Planning multiple dependencies in sentence production

ABSTRACT

One of the defining properties of human language is the abundance of potentially unbounded dependencies between elements in a sentence. And yet, how speakers formulate dependencies in sentence production is still poorly understood. Here we examine the timing of verb planning in sentences involving across-the-board and parasitic gap constructions. Using a new task we call the Sentence-Word Interference task, we show that speakers plan the verb of a secondary clause before sentence onset, but selectively when producing across-the-board sentences and not when producing parasitic gap sentences. Based on this timing contrast, we argue that speakers plan verbs predominantly before the production of their dependents, but only when verbs and their dependents engage in both conceptual and direct syntactic relationships. More broadly, the current study suggests that sentence planning is constrained by syntactic relationships that are not reducible to conceptual relationships or to surface word order.

KEYWORDS:

• Syntax
• Sentence production
• Sentence-word interference
• Across-the-board extraction
• Parasitic gaps

Acknowledgments

We would like to thank Brian Dillon, Lyn Frazier, and Victor Ferreira for their comments and feedback on the earlier version of this paper. The data and analysis scripts used for this study can be found here: https://osf.io/74eqn/?view_only=d0c9852da00642699d338d04a0924eeb.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Similar cases where one of the gaps is missing in ATB constructions have been observed (Goldsmith, 1985; Kehler, 2002; Ross, 1967; Schmerling, 1972, inter alia). However, in many of these cases, the gap position does not exist, rather than it being replaced by an overt NP. Thus, something different is going on in these cases. Though we recognise these exceptions in ATB constructions, we contend that they are different from cases like PG constructions where the second gap is replaced by an overt NP.

2 Note that, under the analysis where the ATB and PG constructions are alike but both involves the null operator (Munn, 1992), it can be predicted that the timing of verb planning is late for both ATB and PG sentences. However, the timing contrast between ATB and PG sentences cannot be predicted under this analysis.

3 Note that the high error rates in Experiment 3a and 3b (and 4) are expected; ATB and PG sentences are infrequent and structurally complex, and recalling rare and complex sentences without changing their form is of course expected to be hard. Indeed, these error rates are comparable to a non-recall-based study that investigated the onset latency and production time of complex sentence production excluding responses with undesired sentence types (33% to 44%, Ferreira & Swets, 2005). However, the overall high error rates do not by themselves interfere with the logic of the current study, and we believe high error rates are simply a cost we have to pay to investigate the production of complex sentences containing infrequent structures in experimental settings.

4 The onset latency in Experiment 3b seems to be generally longer than in Experiment 3a. This could be because the distractor verbs were presented twice in Experiment 3b, once as the first verb to be read aloud right after the sentence memorisation, and once as the recall cue. Participants in Experiment 3b might have generally harder time ignoring the distractor verbs regardless of their relatedness to the target verbs, because both related and unrelated distractor verbs were recently activated for production.

5 Note that 20 ms average difference, though small, is about the average size of semantic interference effect found in typical picture-word interference task, according to the recent meta-analysis on the semantic interference effect (Bürki et al., 2020)