The production of pronominal clitics: Implications for theories of lexical access

Abstract

In three experiments we investigated the locus of the frequency effect in lexical access and the mechanism of gender feature selection. In Experiment 1, participants were asked to produce gender-marked verb plus pronominal clitic utterances in Italian (e.g., “portalo” (bring it [masculine]) in response to a written verb and pictured object. We found that pronominal clitic production is sensitive to the frequency of the noun it replaces. This result locates the effect of word frequency in lexical access at the level where a word's grammatical features are represented. In Experiments 2, 3A, and 3B we used a picture-word interference naming task and found that the gender of a distractor word does not affect the production of gender-marked clitics. This result, together with those of Experiments 3A and 3B, which show a semantic interference effect and the absence of a phonological facilitation effect in clitic production, respectively, allows the inference that the retrieval of grammatical gender is an automatic consequence of lexical node selection and not an independent selection process that operates on the principle of selection-by-activation level.

Acknowledgement

The research reported here was supported in party by NIH grant DC04542 to Alfonso Caramazza. We would like to thank the members of the Harvard Cognitive Neuropsychology Laboratory and especially Albert Costa for many helpful discussion on various aspects of the research reported here. We also thank Brad Mahon and Robin Peterson for helpful comments on an earlier version of this article.

Notes

¹Romance clitics are phonologically unstressed pronouns that can only be placed adjacent to the verb on which they depend.

²When the word following the definite determiner begins with/s + consonant/,/gn/,/z/ or a vowel, the determiner forms are lo (singular) and gli (plural).

³Another grammatical feature that must be selected as part of the information used to determine the form of pronouns is grammatical case. For example, English distinguished nominative and dative case, as in the contrasts she/her, he/him.

⁴Of course, it is possible that lexical access is frequency insensitive or that both levels of lexical representation are frequency sensitive. It is unlikely that lexical access is frequency insensitive, especially given the overwhelming evidence from aphasia, which shows that patients with intact semantic processing nonetheless show strong frequency effects in naming and other word production takes (e.g., Caramazza & Hillis, 1990). It is far less clear that frequency sensitivity would be restricted to only one level of lexical representation if, in fact, there were to be two such levels of representation.

⁵Two points should be noted here. First, it is not obvious that the error-induction parading used by Dell reveals effects of lexical processing as opposed to post-lexical processes, given that the task required subjects to prepare and hold in short-term-memory a phonological sequence, which subsequently had to be produced as fast as possible. It is possible that the effects obtained in this task reveal the strength of association among a set of phonemes—determined in part by word frequency—rather than the process of lexical node selection, Second, a strongly interactive model, such as the one proposed by Dell (1990), does not necessarily require that homophones share a lexical representation in order to obtain a homophone frequency effect. It is possible that feedback from the identical sets of phonemes shared by homophones would suffice to obtain a homophone frequency effect.

⁶The high percentage of discarded data may be attributed to the greater difficulty of this experiment with respect to the others reported here (see below). In this experiment, participants were required to produce in the correct order two clictics in response to two different input stimuli: a picture for the object clitic, a written pronoun for the dative clitic. In the other experiments, participants were only required to produce one clitic.

⁷Actually, the gender marking of adjectives in an NP only surfaces when determiners are absent. Thus compare “de rode auto” and “het rode boek” versus “rode auto” and “rood boek”.

⁸The gender congruency effect is only obtained when participants must produced gender-marked words/morphemes. Thus, for example, La Heij et al. (1998) obtained a gender congruency effect in a picture-word interference production task when Dutch subjects were required to produce det + noun phrases. When participants were required to produce only the noun, no gender congruency effect was observed. This result has been interpreted in two ways. On one interpretation, the absence of a gender congruency effect indicates that gender is not selected when bare nouns must be produced. Since the gender feature is not selected in such cases, there is no opportunity for a congruency effect (Levelt et al., 1999). An alternative interpretation explains the absence of a gender congruency effect in bare noun naming as indicating that the effect is only obtained when gender-marked phonological forms (e.g., determiners) compete for selection. And since participants are not required to produced gender-marked forms there is no opportunity for a gender congruency effect. Miozzo and Caramazza (unpublished raw data) also failed to observe a gender congruency effect when Italian participants were required to produce bare nouns in a picture-word interference naming task.

⁹In two unpublished studies, one with Italian participants (Costa & Caramazza, 2005) and one with Dutch speakers (Perdijk, Spalek, & Schriefers, 2005), that used the same experimental paradigm as Vigliocco et al. (2002) but controlled for response set size, no gender priming effects were observed.

¹⁰Although the existence of the semantic (co-ordinate) interference effect is beyond dispute (Glaser & Glaser, 1989; Glaser & Dűngelholff, 1984; La Heij, 1988; Lupker, 1979, Rosinski, 1977), the cause of the effect remains controversial. One interpretation (e.g., Roelofs, 1992; Starreveld & La Heij, 1996) assumes that it reflects the competitive nature of the lexical selection process. The claim is that the selection process is sensitive to the relative activation level of the target response: the smaller the difference in activation levels between the target and distractor lexical nodes the harder the selection decision. Given this assumption, the semantic interference effect arises because the activation level of the lexical node of a semantically related distractor is assumed to be greater than that of an unrelated distractor. This difference presumably reflects the fact that semantically related disractors received activation not only from the word stimulus but also from the semantically related target picture. Another interpretation (Costa, Alario, & Caramazza, in press; Miozzo & Caramazza, 2003) is that the semantic (coordinate) interference effect has its locus at the level of deciding which semantic representation—the target picture's or the distractor word's— should be selected for lexicalisation. However, for present purposes, the value of the semantic interference effect is as a signature effect that the diatractor word has been processed and its semantic representation has been retrieved.

¹¹We are currently exploring the possibility that clitic selection occurs early by investigating the production of proclitic forms such as “lo porta” (he brings it [mas.], literally: it [mas.] brings [second person, singular]). These proclitics are “attached” to the beginning of the verb.

¹²It could be argued that our failure to observe a gender congruency effect reflects the possibility that distractor words do not strongly activate their gender features in Italian. However, this possibility has no plausible motivation since the activation of a noun's gender is assumed to occur automatically as part of the activation of its lexical node. This assumption has received considerable empirical support since the seminal paper by Schriefers (1993), as demonstrated by the reliable gender congruency effects observed in various languages by various researchers. Thus, in the measure to which this assumption is needed to explain the existence of gender congruency effects in some languages it must also apply to Italian. Otherwise, one would have to argue, without any motivation, that Italian does not allow the automatic spread of activation that is a core feature of all current theories of lexical access.

Additional information

Notes on contributors

Alfonso Caramazza

Correspondence should be addressed to Alfonso Caramazza, Department of Psychology, Harvard University, William James Hall, 33 Kirkland St., Cambridge, MA 02138, USA. [email protected]