The role of morphological markedness in the processing of number and gender agreement in Spanish: an event-related potential investigation

ABSTRACT

Current morphological theory assumes that feature values, such as masculine and feminine or singular and plural, are asymmetrically represented. That is, one member of the opposition (e.g. feminine for gender, plural for number) is assumed to be marked, and the other one, unmarked. The present study examines how these asymmetries impact agreement resolution in Spanish. Agreement was manipulated between a noun acting as head of a relative clause and an adjective located inside the relative clause (e.g. catedral que parecía inmensa “cathedral that looked huge”). Half of the nouns were feminine (marked) and the other half, masculine (unmarked). Half of the nouns were used in the plural (marked) and the other half, in the singular (unmarked). Twenty-seven Spanish native speakers read 240 sentences while their brain activity was recorded with electroencephalography and performed a grammaticality judgment. Results showed that both number and gender violations elicited a central-posterior P600, a component associated with syntactic repair, and a late anterior negativity, argued to reflect working memory costs. Only the P600 was affected by markedness. It started earlier for violations where the mismatching feature was marked. Moreover, it was larger for errors where the mismatching feature was marked, although this amplitude modulation only emerged for number, possibly due to differences in how number and gender cues were realized (i.e. both masculine and feminine showed overt inflection, but singular was uninflected relative to plural). These results suggest that the parser is sensitive to markedness asymmetries in the course of online processing.

KEYWORDS:

• P600
• markedness
• agreement
• number and gender

Acknowledgements

The authors thank Robert Fiorentino, Alison Gabriele, and Ianthi Tsimpli for their feedback on different aspects of this project. We also thank Pilar Gray-Carlos and María de Mier for their help in recruiting participants, Dave Miller for his help with data collection, Steve Politzer-Ahles for his help in preparing the manuscript, and three anonymous reviewers for their insightful comments and suggestions. .

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Harley and Ritter (2002) and Corbett (2000, p. 154) suggest that, for languages with a two-way number distinction, this pattern (i.e. plural being marked with respect to singular) holds true cross-linguistically. For grammatical gender, the issue is more complicated, since there are languages with a two-way gender system where nouns are not classified along the masculine vs. feminine divide. Corbett (1991, pp. 206–207, 220) discusses at least three languages with a masculine–feminine system where the feminine value appears to be the unmarked one (Khasi, Maasai, and Zayse), suggesting that markedness asymmetries for gender may be language-specific.

2. For a review of different theoretical approaches to agreement, the reader is referred to Acuña-Fariña (2009). See also Corbett (2006); Pollard and Sag (1988); Shieber (1986); and Wechsler & Zlatić (2003).

3. This is the case for nouns where no gender-based biological distinction is possible.

4. Data from one additional participant who performed more than two standard deviations below the mean in all ungrammatical conditions of the judgment task were excluded from analysis.

5. Caffarra and Barber (2015) used ERP to compare gender violations with opaque vs. transparent nouns, and found no differences, but they did not manipulate markedness, suggesting that there is no clear account of how markedness impacts gender agreement when distributional cues are absent. In addition, Caffarra and Barber (2015) only tested determiner–noun gender agreement (i.e. linearly and structurally local), but it is possible that the lack of distributional cues impacts gender agreement when the agreeing elements are not adjacent and belong to different phrases.

6. These tests were carried out as follow-ups, and we thank an anonymous reviewer for this suggestion. To measure cloze probability, the sentences were truncated right before the critical word and participants were asked to complete the sentences with the first adjective that came to their mind. Results from 40 native speakers of Spanish who did not participate in the EEG experiment revealed that the cloze probability of the sentences ranged from 0 to .55, and that the mean cloze per condition was low (for gender, conditions 1 and 4: .022; conditions 7 and 10: .025; for number, conditions 1 and 7: .024; conditions 4 and 10: .022) mainly because the sentences were not designed to be semantically constraining. Crucially, the difference across conditions was not significant (p > .2 for both comparisons). In order to measure the plausibility of the sentences, they were truncated at the critical adjective. A total of 40 native speakers of Spanish who did not participate either in the cloze test or in the EEG task rated how plausible the sentences were on a Likert scale ranging from 1 (completely implausible) to 7 (completely plausible). The mean plausibility per condition was above 5.5 for all conditions (for gender, conditions 1 and 4: 5.8; conditions 7 and 10: 5.55; for number, conditions 1 and 7: 5.71; conditions 4 and 10: 5.62) and the difference between conditions was not significant (p > .3 for both comparisons).

7. Follow-ups to the marginal Agreement by Anterior–Posterior interaction showed no effects.

8. A related finding in Coulson et al.’s study is that P600 amplitude was reduced when the proportion of morphosyntactic violations was higher. In this respect, our design included an equal number of “singular noun + plural adjective” and “plural noun + singular adjective” violations, suggesting that – in terms of proportion – one violation type was not any more salient/probable than the other.

9. We thank an anonymous reviewer for this suggestion.

Additional information

Funding

José Alemán Bañón was supported by a postdoctoral fellowship from the Spanish Ministry of Economy and Competitiveness [grant number FPDI-2013-15813].