COGNITIVE NEUROPSYCHOLOGY, 1997, 14 (1), 177–208

How Many Levels of Processing Are

There in Lexical Access?
Alfonso Caramazza
Harvard University, Cambridge, USA

The patterns of semantic errors in speaking and writing are used to constrain
claims about the structure of lexical access mechanisms in speech and written
language production. It is argued that it is not necessary to postulate a
modality-neutral level of lexical representation (lemma) that is intermediate
between lexical-semantic representations and modality-specific lexical
representations. A dual-stage access model is proposed in which the first stage
involves the selection of semantically and syntactically specified,
modality-specific lexical forms, and the second stage involves the selection of
specific phonological (orthographic) content for the selected lexemes.

INTRODUCTION
How are words accessed in language production? Theories of speech produc-
tion are in agreement on two fundamental points: (1) semantic, syntactic, and
lexical form information constitute independent levels of representation, and
(2) these levels of representation are probably accessed sequentially in the
course of language production. The dominant view is that lexical access
involves at least two distinct stages of processing. The first stage involves the
selection of a semantically and syntactically specified lexical representation or
lemma; the second stage involves the selection of its corresponding lexical-
phonological representation or lexeme (e.g. Bock, 1982; Bock & Levelt, 1994;
Burke, MacKay, Worthley, & Wade, 1991; Butterworth, 1989; Dell, 1986; Fay
& Cutler, 1977; Fromkin, 1971; Garrett, 1975, 1980; Harley, 1984; Kempen &

Requests for reprints should be addressed to Alfonso Caramazza, Cognitive Neuropsychology

Laboratory, Department of Psychology, William James Hall, Harvard University, 33 Kirkland St.,
Cambridge, MA 02138, USA (email: caram@ [Link]).
The work reported here was supported by NIH grants NS22201 to AC. A version of this paper
was presented at the Fourteenth European Workshop on Cognitive Neuropsychology, Bressanone,
January 1996. I am grateful to Michele Miozzo and Jennifer Shelton for many helpful comments
on earlier versions of this paper.

Ó 1997 Psychology Press, an imprint of Erlbaum (UK) Taylor & Francis Ltd
178 CARAMAZZA

Huijbers, 1983; Levelt, 1989; MacKay, 1987; Roelofs, 1992; Stemberger,

1
1985) . But this is where the agreement ends. On almost everything else, from
the nature of the information represented at each stage or level of processing,
to the overall number of processing stages, to the manner in which repre-
sentations are selected, there are substantial disagreements among models.
Thus, for example, models differ on whether they assume discrete or interactive
stages of processing, whether they assume componential or holistic repre-
sentations for meaning, whether they assume localist versus distributed repre-
sentations, and whether or not they assume morphological composition.
Despite, or perhaps because of, the many disagreements, this is a vibrant
area of research, and there are a number of active research programmes directed
at articulating the structure and content of lexical-phonological representations
and their associated access mechanisms (see, for example, Dell, 1986; Dell &
O’Seaghdha, 1991; Garrett, 1988; Levelt, 1989; Levelt et al., 1991; MacKay,
1987; Martin, Weisberg, & Saffran, 1989; Meyer, 1990; Shattuck-Hufnagel,
2
1987; Starreveld & La Heij, 1996; Stemberger, 1990) . Many fewer program-
matic efforts have focused specifically on the nature of lemma-level repre-
sentations, perhaps because semantic theory is not nearly as well developed as
phonological theory, and/or perhaps because it is difficult to address questions
about the syntactic properties of words by means of the single-word processing
tasks typically used in psycholinguistic experiments. Nevertheless, in the last
few years a number of efforts have been made to formulate explicit claims about
the content and the processing structure of lemma-level representations and to
provide experimental evidence on these issues (Bock & Eberhard, 1993; Bock
& Levelt, 1994; Jescheniak & Levelt, 1994; Levelt, 1989; Roelofs, 1992, 1993;
Schriefers, 1993; see also Garrett, 1992, for a recent review of neuropsychologi-
cal evidence on this issue).
In this paper, I address the relationship among semantic, syntactic, and word
form representations. I address this issue principally from the perspective of

1
A similar distinction has been made in the neuropsychological literature in order to explain
the dissociation between disorders of lexical-semantic processing and disorders of lexical form
retrieval. Most models of the lexical system assume a two-stage retrieval process in which the
first stage involves accessing a lexical-semantic representation and the second stage involves
accessing modality-specific lexical forms (e.g. Caramazza & Hillis, 1990; Ellis, 1985; Howard &
Franklin, 1988; Humphreys, Riddoch, & Quinlan, 1988). This view should not be confused with
a superficially similar proposal that distinguishes between cognitive and lexical form systems (e.g.
Allport, 1985; Allport & Funnell, 1981). On the latter view, lexical access only involves one stage:
The retrieval of lexical forms from nonlinguistic conceptual information.
2
Although on a much smaller scale, in recent years there have also been a number of attempts
to articulate the structure and content of lexical-orthographic representations (see, for example,
Badecker, Hillis & Caramazza, 1990; Caramazza & Miceli, 1989, 1990; Jonsdottir, Shallice, &
Wise, 1996; Kay & Hanley, 1994; Link & Caramazza, 1994; McCloskey, Badecker, Goodman-
Schulman, & Aliminosa, 1994).
 LEXICAL ACCESS 179

cognitive neuropsychology—that is, by considering the language production

performance of brain-damaged subjects—but I will also discuss some experi-
mental evidence from normal speech production when relevant. The paper is
organised as follows. After a brief review of the arguments and evidence for
the distinction between lemma and lexeme levels of representation, I discuss
the most influential proposal about the nature of lemma representations. A
review of recent experimental evidence concerning the relationship between
syntactic features and semantic and phonological information reveals several
problems with current formulations of lemma-level representations. The prin-
cipal part of the paper follows, and it concerns the analysis of semantic errors
in naming, reading, and writing tasks by brain-damaged subjects. A crucial
aspect of the argument presented here rests on the relationship between
phonological and orthographic lexical forms—the principal issue under con-
sideration in this special issue of Cognitive Neuropsychology. The implications
of the distribution of semantic errors in speaking and writing for the relationship
among semantic, syntactic, and word form representations are explored.
Finally, I present a model of lexical access that can better account for the major
results on lexical access.

THE DISTINCTION BETWEEN LEMMA AND

LEXEME LEVELS OF REPRESENTATION
There are compelling arguments and empirical evidence for distinguishing
between at least two levels of lexical representation. For example, the distinc-
tion between lemma and lexeme representations provides a natural account for
the existence of homonyms: Words that are phonologically and orthographi-
cally identical but which differ in meaning and/or grammatical class (e.g. to
watch/the watch; the bank [money]/the bank [river]). The relevant distinctions
between these words are not at the level of lexical form (since they are identical)
but at the level of semantic and syntactic properties. Thus, if we are to capture
the lexical distinctions between the members of homonym pairs it has to be at
a level other than that of phonological (orthographic) content of the word pairs.
But arguments from the structure of language can only take us so far. A
processing model will have to be based on empirical evidence concerning
language use. Here, too, there is ample evidence in support of a dual-stage
model of lexical access (for reviews see Bock & Levelt, 1994; Butterworth,
1989; Dell, 1986; Garrett, 1988, 1992; Levelt, 1989).
Various sorts of data have been cited in support of the lemma/lexeme
distinction, including naturally occurring and experimentally induced speech
errors or slips of the tongue (e.g. Dell, 1990; Dell & Reich, 1981; Fay & Cutler,
1977; Fromkin, 1971; Garrett, 1975, 1976; Stemberger, 1985). These data are
the most extensive and the earliest evidence proffered in support of the
two-stage model of lexical access. For example, Garrett (1975, 1976) noted that
180 CARAMAZZA

the elements that enter in word and sound exchange errors, such as the examples
shown in Table 1, are subject to different distributional constraints: Word
exchanges tend to involve words of the same grammatical class and occur
between phrases, whereas sound exchanges tend to occur in words of different
grammatical classes within a phrase. Furthermore, sound exchanges but not
word exchanges are usually phonologically similar elements and occur in
similar phonological environments. The fact that word exchanges are con-
strained by grammatical features and not by phonological properties, and the
fact that sound exchanges are constrained by phonological and not semantic or
syntactic properties, has been taken to indicate that separate lexical access
stages are involved in speech production: the first stage retrieves a semantically
and syntactically specified representation; the second stage retrieves a
phonologically specified representation. This conclusion, aside from debates
about whether the two stages are independent or interactive (e.g. Dell &
O’Seaghdha, 1991; Levelt et al., 1991), has remained a constant of all models
of speech production.
Other evidence cited in support of the lemma/lexeme distinction includes
the reaction time data in naming and lexical decision experiments (e.g. Jesche-
niak & Levelt, 1994; Kempen & Huijbers, 1983; Levelt & Maassen, 1981;
Levelt et al., 1991; Schriefers, Meyer, & Levelt, 1990) and the pattern of
hesitation phenomena in normal and aphasic speech (Butterworth, 1979, 1980;
Butterworth & Beatty, 1978). Perhaps the most intuitively appealing evidence
for the lemma/lexeme distinction is the tip-of-the-tongue (TOT) phenome-
non—the feeling of knowing a word that is momentarily inaccessible for
production (R. Brown & McNeill, 1966; Burke et al., 1991; Jones & Langford,
1987; Kohn et al., 1987; Koriat & Lieblich, 1974; Perfect & Hanley, 1992;
Rubin, 1975; and see A.S. Brown, 1991, for review). This phenomenon has
been interpreted as reflecting the failure to retrieve a lexeme in the context of
successful retrieval of its lemma. And, of course, there is the evidence from
aphasia, which shows there are word production disorders that can be attributed
to a deficit in lexical-semantic processing (e.g. Gainotti, 1976; Hillis, Rapp,
Romani, & Caramazza, 1990; Warrington, 1975) and production disorders for
which the deficit can clearly be localised at the level of retrieval of lexical forms

TABLE 1
Examples of Word and Sound Exchange Errors
Word exchange errors: They left and forgot it behind (Garrett, 1988)
I left the briefcase in my cigar (Garrett, 1980)
. . . writing a mother to my letter (Dell & Reich, 1981)
Sound exchange errors: She’ s a real rack pat . . . (Garrett, 1988)
It comes down to a choice of stummer sipends . . . (Garrett, 1980)
. . . lorck yibrary . . . (Dell & Reich, 1981)
 LEXICAL ACCESS 181

in the face of spared lexical-semantic processing (e.g. Caramazza & Hillis,

1990; Howard & Franklin, 1989; Kay & Ellis, 1987). Thus, there is a wide
spectrum of evidence consistent with the two-stage model of lexical access.
Although there appears to be substantial agreement among researchers of
speech production on the distinction between a lemma and a lexeme level of
lexical representation, the content and processing dynamics of lemma repre-
sentations remain relatively unspecified. Although everyone agrees that lem-
mas are modality-independent, semantically and syntactically specified lexical
representations, there are different ways in which one can implement the
distinction between lexical form representations (phonological and ortho-
graphic) and modality-independent, semantic, and syntactic representations.

THE STRUCTURE OF LEMMA REPRESENTATIONS

The clearest proposal concerning the processing structure of lemma repre-
sentations is to be found in a paper by Roelofs (1992) and subsequently adopted
by Bock and Levelt (1994) and Jescheniak and Levelt (1994). Their model
distinguishes among three levels of representation (see Fig. 1): the conceptual,
the lemma, and the lexeme levels. The conceptual level represents lexical
concepts as unitary nodes in a conceptual network. The meaning of a word is
given by the set of labeled connections between a concept node and other nodes
in the network. Each lexical concept node is connected to a lemma node; lemma
nodes are modality-independent units that are connected to a set of syntactic
nodes specifying such properties as grammatical class (Noun, Verb, etc.),
gender, and auxiliary type (be or have). Each lemma node is connected to a
lexeme node which, through its connections to segmental nodes, specifies the
phonological (and orthographic) form of a word. In Fig. 1, the lexical concept
[TIGER] is connected to its Italian lemma node {tigre}, which is connected to
the category node N(oun) and the gender feature F(eminine), and, in turn, the
lemma node is connected to its lexeme /tigre/. Lexical access in this model is
represented by the sequential selection of lemma (and hence the syntactic
features that define a word) and lexeme nodes through spreading activation
emanating from the lexical concept node.
Several properties of this model should be stressed for present purposes.
First, word meaning is represented by unitary concept nodes, with each node
connected to only one lemma node. Spreading activation from the concept node
only activates its corresponding lemma node. However, spreading activation
within the conceptual level from the concept node to connected nodes (e.g.
ANIMAL, STRIPES, WILD, etc.) will result in the partial activation of the
latter nodes and, consequently, the weak activation of their corresponding
lemma nodes. Second, the selection of a word’s lemma node is tantamount to
the selection of the syntactic nodes/features that define that word. And, third,
the selection of a lexeme is mediated by the selection of the word’s grammatical
182 CARAMAZZA

visual
form

FIG. 1. Part of the lexical system showing the relation between lemma and other levels of lexical
representation: The “Syntactic Mediation” hypothesis (see text). The lemma and lexeme levels show
the Italian words for the lexical concepts TIGER, APPLE, TO GO, and TO DRINK. (Adapted from
Bock & Levelt [1994], Jescheniak & Levelt [1994], and Roelofs [1992].)

features. Given the centrality of syntactic information in defining the structure

of lemmas and in accessing lexemes, I will call this model of the structure of
lemmas the “syntactic mediation” (SM) hypothesis.
In various papers, Dell (1986, 1990; Dell & O’Seaghdha, 1991) has proposed
an interactive network model with a very similar hierarchy of lexical levels to
that proposed by Bock and Levelt (1994), Jescheniak and Levelt (1994), and
 LEXICAL ACCESS 183

Roelofs (1992). Dell (1990, pp. 331–332) draws a similar distinction among
semantic, lemma, and lexeme representations: “The lemma node represents the
lexical item as a syntactic/semantic entity. It corresponds to Dell’s (1986) `word
node’ and is assumed to connect directly to conceptual structure and to syntactic
information. Below that, the lexeme node is a single unit representing the
phonological form of the word. This corresponds roughly to the morpheme
node and/or to the set of syllable nodes in Dell (1986) and MacKay (1982). The
lexeme node connects to phonological segments (. . .).” However, this model
differs from that of Levelt and collaborators on at least two crucial points: (1)
in Dell’s model, word meanings are represented in componential form, and (2)
the stages of lexical processing are not discrete but interactive. Nonetheless,
for present purposes, the two models make a common assumption: There is a
level of representation, lemma, that is distinct from lexical-semantic and from
3
lexical-phonological information .
The modality-neutral lemma hypothesis—a hypothesis shared by Roelof’s
discrete stage model and Dell’s interactive model—predicts that retrieval of
syntactic information is necessary for the successful retrieval of lexemes.
However, reviews of experimental results with normal subjects in tasks requir-
ing access to syntactic and word form information and of the performance of
brain-damaged subjects with selective deficits in word production raise con-
siderable difficulties for this hypothesis. I will argue that there are grounds for
rejecting the hypothesis that a modality-neutral lemma node intervenes between
lexical-semantic representations and word forms. However, I will also argue
that there are equally compelling grounds for the autonomy of syntactic
information from semantic and word form representations. These seemingly
conflicting conclusions will be reconciled in a new model of the processing
structure and organisation of lexical knowledge.

THE INDEPENDENCE OF SYNTACTIC FROM

SEMANTIC AND WORD FORM INFORMATION
The neuropsychological literature is replete with evidence about the crucial role
of syntactic information in the organisation of lexical knowledge. The most
celebrated dissociation is that between closed- and open-class words (or func-
tion and content words). Although principally discussed in the context of
agrammatic and paragrammatic speech performance (e.g. Buckingham &
Kertesz, 1976; Goodglass, 1976), the dissociation can also be seen in single-

3
Parenthetically, it should be noted that it is not obvious that Dell can maintain both that there
are interactions between semantic and phonological levels, which he needs to explain the results
showing that word substitutions are affected by both semantic and phonological similarity (Dell
& Reich, 1981), and that there is a modality-neutral level of representation, lemma, intervening
between lexical-phonological and semantic representations.
184 CARAMAZZA

word processing tasks in the context of diverse clinical pictures (Andreewsky

& Seron, 1975; Caramazza, Berndt, & Hart, 1981; Gardner & Zurif, 1975; see
also papers in Coltheart, Patterson, & Marshall, 1980). Thus, for example, two
acquired dyslexic subjects—RG (Beauvois & Déro uesné, 1979) and AM
(Patterson, 1982)—could read most (>92%) content words correctly, including
abstract words, but showed considerable difficulty (only 70% correct) in
reading function words.
The other major grammatical class dissociation is between nouns and verbs
(e.g. Berndt, Mitchum, Haendiges, & Sandson, 1997; Caramazza & Hillis,
1991; Damasio & Tranel, 1993; Daniele, Giustolisi, Silveri, Colosimo, &
Gainotti, 1994; De Renzi & di Pellegrino, 1995; Hillis & Caramazza, 1995;
Kremin & Basso, 1993; McCarthy & Warrington, 1985; Miceli, Silveri, Villa,
& Caramazza, 1984; Miceli, Silveri, Nocentini, & Caramazza, 1988; Zingeser
& Berndt, 1988). There are a number of reports of brain-damaged subjects who
show a selective difficulty in producing nouns in the context of relatively spared
ability to produce verbs, and those who show the opposite pattern of relative
difficulties with these two classes of words. Thus, for example, De Renzi and
di Pellegrino (1995) have described a subject with a frontotemporal lesion who
showed a selective sparing of verbs. For example, in an oral spelling task he
correctly produced over 93% of verbs but only about 45% of nouns and 40%
of function words. His difficulties with words of other grammatical classes
could not be ascribed to a deficit in processing concrete words (Breedin,
Saffran, & Coslett, 1994; Warrington, 1975) because he showed no advantage
for this type of words within grammatical classes. And function words, which
are at least as abstract as any verb, fared no better than concrete nouns. It would
seem that the effect is purely grammatical in nature.
The existence of these grammatical class dissociations implies that syntactic
information is one of the dimensions along which the lexical system is organ-
ised. However, they do not establish precisely where in the lexical system
syntactic information is represented, nor do they clearly specify the relation of
syntactic to semantic and word form representations. Particularly relevant for
the latter purpose is the performance of brain-damaged subjects who show
selective difficulties in producing words of one grammatical class in only one
modality of output (Caramazza & Hillis, 1991; Hillis & Caramazza, 1995; Rapp
& Caramazza, 1997; Rapp, Benzing, & Caramazza, 1995). For example,
subject SJD showed severe difficulties writing the very verbs that she could
easily produce orally. In written and oral production tasks with homonyms (e.g.
to watch/the watch) she correctly produced both the noun and verb forms in
speaking but was only able to write the noun forms correctly (Caramazza &
Hillis, 1991). Other examples of modality-specific grammatical class effects
include subjects who show double dissociations of grammatical class by
modality. Thus, EBA (Hillis & Caramazza, 1995) showed a selective deficit in
recognising written forms of verbs, but a selective difficulty in producing nouns
 LEXICAL ACCESS 185

orally; KSR (Rapp et al., 1995) was selectively impaired in producing nouns
in speaking, but verbs in writing; and PW (Rapp & Caramazza, 1997) has
difficulties producing closed-class words in writing and open-class words in
speaking. The striking commonality in these cases is the remarkable specificity
of the disorder—the deficit involves words of a specific grammatical class in
only one modality of output (or input).
These selective grammatical class deficits, restricted to either oral or written
production, provide the evidence we need to argue that syntactic knowledge is
represented independently of both lexical-semantic and word form information.
We reason as follows: The fact that the deficit is restricted to one modality of
output implies that the lexical-semantic system is intact; furthermore, given that
the lexical-semantic system is intact and given that the impairment in these
subjects is limited to one grammatical class, jointly imply that the deficit must
concern a syntactic level of representation. Thus, we are led to conclude that
lexical-semantic and syntactic information are represented independently.
Evidence for the autonomy of syntactic from word form information is found
in the performance of anomic subjects who are able to provide information
about the syntactic features of words they are unable to produce. Thus, for
example, Henaff Gonon, Bruckert, and Michel (1989) described a French-
speaking anomic patient who could correctly provide the grammatical gender
(13/14 correct) of the nouns he failed to produce in various naming tasks. A
more systematic investigation of this type of dissociation has recently been
reported by Badecker, Miozzo, and Zanuttini (1995). These investigators
described the performance of an anomic subject, Dante, who despite his
inability to produce the names of objects, or even guess the first or last phoneme
in a two-phoneme, forced-choice task, was virtually always able to recall their
gender correctly. In a more recent investigation of the same subject (Miozzo &
Caramazza, in press b), the dissociation between the retrievability of syntactic
and word form information has been extended to verbs. Dante was shown to
be able to provide correctly the auxiliary form of a verb (be or have) he was
unable to retrieve: He correctly recalled the auxiliary of verbs he could not name
in 99% of the trials, but was exactly at chance level in guessing the initial
phoneme of the word. These results clearly show that a word’s syntactic
features are represented independently of its form, thereby permitting their
independent access. However, they leave unanswered the question of how
syntactic features are activated and how they are related to semantic and
phonological information. Thus, for example, do syntactic features mediate
between semantic and phonological information as proposed by the SM hy-
pothesis? Tentative answers to this question are provided by normal subjects’
ability to retrieve gender and phonological information in TOT states.
In a series of experiments (Caramazza & Miozzo, 1997; Miozzo &
Caramazza, in press a), we addressed the question of whether the availability
of syntactic and phonological information are correlated in TOT states. For this
186 CARAMAZZA

purpose we compared subjects’ relative ability to retrieve grammatical gender

and the initial phoneme of words in TOT states. Grammatical gender is a
syntactic feature of nouns that, in Italian, is not deducible from their meaning.
In several experiments, we were able to show that subjects are able to retrieve
both the grammatical gender and the initial phoneme of words in TOT states
with well above chance-level accuracy but that performance for the two features
is uncorrelated. These results demonstrate that, contrary to the SM hypothesis,
access of a word’s phonological features does not strictly depend on the prior
4
access of its syntactic features .
Converging evidence in support of the claim that access of a lexeme
representation does not require successful access of its syntactic features can
be found in a recent series of experiments by Jescheniak and Levelt (1994). In
a study designed to explore the locus of the frequency effect in speech
production, they had Dutch subjects either name a picture (Expt. 1) or make a
gender decision about the name of the picture (Expt. 4). The results of the
experiments are very clear: The frequency effect remains over repeated trials
in the naming condition, but is not sustained in the gender decision task. In the
gender condition they obtained a clear “priming” effect: It appears that once
gender is retrieved, its subsequent retrieval is independent of the word’s
frequency. For present purposes, the important result concerns a further
manipulation introduced by Jescheniak and Levelt. They wanted to determine
whether the gender priming effect they observed in Expt. 4 depended on the
retrieval of the lexeme or whether the direct retrieval of gender information was
required. Clearly, if retrieval of a word’s lexeme requires the prior selection of
(all) its syntactic features, then the retrieval of the lexeme should lead to gender
priming because gender, too, has been selected. To address this issue, they
carried out the following two experiments. In Experiment 5a, they first gave
subjects two blocks of naming trials followed by two blocks of a gender
decision task with the same items; in Experiment 5b, they had the same
arrangement of blocks of naming and gender decision trials but this time the
naming task required subjects to produce the full noun phrase (article + noun)

4
This conclusion may be too strong. It is entirely possible that there are fundamental
differences in the way in which different grammatical features of a word are represented and
accessed. Thus, for example, we might want to distinguish between what I will call “intrinsic”
and “extrinsic” grammatical features. The former, intrinsic features, refers to those properties that
are inherently associated with a word (e.g. grammatical class and gender); the latter, extrinsic
features, refers to those grammatical properties that are contextually determined (e.g. number and
tense). And within the set of intrinsic features, we may want to distinguish between the more and
the less arbitrary features: “gender” is a purely arbitrary feature; “noun” is not nearly as arbitrary
given the meaning of the word. It may turn out that the accessibility of the different types of
grammatical features for any one word is not uniform.
 LEXICAL ACCESS 187

in naming objects. Facilitation in the gender task was only obtained in Experi-
ment 5b. That is, facilitation in the gender decision task was only obtained when
the preceding trials required subjects to retrieve gender information explicitly;
the mere retrieval of a word’s lexeme (Expt. 5a) did not facilitate access to its
gender information. These results suggest that phonological information can
be accessed “independently” of grammatical information, and may even imply,
contrary to the interpretation preferred by Jescheniak and Levelt, that the
selection of a lemma does not lead “automatically” to the activation of its
associated grammatical features.
There is also indirect evidence from the spontaneous speech of brain-
damaged subjects that is consistent with the claim that lexemes can be accessed
without correct access to their gender features . These subjects make syntactic
agreement errors despite their ability to retrieve the phonological form of
words. For example, the Italian speaker, FS (Miceli & Caramazza, 1988)
produced utterances such as the following: “Poi io ascolto il (masculine
singular) televisione (fem. singular)—then I listen [to] the television”; “ . . .
perché il (masc. sing.) giornate (fem. plural) [sono] lungo (masc. sing.)— . . .
because the days are long.” In this example, FS correctly retrieved the lexemes
televisione and giornate but not the gender information needed to select the
correct article (la and le, respectively) and the proper inflection on the adjective
(lunghe, in this case). These examples suggest the possibility that the retrieval
of lexemes may be spared in the face of damage to grammatical features.
A similar case has been described by Cubelli and Perizzi (1996). They report
a patient who produced utterances such as “Il nonno è seduto sul (masc. sing.)
panchino (neologism)”—the grandfather is seated on the bench—instead of “Il
nonno è seduto sulla (fem. sing.) panchina (fem. sing.)”; and “La palla è sotto
il (masc. sing.) sedio (neologism)”—the ball is under the chair—instead of “La
palla è sotto la (fem. sing.) sedia (fem. sing.).” This case, too, clearly demon-
strates that the phonological representation of a word (albeit deformed) can be
accessed despite access of the wrong syntactic features.
The final example I will consider here concerns a subject who made
article/noun agreement errors involving the mass/count distinction (Semenza,
Mondini, & Cappelletti, 1995). This subject produced utterances such as “Nella
cucina c’è sempre una panna”—In the kitchen there’s always one cream;
instead of “Nella cucina c’è sempre della panna”—In the kitchen there’s
always some cream; and “Una farina nel sacco (. . .) ”—one flour in the bag .
. . ; instead of “La farina nel sacco (. . .) ”—The flour in the bag . . . . These
examples demonstrate that the subject could correctly access the phonological
representation of words despite the occasional failure to access (some of) their
syntactic features.
In short, the agreement errors produced by the three cases briefly reviewed
here suggest that access of a word’s syntactic features is not required in order
to activate its lexeme.
188 CARAMAZZA

INTERIM SUMMARY

There is ample evidence in support of two-stage models of lexical access. None

of the results I have reviewed challenges this view of the lexical system. To the
contrary, the results further confirm the validity of distinguishing between a
level of processing where word forms are represented and a level or levels of
processing where lexical-semantic and syntactic information are represented.
However, the evidence also indicates that access to a word’s lemma level does
not automatically lead to access of its syntactic features (Jescheniak & Levelt,
1994; Miozzo & Caramazza, in press a). Although the latter results are prob-
lematic for the view that syntactically specified lemma representations mediate
access of lexical forms, a seemingly simple modification of the model could
easily accommodate the recalcitrant results we have reviewed.
The proposed modification may be readily understood by considering the
schematic representation of the model shown in Fig. 1. The model postulates
the existence of autonomous lemma nodes that are directly activated by their
corresponding lexical concept nodes; the lemma nodes, in turn, activate their
corresponding lexeme nodes. In the original formulation of the model, the
activation of the lemma node automatically spreads to its associated syntactic
nodes. In this model, selection of a lemma node entails the selection of a set of
syntactic features. If one drops the assumption that the selection of the lemma
node implies the selection of its syntactic features and assumes instead that the
selection of the lemma node merely makes it possible for the subsequent, but
not necessary, selection of syntactic features, then, it might be possible to
accommodate the observation that access of a word’s phonological features
does not presuppose access of its syntactic features. The modification enter-
tained here essentially argues that selection of lemma is formally independent
of the selection of its syntactic features. This move saves the syntactic media-
tion hypothesis but renders the motivation for postulating an autonomous
lemma node less than compelling. The original motivation for assuming an
autonomous lemma level was to capture at once the autonomy of syntactic
information and the dependence of lexeme representations on their syntacti-
cally specified lexical representations. If one were to abandon the assumption
that access to the lemma node entailed the automatic selection of its syntactic
features, it would be unclear why one would want to have such a node in the
first place—its role would have been reduced to a contentless waystation to
syntactic and phonological representations: The lemma node would have been
rendered superfluous.
Putting aside these considerations and granting for the moment the plausi-
bility of the new formulation of the organisation and processing structure of the
lemma level, there are empirical reasons for rejecting the postulation of a
contentless lemma node. The evidence comes from the contrasting patterns of
lexical production errors in speaking and writing. These data have not pre-
 LEXICAL ACCESS 189

viously been given due consideration in the development of models of lexical

access (but see an excellent review in Garrett, 1992). But, as I hope to show in
what follows, the analysis of the contrasting patterns of performance in the
selection of phonological and orthographic lexical forms in brain-damaged
subjects can help reduce the range of plausible theories of lexical access.
In order to use lexical access performance in spelling tasks to constrain
claims about the processing structure of the lemma and lexeme levels of
representation, it must first be established that the relation of lemma to ortho-
graphic word forms is direct and not mediated by access of their corresponding
phonological forms. In other words, it must first be demonstrated that access
of orthographic lexemes (O-lexemes) occurs independently of access of
phonological lexemes (P-lexemes) (see Fig. 2 for a schematic representation of
the two hypotheses). The reason for this preliminary step in evaluating the
evidence from written language production is simple: If spelling is mediated
by phonology, then spelling performance will not directly reflect the interac-
tion between lemma and lexical form access, but the complex interaction
between lemma and P-lexemes and between P-lexemes and O-lexemes. How-
ever, if it could be established that access of O-lexemes is not mediated by the
prior access of their associated P-lexemes, then we would be able to use lexical
errors in written production to inform theories of the relation between lemma
5
and lexeme levels of representation .

THE AUTONOMY OF LEXICAL-ORTHOGRAPHIC

REPRESENTATIONS
Arguably, the clearest evidence for the autonomy of orthography in language
production comes from the neuropsychological literature. If written production
is phonologically mediated, we would expect that damage to the phonological
system should necessarily result in deficits in written language production.
However, it has been observed repeatedly that the ability to spell is often
preserved even though phonological production is severely impaired (e.g.
Alajouanine & Lhermitte, 1960; Assal, Buttet, & Jolivet, 1981; Basso,
Taborelli, & Vignolo, 1978; Caramazza, Berndt, & Basili, 1983; Ellis, Miller,
& Sin, 1983; Hanley & McDonnell, this issue; Hier & Mohr, 1977; Lecours &
Rouillon, 1976; Lhermitte & Dér ouesné, 1974; Patterson & Shewell, 1987;
Shelton & Weinrich, this issue). Although this dissociation has often been
interpreted as sufficient evidence for the hypothesis of phonologically unme-
diated (direct) access of O-lexemes (e.g. Allport & Funnell, 1981), this conclu-
sion would only be warranted if it could be shown that the deficit in speech

5
This conclusion does not entail that the relation between the lemma level and the correspond-
ing P-lexeme and O-lexeme levels must be identical. However, it is more parsimonious to begin
with this assumption.
190 CARAMAZZA

FIG. 2. Schematic representation showing the relation between lemma and lexeme representations.
Panel A shows the phonological mediation hypothesis; panel B shows the orthographic autonomy
hypothesis.

production was not the result of damage to post-lexical phonological processes

but of damage directly to the phonological lexicon. Otherwise, it could be
argued that O-lexeme access is mediated by P-lexeme access and that the
observed impairment in speech production was merely the result of damage to
post-lexical phonological processes. However, there are a number of results
that are not subject to these reservations. The clearest examples are those where
brain-damaged subjects make semantic errors in oral naming but not in written
naming (e.g. subjects RGB and HW: Caramazza & Hillis, 1990) or semantic
errors in written naming but not in oral naming (e.g. subject SJD: Caramazza
& Hillis, 1991). The production of lexical (semantic) errors indicates a lexical
as opposed to a post-lexical process as the locus of damage; the fact that the
semantic errors occur in only one modality of output indicates that P-lexemes
and O-lexemes are activated independently by their lemma-level repre-
sentations. In other words, the fact that subjects RGB and HW were able to
retrieve the correct lexical form in writing despite the production of semantic
errors in speaking demonstrates that access of O-lexemes is not mediated by
 LEXICAL ACCESS 191

prior access of P-lexemes (see also Hanley & McDonnell, this issue; Shelton
& Weinrich, this issue; and Rapp & Caramazza, 1997, for discussion of other
relevant evidence).
One other type of evidence in support of the orthographic autonomy hy-
pothesis is the observation of different oral and written semantic errors in
response to the same object in double, sequential naming tasks. WMA (Miceli,
Benvegnù, Capasso, & Caramazza, this issue) and PW (Rapp, Benzing, &
Caramazza, this issue) produced semantic errors both in speaking and in
writing. Crucially for present purposes, in double naming tasks they produced
inconsistent lexical responses in oral and written naming of the same pic-
ture—for example, they might produce a correct response in writing followed
by a “don’t know” or a semantic error in oral naming—and they occasionally
produced different semantic errors in oral and written naming. For example, in
response to a picture of tweezers, PW orally named it “pliers,” wrote “needle,”
and then orally named it again “pliers.” Similarly, the Italian subject WMA, in
response to a picture of a cook (cuoco) said “pietanza” (dish) but wrote
“forchette” (forks), and in response to a picture of peppers (peperoni) wrote
“tomato” (pomodoro) but said “carciofo” (artichoke). These patterns of per-
formance undermine the phonological mediation hypothesis of written lan-
guage production, and they suggest instead that lemma representations
independently activate their associated O-lexeme and P-lexeme repre-
sentations. Thus, we can safely proceed to interpret the implications of con-
trasting patterns of P-lexeme and O-lexeme selection errors for models of
lexical access and, more specifically, for theories of the structure of lemma-
level representations.

DISPENSING WITH THE CONTENTLESS

LEMMA NODE
The evidence that will be used for this purpose is the pattern of dissociations
of lexical errors in speaking and writing. There are several steps to the
argument. I begin by briefly summarising the relevant facts.
As already noted, there are brain-damaged subjects who make semantic
errors in only one modality of output. The deficit in these patients can be located
unambiguously in the language production system at a stage beyond the
lexical-semantic level. Especially convincing in this regard is the fact that when
subjects were asked to read aloud and define words, they often made semantic
errors in oral reading but invariably went on to provide the correct definition
(Caramazza & Hillis, 1990). For example, HW read the word pirate as “money”
but then went on to define the word she was supposed to read as “Has a thing
over its eye . . . I would say that they don’t have any anymore, but they do in
business. He wants your money and your gold.” And RGB read pharmacist as
192 CARAMAZZA

“drugs” and went on to define the stimulus as “He gives you your prescrip-
tions.” Observations such as these, together with the fact that subjects per-
formed virtually flawlessly in various comprehension tasks, confine the deficit
to output processes. More important for present purposes is the observation that
semantic errors were restricted to one modality of output. This implies that
lemma-level representations were correctly activated. That is, since subjects
could consistently select the correct lexemes in one modality of output itfollows
that their associated lemmas must have been correctly accessed, or it would not
have been possible to produce the correct lexemes in the first place. A further
relevant observation is that their semantic errors occurred in the context of
unimpaired post-lexical processes. Thus, for example, HW only occasionally
produced an articulatory error; RGB spoke fluently and without phonological
or articulatory errors; and SJD only rarely made spelling errors. In all three
cases, their errors consisted of fluently produced semantic substitutions. These
facts rule out a post-lexical deficit as a determinant of the contrasting patterns
of performance in oral and written naming. Thus, the locus of damage in these
subjects is at a point between the correctly selected lemma-level representations
and their modality-specific lexeme representations.
Having confined the possible locus of damage in subjects HW, RGB, and
SJD to a point between the lemma and the lexeme levels of representation, we
are confronted with a puzzle: If the correct lemma has been selected, how can
the inaccessibility of a modality-specific lexeme result in a semantic error? It
is not immediately apparent how models that postulate a “contentless” node
between lexical-semantic and lexeme levels of representation can account for
the occurrence of semantic errors in a single modality of output. That is, it is
not clear what happens when a correctly selected lemma node fails to activate
either its P- or O-lexeme. Since the lemma node has a discrete, one-to-one
relationship to its P- and O-lexemes, the expectation ought to be that failure to
activate one of its modality-specific lexemes would result in the absence of a
response and not in the production of a semantic error. Nonetheless, although
it does not flow naturally from its basic architecture, the lemma-as-abstract-
node hypothesis could be made to account for the modality-specific semantic
errors as follows.
It could be argued that damage to the connections between lemma and
lexemes in one modality leads to the reselection of another lemma because of
spreading activation within the conceptual system. The chain of events might
be as follows: the correctly selected lexical concept activates the correct lemma,
leading to its selection; the correctly selected lemma activates its associated P-
and O-lexemes; if one of these lexemes cannot reach threshold because of
damage within that level of representation, a different lexical-concept node is
selected from among the set of nodes that have become activated by the
spreading activation from the originally selected lexical-concept node; the
newly selected lexical-concept node will activate its associated lemma node,
 LEXICAL ACCESS 193

and so on. The way in which the abstract lemma hypothesis can motivate the
reselection of another lemma is by proposing that this process is undertaken
whenever the appropriate lexeme node does not reach threshold.
This account is further strained when we consider the performance of
subjects WMA (Miceli et al., this issue) and PW (Rapp et al., this issue), who
make different oral and written semantic errors in naming the same object. Here
the story would take the following form: The reason a semantic error is made
in one modality is, as previously discussed, because the target lexeme in that
modality could not reach threshold and therefore a new cycle of lexical selection
beginning with a related lexical-concept node would have to be undertaken. To
explain the production of a semantic error in the other modality we would have
to argue that in that modality, too, the target lexeme could not reach threshold
and therefore yet another cycle of lexical selection would have to be undertaken
for production in that modality. What has to be explained, however, is why a
different lexical concept is selected the second time. We know that the reason
for this is not because the process of selection of alternative lexical concepts
from one trial to another is random. This is shown by the fact that consecutive
responses in the same modality always resulted in the same response—it was
only across modalities that different lexical responses were produced to the
same stimulus. Thus, the selection of a different semantically related lexeme
must be due to the failure to select successfully the other modality lexeme of
the previously selected lemma. An example should make this clear. Subject PW
produced the following sequence of responses to a picture of tweezers: “pliers,”
“needle,” and “pliers,” in oral, written, and oral production respectively. To
explain this performance we would have to assume that the P-lexeme repre-
sentation /tweezers/ could not reach threshold and consequently the next
lexical-concept node selected was “pliers.” This cycle could be brought to
completion because the P-lexeme /pliers/ could reach threshold and was
selected. We would then have to argue that in trying to write the name of
tweezers, PW again selected the lexical-concept node correctly and proceeded
as before, only to encounter difficulty in selecting the O-lexeme for tweezers.
At this point we would expect that the same alternative lexical-concept node as
the one selected in the previous trial (pliers) would again be selected (as in
within-modality consecutive trials). But since the subject produced a different
semantic error in writing, we must assume that the O-lexeme for “pliers” could
not reach threshold and therefore a completely new cycle of lexical selection
had to be undertaken. In the new cycle the lexical-concept node “needle” was
selected and this time the cycle could be brought to completion.
The two cases considered here do not constitute knock-down arguments
against the abstract lemma hypothesis. And it might even be argued that it has
been shown that the abstract lemma hypothesis is able to account for the
supposedly recalcitrant results. However, the way in which the model has been
forced to account for the results provides the most compelling argument against
194 CARAMAZZA

the hypothesis. The explanation proposed for the contrasting patterns of seman-
tic errors across modalities of output gives absolutely no role to the abstract
lemma node. Had we completely omitted all mention of this node, we would
have produced a formally equivalent argument. That is, the burden of explana-
tion for the existence of contrasting semantic errors in oral and written naming
was completely carried by the “interaction” between lexical-semantic and
lexeme representations. Thus, once again, we find that the notion of an abstract
lemma node may be quite superfluous.
To this point I have presented various results and arguments against the
lemma-as-abstract-node hypothesis. Singly, each observation or argument may
be insufficient to reject this hypothesis since each observation may be chal-
lenged or alternative interpretations offered. However, the combined weight of
all the evidence is not easily dismissed on one pretext or another, and is
sufficient to raise profound scepticism about the explanatory value of the
hypothesis. This scepticism about the validity of the syntactic mediation
hypothesis of lemma-level representation encourages the consideration of
alternative forms of organisation of the lexical system. One such possible
alternative is considered next.

THE INDEPENDENT NETWORK MODEL OF

LEXICAL ACCESS
The Independent Network (IN) model of the lexicon assumes that lexical
knowledge is organised in sets of independent networks connected to each other
by a modality-specific lexical node. The lexical-semantic network represents
word meanings as sets of semantic properties, features, or predicates. The
lexical-syntactic network represents a word’s syntactic features such as
grammatical category, gender, auxiliary type, tense, and so on. The nodes in
this network are organised in subnetworks corresponding to the different
syntactic functions. Thus, there is a subnetwork consisting of category nodes
(N, V, etc.); one consisting of gender nodes (M, F); one consisting of auxiliary
types (be, have); and so on. Nodes within a subnetwork have inhibitory links
since they are in competition. The P- and the O-lexeme networks consist of the
modality-specific representations of lexical items (more specifically, lexical
stems). Nodes in these networks are also linked inhibitorily since they are in
competition.
The production of a word involves the following sequence of events. A
selected lexical-semantic representation propagates activation toward the lexi-
6
cal-syntactic and the P- and O-lexeme networks . Not all syntactic features can
be activated by the semantic network. For example, with the exception of

6
The lexical-syntactic network also receives input from outside the lexical system—from
sentence generation mechanisms. However, these inputs will not be considered here.
 LEXICAL ACCESS 195

natural, gender-marked words (e.g. uomo [man] in Italian), gender features do

not receive activation from the semantic network. However, grammatical
category and verb tense features, for example, do receive activation from the
semantic network (see footnote 4, p. 186). Under normal circumstances the
7
activation of syntactic nodes from the semantic network is not sufficient for a
grammatical feature to reach threshold. Selection of the full set of grammatical
features of a word requires the prior activation and selection of the modality-
specific lexical node.
Activation and selection of a modality-specific lexical form (P- and O-lex-
emes) results in activation of its associated phonological and orthographic
properties. Thus, in this model, the selection of grammatical features typically
occurs temporally prior to the selection of the specific phonological and
8
orthographic content of a word . However, since the selection of the lexeme
node does not depend on the prior selection of its associated syntactic features,
the phonological and orthographic content of the lexeme nodes may, under
special circumstances (e.g. TOT states, brain damage), become available
independently of their grammatical features. A schematic representation of the
IN model is shown in Fig. 3.
The IN model shares many properties with other models of the lexical
system: lexical-semantic information is represented independently of syntactic
and word form representations as in the models proposed by Bock and Levelt
(1994), Dell (1990), and Roelofs (1992); lexical-semantics is componential as
in Butterworth (1989) and Dell (1986), but unlike Garrett (1992), Jescheniak
and Levelt (1994), and Roelofs (1992); it is a forward activation model as in
Butterworth (1989) and Roelofs (1992), but unlike Dell (1986) and Stemberger
(1985); however, unlike all previous models, the activation from the selected
lexical-semantic representation spreads simultaneously and independently to
the lexical-syntactic and the word-form networks. Furthermore, unlike these
earlier models it does not postulate a modality-neutral lemma representation;
instead, it only postulates direct links between lexical-semantic representations
and modality-specific (phonological and orthographic) lexical representations.
The latter representations may be called phonological lexemes (P-lexemes) and
orthographic lexemes (O-lexemes), if we wanted to stress the modality-specific
nature of the representations, but they could also be called phonological lemma
(P-lemma) and orthographic lemma (O-lemma), if we wanted to stress the fact

7
We can think of the effect of the activation propagated from the semantic network to the
syntactic feature network as a form of priming of the target features that will eventually be selected
when additional activation is provided by the lexeme node.
8
The fact that the model assumes that syntactic nodes reach threshold before lexeme-level
representations does not mean that subjects should be able to “report” syntactic information more
rapidly than word form information. Whereas the production of word forms is a natural function
of the lexical system, the metalinguistic task of reporting syntactic features may be quite difficult
and slow.
196 CARAMAZZA

PANEL A

FIG. 3A. Panel A (see also Panel B, Fig. 3B, opposite) presents a schematic representation of the
Independent Network model showing the relation among semantic, syntactic, and lexical form repre-
sentations.

that they are semantically and syntactically specified lexical representations. I

will use the terms P-lexeme and O-lexeme.
The IN model can readily account for the results we have reviewed in this
paper. It accounts for the contrast between word exchanges and sound ex-
changes by assuming that lexical access occurs in two separate stages: the first
stage involves the selection of a modality-specific, syntactically and semanti-
cally specified representation; the second stage involves the selection of the
lexeme’s phonological (orthographic) content. The general features of the TOT
phenomenon are also explained by assuming that lexical access occurs in two
 LEXICAL ACCESS 197

FIG. 3B. Panel B (see also Panel A, Fig. 3A, opposite) shows a more detailed representation of the
model. The O-lexeme network is not shown in this figure to avoid excessiv e crowding. The flow of
information is from semantic to lexeme and syntactic networks and then on to segmental information.
N=noun; V=verb; Adj=adjective; M=masculine; F=feminine; CN=count noun; Ms=mass noun. Dotted
lines indicate weak activation. Links within a network are inhibitory.

stages. The more specific results concerning the nondependence of phonologi-

cal information on the prior retrieval of syntactic features (Caramazza &
Miozzo, 1997; Miozzo & Caramazza, in press a) follow naturally from the
assumption that the selection of a lexeme does not guarantee the selection of
the full set of its associated syntactic features. And the results of Miozzo and
his collaborators (Badecker et al., 1995; Miozzo & Caramazza, in press)
concerning the performance of Dante, which showed that gender and auxiliary
198 CARAMAZZA

type information was consistently available in the face of a complete inability

to provide information about the phonological features of inaccessible noun
and verb lexemes, are explained by assuming a deficit in accessing the
phonological content of the correctly selected lexeme representations.
The IN model naturally accounts for the pattern of dissociations and asso-
ciations of semantic errors across modalities of output. Semantic errors can
arise either from damage to the lexical-semantic level or from damage in
accessing lexeme representations. As for the other models considered here, in
the IN model, damage to the lexical-semantic level should result in quantita-
tively and qualitatively similar semantic errors in all lexical processing tasks.
This pattern of performance has been documented in subjects with dementing
disorders (e.g. Chertkow & Bub, 1990; Warrington, 1975) as well as in
subjects who have sustained extensive focal brain damage (e.g. Hillis et al.,
1990).
More pertinent to the present discussion are those cases of semantic errors
restricted to one modality of output. This pattern of performance follows
naturally from two characteristics of the IN model: the assumption that mean-
ings are componential, and the assumption that a selected lexical-semantic
representation activates in parallel the lexemes of all words that share semantic
features with the selected meaning (Caramazza & Hillis, 1990; Dell, 1986;
Hillis & Caramazza, 1995). A consequence of these assumptions is that under
normal circumstances the cohort of activated lexemes is defined semanti-
cally—all the entries that share semantic features. In the eventuality of a
problem in accessing the target lexeme, either because of damage directly at
the lexeme level or because of damage to the connections from the semantic
network to the lexeme level, the lexeme with the highest activation will be
selected. This lexeme has a high probability of being semantically related to
the target response.
The proposed structure of semantic/lexeme interaction in the IN model also
naturally explains the pattern of contrasting semantic errors across oral and
written naming in some brain-damaged subjects. On the assumption of damage
to both the P- and O-lexeme networks, resulting in the temporary inaccessibility
of some lexemes (typically low-frequency items), the most active lexeme in
each network will be produced in response to the activation from the selected
lexical-semantic representation. The selected lexemes will be semantically
related to the target response. However, since the lexeme networks are activated
independently of each other, and directly from the selected lexical-semantic
representation, there is no expectation that the selected lexemes in the two
networks will always be the same (although they are more likely than not to be
the same if such factors as frequency and semantic similarity determine the
selection of the lexeme). In other words, a natural consequence of the functional
architecture and processing dynamics of the IN model is precisely the pattern
of performance observed in subjects WMA (Miceli et al., this issue) and PW
 LEXICAL ACCESS 199

(Rapp et al., this issue) in conditions where access to the P- and O-lexeme
networks is impaired.
Finally, in order for the IN model to account for the selective deficits for
words of one grammatical class in only one modality of output, we must make
an additional assumption about the organisation of lexical knowledge in the
brain. We must assume either that syntactic information about different gram-
matical classes is represented in different areas of the brain or that lexeme
representations are organised by grammatical class in different parts of the
9
brain . We could, then, explain the patterns of spared and impaired production
of nouns, verbs, and function words as the result of selective damage to either
one of the spatially segregated networks. Some similar assumption is needed
by all models of lexical production.
It would seem that the IN model can readily and naturally account for the
full pattern of results discussed in this paper. However a problem remains to
be addressed. In the IN model, much of the burden for explaining the observed
patterns of semantic errors in brain-damaged subjects is carried by the assump-
tion that meanings are componential—it is because of the componentiality
assumption that a semantic representation activates multiple lexemes. This
assumption has not gone unchallenged over the years (e.g. Fodor, Garrett,
Walker, & Parkes, 1980). Levelt (1989, 1992) has recently raised a further
objection in the form of what he calls the “hyperonym” problem: Why don’t
people speak in hyperonyms (i.e. produce “animal” instead of “dog,” “furni-
ture” instead of “table,” “artefact” instead of “furniture,” and so on) if meanings
are componential? This is a non-trivial problem that must be addressed by
proponents of componential theories of meaning. Indeed, as Levelt (1992) has
correctly noted, a simple, unadorned componential theory of lexical semantics
is fatally flawed. Here I propose a solution to this problem within a parallel
activation, componential theory of meaning.

A SOLUTION TO THE HYPERONYM PROBLEM

Levelt (1992, p. 6) states the hyperonym problem as follows: “When lemma
A’s meaning entails lemma B’s meaning, B is a hyperonym of A. If A’s
conceptual conditions are met, then B’s are necessarily also satisfied. Hence if
A is the correct lemma, B will (also) be retrieved.” Thus, if a speaker intended
to produce “dog,” the set of semantic features selected at the level of the
semantic network would fully satisfy not only the word “dog” but also “ani-
mal.” Why, then, don’t speakers say animal when they intend to say dog?
Levelt (1992) has proposed two solutions. One solution, within a compo-
nential account of meaning, is to adopt what he has called the “principle of

9
An implication of the latter assumption is that there are independent lexeme representations
for homonyms such as to play/the play (the position that was taken in Caramazza & Hillis, 1991).
200 CARAMAZZA

specificity.” This principle states that given that the activation conditions for
multiple lexemes are satisfied by the selected lexical-semantic representation,
speakers select the most specific lexeme. Thus, speakers intending to say
“table” will select “table” instead of “furniture”; and speakers intending to say
“furniture” will select “furniture” instead of “artefact.” However, as Levelt
notes, it is not obvious how to implement such a principle in current network
models of language production.
The other solution proposed by Levelt is to give up altogether the assumption
of componentiality and adopt a holistic conception of semantics, where word
meanings are represented by lexical-concept nodes and the set of labelled
connections among the concept nodes (e.g. Collins & Quillian, 1969). On this
view, there is a node in conceptual memory corresponding to every lexical entry
in the language. These nodes are directly connected to their corresponding
lemma representation in a one-to-one fashion (as already discussed). Thus, the
conceptual node DOG would be connected to its lemma “dog,” and the
conceptual node ANIMAL would be connected to its lemma “animal.” This
solution to the hyperonym problem may face other problems, however (see
Bierwisch & Schreuder, 1992, for discussion).
There is another solution possible within the IN model that does not require
the implementation of the “principle of specificity.” The solution is quite
simple, and follows naturally from the assumptions of componentiality and
parallel activation that form part of the basic structure of the IN model. All that
needs to be assumed is that the amount of activation passed onto the next level
by any one “feature” is a weighted proportion of the number of selected
features. So, for example, if the meaning of a word is represented by 10 features,
the amount of activation passed on by each feature is (roughly) 1/10th of the
amount of activation propagated from the lexical-semantic network to the
10
lexeme level . A further assumption is that the amount of activation normally
needed by the activated lexemes to reach threshold is the full unit of activation
propagated from the lexical-semantic network. Consequently, the lexeme most
likely to reach threshold (ignoring other factors such as resting state levels of
different lexemes) will be the one that receives activation from all the selected
semantic features —the only one that receives the full complement of activation
propagated from the lexical-semantic level. The hyperonym lexeme will, by
definition, receive activation from only a fraction of the selected features and
therefore not enough to reach threshold (see example in Fig. 4). On this view,
it is not even necessarily the case that the most active alternative lexeme to the
target lexeme will be the hyperonym; a very similar co-hyponym (e.g. dog,
given that cat has been selected at the lexical-semantic level) could reach a
greater level of activation than the hyperonym (animal).

10
I say roughly because the amount of activation propagated by each feature may also be
weighted by its importance to the meaning of the word.
 LEXICAL ACCESS 201

FIG. 4. Example of activation levels reached by various P-lexemes in response to the selection of the
semantic representation CAT.

The proposed solution to the hyperonym problem does not solve the inverse
problem also associated with componential theories of meaning—the “hypo-
nym problem.” If people intend to say “animal,” why don’t they sometimes say
“dog” or “elephant” instead? The solution proposed for the hyperonym problem
could lead to hyperonym errors—producing “dog” instead of “animal”—since
the amount of activation reaching the hyponym (dog) would be the same as that
reaching the intended target (animal). So, why don’t people regularly make
hyponym errors? To solve this problem we must make an additional assump-
tion: The maximum amount of activation contributed by a single link to a node
is a direct function of the number of links that feed into the node. Specifically,
we assume that the maximum activation contributed by any one connecting link
to a lexeme node that has N links feeding into it will be 1/Nth the total activation
202 CARAMAZZA

needed for selection of a lexeme node (or some other arbitrary value just greater
than threshold). In this way, the activation reaching the lexeme “dog” when the
semantic representation ANIMAL has been selected will necessarily be smaller
than the activation reaching the lexeme “animal.”

CONCLUSION
In this paper I have been concerned with whether it is necessary to postulate a
modality-neutral level of lexical representation (lemma) that is intermediate
between lexical-semantic representations and modality-specific lexical repre-
sentations. I have argued that there is no compelling reason for postulating such
a level of representation and that, in fact, its postulation has undesirable
empirical consequences. I have also attempted to articulate a hypothesis about
the relations among lexical-semantic, syntactic, and word form representations
that is compatible with the facts of cognitive neuropsychology, normal speech
errors and experimental evidence with normal subjects.
Two facts, in particular, have been stressed in this criticism of the lemma-
as-abstract-node hypothesis: the existence of brain-damaged subjects with
selective grammatical class deficits restricted to either oral or written produc-
tion, and the existence of subjects who make semantic errors only in speaking
or only in writing. The reason for stressing the importance of these facts is
because the peculiar nature of these dissociations allows us to specify relatively
precisely the locus of functional damage in these subjects. Thus, we can be
confident that the locus of damage in subjects who only make semantic errors
in speaking, say, must directly concern the inaccessibility of P-lexeme repre-
sentations. The reason for this conclusion is simple: As the subject can produce
the target response correctly in the other modality of output—writing, in this
example—we must conclude that the lexical-semantic level of representation
is undamaged. Furthermore, since the errors are well-formed lexical substitu-
tions, we must conclude that lexical-phonological knowledge is also undam-
aged. The only remaining possible locus of damage is at the level of the
connections leading to the activation of lexeme representations. And since a
consequence of this damage is the production of semantic errors, we are invited
to infer that the damaged relation concerns a mapping between meaning and
lexical forms—that is, between lexical-semantic representations and P- and
O-lexeme representations. This conclusion is incompatible with the syntactic
mediation hypothesis of lexeme access proposed by Bock and Levelt (1994),
Dell (1990), Jescheniak and Levelt (1994), and Roelofs (1992).
Another important source of evidence for the organisation of the lexical
system concerns the observation that selective grammatical class deficits, such
as difficulties in producing verbs, can be restricted to either oral or written
production. Here again we can rule out damage to the lexical-semantic level
because the subjects could produce verbs correctly in the unaffected output
 LEXICAL ACCESS 203

modality. Furthermore, the fact that the modality-specific deficit was restricted
to only one class of words allows us to exclude a post-lexeme deficit as the
cause of the observed impairment. We are thus invited to draw the following
inferences: lexical-semantic and grammatical information are independent
since we can damage one without affecting the other; and syntactic and word
form information are also independent of each other for the same reason.
The two conclusions reached on the basis of the neuropsychological evi-
dence find further support from research with neurologically intact subjects. It
has been found that access to a word’s lexical-semantic representation does not
guarantee access to its syntactic features, and that access to the word’s
phonological features can occur independently of access to its syntactic features
(at least for gender; Caramazza & Miozzo, 1997; Miozzo & Caramazza, in press
a). Also, in a study involving the analysis of priming effects in object naming
and gender decision tasks (Jescheniak & Levelt, 1994), it has been found that
access of a word’s lexeme facilitated a second retrieval of that lexeme but not
access of its syntactic features. Thus, experimental evidence with normal
subjects converges with that from neuropsychological investigations in show-
ing that access of lexeme representations does not depend on the prior selection
of (all) the grammatical features of a word.
The evidence reviewed here finds a plausible account within a model of
lexical access that shares many features with current theories of the lexical
system but differs from them in some important respects. The central assump-
tions of the Independent Network (IN) hypothesis are:
1. The lexical-semantic, syntactic, and modality-specific form repre-
sentations of a word are independently stored in separate networks.
2. P-lexeme and O-lexeme representations are independently activated by
semantic representations.
3. Lexical-semantics is componential and itactivates in parallel the syntactic
nodes and the P- and O-lexemes, but whereas the former nodes only
receive enough activation to be primed, the lexeme nodes can receive
enough activation for independent selection.
4. The selected lexical-semantic representation activates in parallel all the
lexemes of words that share semantic features with the selected lemma.
5. Activation from selected lexeme converges on the grammatical features
already primed by activation from the lexical-semantic network.
The fact that there is a direct link between the lexical-semantic and the lexeme
levels provides a natural explanation for the occurrence of semantic errors only
in writing or in speaking; the fact that syntactic features are represented
autonomously allows a natural explanation for the occurrence of selective
deficits of grammatical classes in only one modality of output.
There are several empirical and theoretical issues that have not been ad-
dressed here. For example, what is the time course of activation of different
204 CARAMAZZA

levels of representation? And, what are the dynamics (discrete vs. continuous;
strictly forward vs. forward and backward propagation) of activation and
selection of representations at different levels of processing? These issues are
currently the focus of concerted experimental and theoretical analysis (e.g. Dell
& O’Seaghdha, 1991; Levelt et al., 1991). It remains to be seen whether closer
scrutiny of the IN hypothesis in light of these other issues will confirm the
optimistic conclusion reached here.

REFERENCES
Alajouanine, T., & Lhermitte, F. (1960). Les troubles des activités expressives du langage dans
l’aphasie. Leurs relations avec les apraxies. Revue Neurologique, 102 , 604–629.
Allport, D.A. (1985). Distributed memory, modular systems and dysphasia. In S.K. Newman &
R. Epstein (Eds.), Current perspectives in dysphasia . Edinburgh: Churchill Livingstone.
Allport, D.A., & Funnell, E. (1981). Components of the mental lexicon. Philosophical
Transactions of the Royal Society of London B, 295, 397–410.
Andreewsky, E., & Seron, X. (1975). Implicit processing of grammatical rules in a classical case
of agrammatism. Cortex, 11, 379–390.
Assal, G., Buttet, J., & Jolivet, R. (1981). Dissociations in aphasia: A case report. Brain and
Language, 13, 223–240.
Badecker, W., Hillis, A.E., & Caramazza, A. (1990). Lexical morphology and its role in the writing
process: Evidence from a case of acquired dysgraphia. Cognition, 34, 205–243.
Badecker, W., Miozzo, M., & Zanuttini, R. (1995). The two-stage model of lexical retrieval:
Evidence from a case of anomia with selective preservation of grammatical gender. Cognition,
57, 193–216.
Basso, A., Taborelli, A., & Vignolo, L.A. (1978). Dissociated disorders of speaking and writing
in aphasia. Journal of Neurology, Neurosurgery, and Psychiatry, 41 , 556–563.
Beauvois, M.F., & Dérousné , J. (1979). Lexical or orthographic agraphia. Brain, 104, 21–49.
Berndt, R.S., Mitchum, C.S., Haendiges, A.N., & Sandson, J. (1997). Verb retrieval in aphasia.
1. Characterizing single word impairments. Brain and Language, 56, 68–106.
Bierwisch, M., & Schreuder, R. (1992). From concepts to lexical items. Cognition, 42 , 23–60.
Bock, J.K. (1982). Toward a cognitive psychology of syntax: Information processing contributions
to sentence formulations. Psychological Review, 89, 1–47.
Bock, K., & Eberhard, K.M. (1993). Meaning, sound and syntax in English number agreement.
Language and Cognitive Processes, 8, 57–99.
Bock, K., & Levelt, W. (1994). Language production: Grammatical encoding. In M.A.
Gernsbacher (Ed.), Handbook of psycholinguistics. San Diego, CA: Academic Press.
Breedin, S.D., Saffran, E.M., & Coslett, H.B. (1994). Reversal of the concreteness effect in a
patient with semantic dementia. Cognitive Neuropsychology, 11, 617–660.
Brown, A.S. (1991). A review of the tip-of-the-tongue experience. Psychological Bulletin, 109 ,
204–223.
Brown, R., & McNeill, D. (1966). The “tip-of-the-tongue” phenomenon. Journal of Verbal
Learning and Verbal Behavior, 5, 325–337.
Buckingham, H.W., & Kertesz, A. (1976). Neologistic jargon aphasia. Amsterdam: Swets &
Zeitlinger.
Burke, D., MacKay, D.G., Worthley, J.S., & Wade, E. (1991). On the tip of the tongue: What
causes word finding failures in young and older adults? Journal of Memory and Language,
30, 542–579.
Butterworth, B. (1979). Hesitation and the production of verbal paraphasias and neologisms in
jargon aphasia. Brain and Language, 8 , 133–161.
 LEXICAL ACCESS 205

Butterworth, B. (1980). Evidences from pauses in speech. In B. Butterworth (Ed.), Language

production. Vol. 1: Speech and talk. London: Academic Press.
Butterworth, B. (1989). Lexical access in speech production. In W. Marslen-Wilson (Ed.), Lexical
representation and process. Cambridge, MA: MIT Press.
Butterworth, B., & Beatty, G. (1978). Gesture and silence as indicators of planning in speech. In
R. Campbell & G.T. Smith (Eds.), Recent advances in the psychology of language: Formal
and experimental approaches. New York: Plenum Press.
Caramazza, A. (1988). Some aspects of language processing revealed through the analysis of
acquired aphasia: The lexical system. Annual Review of Neurosciences, 11, 395–421.
Caramazza, A., Berndt, R.S., & Basili, A.G. (1983). The selective impairment of phonological
processing: A case study. Brain and Language, 18, 128–174.
Caramazza, A., Berndt, R.S., & Hart, J. (1981). “Agrammatic” reading. In F.J. Pirozzolo & M.C.
Wittrock (Eds.), Neuropsychological and cognitive processing in reading. New York:
Academic Press.
Caramazza, A., & Hillis, A.E. (1990). Where do semantic errors come from? Cortex, 26, 95–122.
Caramazza, A., & Hillis, A.E. (1991). Lexical organization of nouns and verbs in the brain. Nature,
349, 788–790.
Caramazza, A., & Miceli, G. (1989). Orthographic structure, the graphemic buffer and the spelling
process. In C. von Euler, I. Lundberg, & G. Lennerstrand (Eds.), Brain and reading. London:
Stockton Press.
Caramazza, A., & Miceli, G. (1990). The structure of graphemic representations. Cognition, 37,
243–297.
Caramazza, A., & Miozzo, M. (1997). The relation between syntactic and phonological knowledge
in lexical access: Evidence from the `tip-of-the-tongue’ phenomenon. Manuscript, Harvard
University, Cambridge, MA.
Chertkow, H., & Bub, D. (1990). Semantic memory loss in dementia of Alzheimer’s type. Brain,
113, 397–417.
Collins, A.M., & Quillian, M.R. (1969). Retrieval time from semantic memory. Journal of Verbal
Learning and Verbal Behavior, 8, 240–247.
Coltheart, M., Patterson, K.E., & Marshall, J. (Eds.) (1980). Deep dyslexia . London: Routledge
& Kegan Paul.
Cubelli, R., & Perizzi, S. (1996). Selective loss of grammatical gender of nouns in aphasia. Paper
presented at the Istituto di Psicologia del CNR, Roma, December 10.
Damasio, A.R., & Tranel, D. (1993). Nouns and verbs are retrieved with differently distributed
neural systems. Proceedings of the National Academy of Sciences of the United States of
America, 90, 4857–4960.
Daniele, A., Silveri, M.C., Giustolisi, L, Colosimo, C., & Gainotti, G. (1994). Evidence for a
possible neuroanatomical basis for lexical processing of nouns and verbs. Neuropsychologia,
32, 1325–1341.
Dell, G.S. (1986). A spreading activation theory of retrieval in sentence production. Psychological
Review, 93, 283–321.
Dell, G.S. (1988). The retrieval of phonological forms in production: Test of predictions from a
connectionist model. Journal of Memory and Language, 27, 124–142.
Dell, G.S. (1990). Effects of frequency and vocabulary type on phonological speech errors.
Language and Cognitive Processes, 4, 313–349.
Dell, G.S., Juliano, C., & Govindjee, A. (1993). Structure and content in language production: A
theory of frame constraints in phonological speech errors. Cognitive Science, 17 , 149–195.
Dell, G.S., & O’Seaghdha, P.G. (1991). Mediated and convergent lexical priming in language
production: A comment on Levelt et al. (1991). Psychological Review, 98, 604–614.
Dell, G.S., & Reich, P.A. (1981). Stages in sentence production: An analysis of speech error data.
Journal of Verbal Learning and Verbal Behavior, 20, 611–629.
206 CARAMAZZA

De Renzi, E., & di Pellegrino, G. (1995). Sparing of verbs and preserved but ineffectual reading
in patient with impaired word production. Cortex, 31, 619–636.
Ellis, A.W. (1985). The production of spoken words: A cognitive neuropsychological perspective.
In A.W. Ellis (Ed.), Progress in the psychology of language, Vol. 2. London: Lawrence
Erlbaum Associates Ltd.
Ellis, A.W., Miller, D., & Sin, G. (1983). Wernicke’s aphasia and normal language processing:
A case study in cognitive neuropsychology. Cognition, 15, 111–144.
Fay, D., & Cutler, A. (1977). Malapropisms and the structure of the mental lexicon. Linguistic
Inquiry, 8, 505–520.
Fodor, J.A., Garrett, M.F., Walker, E.C.T., & Parkes, C.H. (1980). Against definitions. Cognition,
8, 263–367.
Fromkin, V.A. (1971). The non-anomalous of anomalous utterances. Language, 47 , 27–52.
Gainotti, G. (1976). The relationship between semantic impairment in comprehension and naming
in aphasic patients. British Journal of Disorders of Communication, 11 , 57–61.
Gardner, H., & Zurif, E. (1975). Bee but not be: Oral reading of single words in aphasia and alexia.
Neuropsychologia, 13, 181–190.
Garrett, M.F. (1975). The analysis of sentence production. In G. Bower (Ed.), The psychology of
learning and motivation: Advances in research and theory. Vol. 9. New York: Academic Press.
Garrett, M.F. (1976). Syntactic processes in sentence production. In R. Wales & E. Walker (Eds.),
New approaches to language mechanisms. Amsterdam: North-Holland.
Garrett, M.F. (1980). Levels of processing in sentence production. In B. Butterworth (Ed.),
Language production. Vol. 1: Speech and talk. London: Academic Press.
Garrett, M.F. (1988). Processes in language production. In F.J. Newmeyer (Ed.), The Cambridge
survey of linguistics. Vol. 3: Biological and psychological aspects of language. Cambridge,
MA: Harvard University Press.
Garrett, M.F. (1992). Disorders of lexical selection. Cognition, 42, 143–180.
Goodglass, H. (1976). Agrammatism. In H. Whitaker & H.A. Whitaker (Eds.), Studies in
neurolinguistics, Vol. 1. New York: Academic Press.
Harley, T.A. (1984). A critique of top-down independent levels of speech production. Evidence
from non-plan-internal speech errors. Cognitive Science, 8, 191–219.
Henaff Gonon, M., Bruckert, R., & Michel, F. (1989). Lexicalization in an anomic patient.
Neuropsychologia, 27, 391–407.
Hier, D.B., & Mohr, J.P. (1977). Incongruous oral and written naming. Brain and Language, 4 ,
115–126.
Hillis, A.E., & Caramazza, A. (1995). Representation of grammatical categories of words in the
brain. Journal of Cognitive Neuroscience, 7, 396–407.
Hillis, A.E., Rapp, B., Romani, C., & Caramazza, A. (1990). Selective impairments of semantics
in lexical processing. Cognitive Neuropsychology, 7, 191–243.
Howard, D., & Franklin, S. (1988). Memory without rehearsal. In T. Shallice & G. Vallar (Eds.),
The neurological impairment of short-term memory. Cambridge: Cambridge University Press.
Howard, D., & Franklin, S. (1989). Missing the meaning? A cognitive neuropsychological study
of the processing of words by an aphasic patient. Cambridge, MA: MIT Press.
Humphreys, G.W., Riddoch, M.J., & Quinlan, P.T. (1988). Cascade processing in picture
identification. Cognitive Neuropsychology, 5, 67–103.
Jescheniak, J.D., & Levelt, W.J.M. (1994). Word frequency effects in speech production: Retrieval
of syntactic information and of phonological form. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 20, 824–843.
Jonsdottiir, M.K., Shallice, T., & Wise, R. (1996). Phonological mediation and the graphemic
buffer disorder in spelling: Cross language differences? Cognition, 59, 169–197.
Jones, G.V., & Langford, S. (1987). Phonological blocking in the tip of the tongue state. Cognition,
26, 115–122.
 LEXICAL ACCESS 207

Kay, J., & Ellis, A. (1987). A cognitive neuropsychological case study of anomia. Brain, 110,
613–629.
Kay, J., & Hanley, R. (1994). Peripheral disorders of spelling: The role of the graphemic buffer.
In G.D.A. Brown & N.C. Ellis (Eds.), Handbookof spelling: Theory, process and intervention.
Chichester, UK: Wiley.
Kempen, G., & Huijbers, P. (1983). The lexicalization process in sentence production and naming:
Indirect election of words. Cognition, 14, 185–209.
Kohn, S.E., Wingfield, A., Menn, L., Goodglass, H., Berko Gleason, J., & Hyde, M. (1987).
Lexical retrieval: The tip of the tongue phenomenon. Applied Psycholinguistics, 8 , 245–266.
Koriat, A., & Lieblich, I. (1974). What does a person in a “TOT” state know that a person in “don’t
know” state doesn’t know. Memory and Cognition, 2 , 647–655.
Lecours, A.R., & Rouillon, F. (1976). Neurolinguistic analysis of jargonaphasia and
jargonagraphia. In H. Whitaker & H.A. Whitaker (Eds.), Studies in neurolinguistics, Vol. 2.
New York: Academic Press.
Levelt, W.J.M. (1989). Speaking: From intention to articulation. Cambridge, MA: MIT Press.
Levelt, W.J.M. (1992). Accessing words in speech production: Stages of processing and
representations. Cognition, 42, 1–22.
Levelt, W.J.M., & Maassen, B. (1981). Lexical search and order of mention in sentence
production. In W. Klein & W.J.M. Levelt (Eds.), Crossing the boundaries of linguistics.
Studies presented to Manfred Bierwisch. Dordrecht: Reidel.
Levelt, W.J.M., Schriefers, H., Vorberg, D., Meyer, A.S., Pechmann, T., & Havinga, J. (1991).
The time course of lexical access in speech production: A study of picture naming.
Psychological Review, 98, 122–142.
Lhermitte, F., & Dérousné, J. (1974). Paraphasies et jargonaphasies dans le langage oral avec
conservation du langage écrit . Genèse des neologismes. Revue Neurologique, 130, 21–38.
Link, K., & Caramazza, A. (1994). Orthographic structure and the spelling processes. In G.A.
Brown & N.C. Ellis (Eds.), The handbook of normal and distributed spelling development:
Theory, process and intervention. Chichester, UK: Wiley.
MacKay, D.G. (1982). The problem of flexibility, fluency, and speed-accuracy trade-off in skilled
behavior. Psychological Review, 89, 483–506.
MacKay, D.G. (1987). The organization of perception and action: A theory for language and
other cognitive skills. New York: Springer.
Martin, N., Weisberg, R.W., & Saffran, E.M. (1989). Variables influencing the occurrence of
naming errors: Implications for models of lexical retrieval. Journal of Memory and Language,
28, 462–485.
McCarthy, R., & Warrington, E.W. (1985). Category-specificity in an agrammatic patient: The
relative impairment of verb retrieval and comprehension. Neuropsychologia, 23, 709–727.
McCloskey, M., Badecker, W., Goodman-Shulman, R.A., & Aliminosa, D. (1994). The structure
of graphic representations in spelling: Evidence from a case of acquired dysgraphia. Cognitive
Neuropsychology, 11, 341–392.
Meyer, A.S. (1990). The time course of phonological encoding in language production: The
encoding of successive syllables of a word. Journal of Memory and Language, 7, 269–282.
Miceli, G., & Caramazza, A. (1988). Dissociation of inflectional and derivational morphology.
Brain and Language, 35, 24–65.
Miceli, G., Silveri, M.C., Villa, G., & Caramazza, A. (1984). On the basis for agrammatic’s
difficulty in producing main verbs. Cortex, 20, 207–220.
Miceli, G., Silveri, M.C., Nocentini, U., & Caramazza, A. (1988). Patterns of dissociation in
comprehension and production of nouns and verbs. Aphasiology, 1(2), 351–358.
Miozzo, M., & Caramazza, A. (in press a). The retrieval of lexical-syntacti c features in
tip-of-the-tongue states. Journal of Experimental Psychology: Learning, Memory, and
Cognition.
208 CARAMAZZA

Miozzo, M., & Caramazza, A. (in press b). On knowing the auxiliary of a verb that cannot be
named: Evidence for the independence of grammatical and phonological aspects of lexical
knowledge. Journal of Cognitive Neuroscience.
Patterson, K.E. (1982). The relation between reading and phonological coding: Further
neuropsychological observations. In A.W. Ellis (Ed.), Normality and pathology in cognitive
functions. London: Academic Press.
Patterson, K.E., & Shewell, C. (1987). Speak and spell: Dissociations and word-class effects. In
M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive neuropsychology of language. London:
Lawrence Erlbaum Associates Ltd.
Perfect, T.J., & Hanley, J.R. (1992). The tip-of-the-tongue phenomenon: Do experimenter-
presented interlopers have any effect? Cognition, 45, 55–75.
Rapp, B., Benzing, L., & Caramazza, A. (1995). The modality-specific representation of
grammatical category. Paper presented at the Meeting of the Psychonomic Society, Los
Angeles, CA.
Rapp, B., & Caramazza, A. (1997). The modality-specific organization of grammatical categories:
Evidence from impaired spoken and written sentence production. Brain and Language, 56,
248–286.
Roelofs, A. (1992). A spreading-activation theory of lemma retrieval in speaking. Cognition, 42,
107–142.
Roelofs, A. (1993). Testing a non-decompositional theory of lemma retrieval in speaking:
Retrieval of verbs. Cognition, 42, 107–142.
Rubin, D.C. (1975). Within-word structure in tip-of-the-tongue phenomenon. Journal of Verbal
Learning and Verbal Behavior, 14, 392–397.
Sartori, G., Job, R., & Coltheart, M. (1993). The organization of object knowledge: Evidence from
neuropsychology. In D.E. Meyer & S. Kornblum (Eds.), Attention and performance, XIV.
Hillsdale, NJ: Lawrence Erlbaum Associates Inc.
Schriefers, H. (1993). Syntactic processes in the production of noun phrases. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 19, 841–850.
Schriefers, H., Meyer, A.S., & Levelt, W.J.M. (1990). Exploring the time course of lexical access
in language production: Picture-word interference studies. Journal of Memory and Language,
29, 86–102.
Semenza, C., Mondini, S., & Capelletti, M. (in press). The grammatical properties of mass nouns:
An aphasia case study. Neuropsychologia.
Shattuck-Hufnagel, S. (1987). The role of word onset consonants in speech production planning:
New evidence from speech error patterns. In E. Keller & M. Gopnik (Eds.), Sensory processes
in language. Hillsdale, NJ: Lawrence Erlbaum Associates Inc.
Starreveld, P.A., & La Heij, W. (1996). Semantic interference, orthographic facilitation, and their
interaction in naming tasks. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 21, 686–698.
Stemberger, J.P. (1985). An interactive activation model of language production. In A.W. Ellis
(Ed.), Progress in the psychology of language, Vol. 1. Hillsdale, NJ: Lawrence Erlbaum
Associates Inc.
Stemberger, J.P. (1990). Wordshape errors in language production. Cognition, 35 , 123–158.
Vigliocco, G., Silverberg, N.B., Garrett, M.F., & Antonini, T. (1995). Retrieval of grammatical
information in anomia and during tip-of-the-tongue states. Poster presented at the Academy
of Aphasia Conference, San Diego, CA.
Warrington, E. (1975). The selective impairment of semantic memory. Quarterly Journal of
Experimental Psychology, 27, 635–657.
Zingeser, L.B., & Berndt, R.S. (1988). Retrieval of nouns and verbs in agrammatism and anomia.
Brain and Language, 39, 14–32.