provided support for major theories of speech production. Findings from anomic patients have indicate the value of two-stage theories of lexicalisation. Agrammatic aphasics and jargon aphasics provide evidence for separate stages of syntactic planning and content-word retrieval in speech production. As Harley (1995) pointed out, what we have here is a double dissocation, which is a particularly powerful kind of evidence that two different sets of processes are involved. This double dissociation is consistent with theories such as those of Dell and of Levelt et al.


Are different language functions localised in specific areas of the brain? Several attempts to answer that question have been made over the past 140 years or so. Most of these attempts have been rather inconclusive, but the development of various brain-scanning techniques has led to advances in our understanding of the localisation of language functions. In this section, we will consider only a small fraction of the relevant evidence.

Non-fluent aphasia

Paul Broca studied a patient, Leborgne, who suffered from great problems with speech production but seemed to understand what was said to him. Postmortem examination of this patient and of other patients with similar speech problems suggested to Broca that damage to certain parts of the left hemisphere of the brain was responsible for the deficient speech. The so-called Broca's area consists mainly of "posterior aspects of the third frontal convolution and adjacent inferior aspects of the precentral gyrus" (Caplan, 1994, p. 1035). Patients with Broca's aphasia are now known as non-fluent aphasics or agrammatic aphasics (see earlier).

The evidence indicates that matters are more complex than was assumed by Broca. For example, Willmes and Poeck (1993) found that only 59% of patients with non-fluent aphasia had lesions or damage in Broca's area, and only 35% of patients with lesions involving Broca's area had non-fluent aphasia. Dronkers (1996) reported that only 10 out of 22 patients with lesions in Broca's area were suffering from non-fluent aphasia. In addition, he found that all of the patients with non-fluent aphasia had damage to the insular cortex, which does not form part of Broca's area.

Various PET studies on normal individuals indicate that Broca's area is involved in speech production. For example, Wise et al. (1991) gave their participants the task of silent generation of verbs as uses for a noun. This task produced activation of Broca's area in the left hemisphere. Chertkow et al. (1993) used the task of silent picture naming, and observed activation of Broca's area. However, as Howard (1997, p. 294) pointed out, "Activations of Broca's area seems less frequently found in tasks involving overt speech production, possibly because activation in this region is masked by the bilateral activation of the adjacent articulatory motor cortex."

Fluent aphasia

Carl Wernicke studied patients who had great problems in understanding spoken language, but who could speak fluently although not very meaningfully. Postmortem examination led Wernicke to identify damage to part of the left hemisphere of the brain as responsible for the comprehension problems of these patients. Wernicke's area consists of the "posterior half of the first temporal gyrus and possibly adjacent cortex" (Caplan, 1994, p. 1035). The disorder that used to be known as Wernicke's aphasia is now known as fluent aphasia or jargon aphasia (see earlier).

De Bleser (1988) studied 6 very clear cases of fluent aphasia, 7 very clear cases of non-fluent aphasia, and 33 additional aphasic patients. The sites of brain damage were assessed by means of computerised tomography (CT) scans, which allowed the patients to be put into three groups: (1) damage to frontal areas including Broca's area; (2) damage to temporo-parietal areas including Wernicke's area; and (3) large lesions including both Broca's and Wernicke's areas. Four of the six patients with fluent aphasia had damage only to Wernicke's area, but the other two had lesions in Broca's area as well as in Wernicke's area. Of the seven non-fluent aphasic patients, four had damage to Broca's area, but the others had damage to Wernicke's area.

Willmes and Poeck (1993) also used CT scans. They found that 90% of patients with fluent aphasia had lesions in Wernicke's area. However, only 48% of patients with damage in Wernicke's area had fluent aphasia.

PET studies have provided clearer evidence of the involvement of Wernicke's area in speech comprehension. For example, Howard et al. (1992) compared two conditions in which the participants either repeated real words or listened to reversed words and said the same word to each stimulus. As predicted, there was greater activation of Wernicke's area in the former condition.


Why is there more evidence for localisation of language functions in brain-scanning (e.g., PET) studies on normals than in studies of brain-damaged patients? We can readily observe individual differences in localisation with brain-damaged patients, but similar individual differences are less apparent when information about brain activation is averaged across many people. As Howard (1997, p. 298) pointed out:

While studies of motor and sensory processes produce changes in rCBF [regional cerebral blood flow] across a group of subjects of up to 30% in specific locations, language studies typically find significant changes of only 5-10%. This is exactly the pattern which one would predict if sensory and motor processes show very consistent localisation across subjects, while there is a great deal of variability of language functions.

What conclusions about speech production can we draw from the findings of cognitive neuroscience? Howard (1997, p. 288) suggested the following answer to that question:

Language functions, while localised in individual subjects, are not consistently localised in different individuals.. .higher cortical functions such as language may have a certain amount of freedom in the areas of cortex devoted to them.


Writing involves the retrieval and organisation of information stored in long-term memory. In addition, it involves complex thought processes. As Kellogg (1994, p. 13) expressed it, "I regard thinking and writing as twins of mental life. The study of the more expressive twin, writing, can offer insights into the psychology of thinking, the more reserved member of the pair." Thus, although writing is an important topic in its own right (no pun intended!), it is not separate from other cognitive activities.

Theoretical considerations

Hayes and Flower (1986) identified planning, sentence generation, and revising as the key processes in writing:

• The planning process involves producing ideas and organising them into a writing plan to satisfy the goals the writer is seeking to achieve.

• The sentence-generation process involves turning the writing plan into the actual writing of sentences.

• The revision process involves evaluating what has been written; this process can operate at a relatively specific level (e.g., individual words) or at a more general level (e.g., the structural coherence of the writing).

• The natural sequence is planning, sentence generation, and revision, but writers often deviate from this sequence.

Hayes and Flower (1980) used protocol analysis to identify the processes involved in writing. This involves tape recordings being made of writers verbalising their thoughts during writing. They studied a writer who was particularly aware of the processes he used while writing. He started by generating information, then proceeded to organisation, and then to sentence generation. There was a definite progression over time from unorganised fragments to more structured items, and then to complete sentences. As expected, the flow of processing was often interrupted by the sentence-generation and revision processes.

Another method is directed retrospection. Writers are stopped at various times during the writing process and asked to categorise what they were just doing (e.g., planning; sentence generation; revision). Kellogg (1994) discussed studies involving directed retrospection. On average, writers devoted about 30% of their time to planning, 50% to sentence generation, and 20% to revision.


Writing plans depend heavily on the writer's knowledge. Alexander, Schallert, and Hare (1991) identified three kinds of relevant knowledge:

1. Conceptual knowledge: information about concepts and schemas stored in long-term memory.

Business Correspondence

Business Correspondence

24 chapters on preparing to write the letter and finding the proper viewpoint how to open the letter, present the proposition convincingly, make an effective close how to acquire a forceful style and inject originality how to adapt selling appeal to different prospects and get orders by letter proved principles and practical schemes illustrated by extracts from 217 actual letter.

Get My Free Ebook

Post a comment