Word Production

  1. A model for word production
  2. Applying the model to a patient
  3. Speech-error evidence
1. A model for word production
Language production might be thought of as the flip side of language comprehension; that is, in comprehension, you have to take the sounds, i.e., phonemes, that you hear, group them into words, and arrive at their meaning. In production, you begin with a meaning that you wish to convey, and you have to select the appropriate words before transforming them into sounds.

How might you arrive at a blueprint, or model, of how language is produced? The above description specifies roughly the stages involved in production and mentions at least on type of unit that is processed in the stages, phonemes. If you were building a talker, what other kinds of information would it process?

Figure 1 is a depiction of a model of how language is produced (Bock & Levelt, 1994). Production begins when a message-level representation is formulated. Here a speaker forms a pre-linguistic concept of what he or she would like to say; for example, the speaker might decide to tell you about what she did last night, but has not specified which words she will use. The next step is to select the words that are appropriate for the meaning (the lexical-semantic level), followed by specifying the sound of the word (phonological form) and eventually outputting the utterance via the articulatory organs in the mouth, throat, and nose.


Figure. 1 Adapted from Bock & Levelt (1994) Model of Language Production
(Image of sheep courtesy of )

2. Applying the model to a patient
Now that we have the steps of production specified, how can we know more about what occurs at those levels, as well as the units that are processed by them? One way to get evidence for the levels in a model of speech production is with patient data. Individuals who have suffered some sort of damage to their brains can provide insight into how language production normally works by showing us what they can and cannot produce.

One person who has been studied for his patterns of production is MS (Martin, Lesch, & Bartha, 1999). MS suffers from encephalitis, which has led to damage in the bilateral temporal areas of his brain. This impairment has caused him some difficulty with producing language, which he has generously and good-naturedly agreed to allow researchers to study.

An overall impression you might have of his responses is that he seems to know a lot about the meaning of a pictured item, but he just can’t seem to access the correct word to convey that meaning. He can describe what a cane looks like and how it functions, but he can’t say the word, "cane". A preponderance of such responses suggests that he has difficulty advancing from the lexical-semantic level to the phonological level; it may be that the connections between the two levels are impaired.

Connections can be bidirectional; that is, information can spread
downward or upward. Refer again to the model of production in Figure 1. What path did he take to be able to respond correctly to "comb" in the end? After saying, "I use it to comb", he then had the correct phonological form and constituent phonemes, and they provided feedback to the lexical-semantic level to activate the correct noun entry.

3. Speech-error evidence
MS’s data are consistent with the model’s depiction of the levels of speech production. By showing how a system in a patient has broken down, we saw how semantic and phonological levels can be separate. How might we get evidence for the units that are processed by those levels?

The experience you just had attempted to get your production system to break down (with no permanent damage done!) to suggest what the units of production are. Did you make any errors? How can you typify them? This task does a good job of creating the processing circumstances that lead to error in everyday conversation; e.g, having a limit on time, or trying to talk quickly. Any errors you made probably involved switching phonemes around; but phonemes, morphemes (the smallest meaningful unit in a word; e.g, the word "walked" has two morphemes, "walk" and "ed" ), and words move around, suggesting that they are individual units that are processed at various stages in production. In experiments conducted in laboratories, errors are elicited about 10-30% of the time. About how often did you err?