Andrew Radford - Linguistics An Introduction [Second Edition]

Failures of lexical access, as, for example, in speech errors, provide another important source of evidence for discovering the internal structure of the mental lexicon, and we shall now introduce some of the key issues in this area of speech production (see section 7 for speech errors involving phonological units).

Three classes of speech errors, illustrated in ( 197) to (199 ), can be distinguished for our purposes:

(197)

Blends: two words are fused into one.

a.

Irvine is quite clear (← close and near)

b.

At the end of today’s lection (← lecture and lesson)

c.

to determine whatch (← what and which)

(198)

Substitutions: mis-selections of words

a.

He’s a high grader (← low)

b.

Don’t burn your toes (← fingers)

c.

I just put it in the oven at a very low speed (← temperature)

(199)

Word exchanges: two words within the speaker’s utterance are exchanged

a.

You can't cut rain in the trees

b. This

spring has a seat in it

Victoria Fromkin and Anne Cutler have collected speech errors over many years, and the anthologies they have put together provide the most extensive database of naturally occurring speech errors we have; the examples mentioned above were taken from these collections.

Speech errors in lexical access all involve failures of lemma retrieval, but the mechanisms underlying blends, substitutions and exchanges are different. In

general, a speech error occurs when lemma selection is disturbed by the simultaneous activation of two elements. Consider, for instance, blends, and notice from the examples in (197) that the two words forming the basis for the blend are

208

words

roughly equivalent in meaning. Thus, in (197b), lemma selection is disturbed by the fact that the two closely related elements lecture and lesson are active at the same time. But at which processing level are these two elements active? Given

figure 39, there are two possibilities: at the conceptual level and/or at the lemma level. To answer this question, we must have a closer look at the meaning relations that hold between the two elements activated in a speech error.

As noted, in blends the two elements are very similar in meaning and are usually of the same syntactic category. We hardly ever find antonym blends, i.e. fusion of two words that have opposite meanings (e.g. harsy ← easy/hard), or blends in which one element is a superordinate term for the other one (e.g. dealsman ←

dealer/salesman). In an extensive published list of blends, for example, there was not a single antonym blend and just three involving a hyponym and its

superordinate.

Compare this with the elements involved in substitution errors in (198). The most common cases of this type involve antonyms (198a) or other semantic relations. For example, fingers and toes are co-meronyms, each entering into the relation of meronymy with body. Moreover, there is a clear frequency effect in substitutions: high-frequency words are more likely to substitute for a low-frequency word, but not the other way round, and it has been found that in

74 per cent of a large corpus of substitutions, the intruding element had a higher frequency than the correct item, with only 26 per cent of cases involving a

lower-frequency item substituted for one of higher frequency.

Finally, in word exchanges the two elements that are exchanged are typically unrelated in meaning. Rather, they express different concepts, as, for example, in

(199a, b).

Let us briefly summarise the similarities and differences between these three kinds of speech errors in figure 42.

Given these facts, we conclude that the explanation of word exchanges differs radically from the other two kinds of speech errors. Word exchanges result from different sentence fragments being active at the same time. For example, in

producing (199a), there is a point at which the slot for the object of the verb cut and the slot for the object of the preposition in have to be filled, and at this point two candidate fillers, rain and trees, are simultaneously active and are somehow exchanged. Thus, the two elements that are involved in word exchanges are neither conceptually nor semantically related; rather they are syntactically related. They belong to different phrases, but they have similar syntactic functions in their phrases.

Figure 42 Differences between types of speech errors

Lexical processing and the mental lexicon

209

How do substitutions come about? Take example (198b) for illustration. In this case, the speaker wanted to convey a message involving the concept of a finger.

Given figure 39, this concept activates the lemma finger. In the mental lexicon, lemmas that are semantically related are closely associated (cf. the discussion of spreading activation earlier in this section). Thus, the lemma toe is a close associate of the lemma finger in the mental lexicon. For some reason, the activation of toe is stronger in this case than that of finger, and this produces the substitution. The kinds of errors that occur in word substitutions are familiar from word-association experiments in which subjects are asked to freely associate to a given stimulus. In such experiments, responses such as last as a response to first, wine to beer, later to earlier and sun to moon are typically found. These responses reflect the semantic structure of the mental lexicon, for example, the fact that a given lemma is closely connected to its antonym(s), synonym(s)

co-hyponym(s), etc. The same can be said about word substitutions: generally speaking, word substitutions reflect semantic relations in the mental lexicon.

Consider finally how we might explain the occurrence of blends. Blends occur between two words that are broadly similar in meaning, but unlike in the case of substitutions, semantic relations such as antonymy, hyponymy and meronymy

appear to be irrelevant. Thus, as noted, antonym blends and blends involving a word and its superordinate are extremely rare. This suggests that in blends the intrusion of the second element occurs at the conceptual level, rather than in the mental lexicon. Take, for example, (197b). The message fragment the speaker wants to convey at this point, namely selecting a reference point of the school/

university day, would be compatible with using both concepts, LECTURE and

LESSON. These two concepts are closely related and are simultaneously acti-

vated. Subsequently, they both activate their corresponding lemmas (see figure

39). Both lemmas are retrieved and inserted into the same slot. In short, blends result from conceptual intrusion. Viewed from the perspective we have sketched above, speech errors are not a random phenomenon; they reflect levels of representation in the mental lexicon (exercises 3, 4 and 5).

A rather different set of issues concerning the structure of the mental lexicon arises in connection with our observations at the end of section 12. There we noted that there is psycholinguistic evidence which supports the idea that the notion of prototype plays a role in lexical organisation, and we shall now briefly discuss a small sample of this evidence.

Recall that within the category of birds, a robin appears to be prototypical, particularly if it is contrasted with an ostrich. We suggested that this might be a reflection of the lexical entry for robin being ‘closer’ than that for ostrich to that for bird. One piece of evidence supporting this view is very easily obtained. If subjects are simply asked to list the names of birds, then typically robin, sparrow and eagle will appear early on such lists, whereas ostrich, emu and chicken will appear late, if at all. If we suppose that presentation and processing of the word bird produces activation which spreads ‘outwards’ from the lexical entry for bird, becoming less effective the further it travels, we have a ready explanation for this