There and back again: An acquisition study

ABSTRACT

When again modifies an English goal-PP construction, the sentence is ambiguous between a repetitive and a restitutive reading. Interestingly, languages vary in whether their counterpart to English again permits a restitutive reading with goal-PP constructions (Beck 2005; Beck & Snyder 2001). This article explores how English-speaking children acquire restitutive again with goal-PP constructions, given the cross-linguistic variation. An experiment shows that 4- to 5-year-old children can understand these restitutive readings. However, examining the parental input of four children, we find that parental uses of restitutive again with goal-PP constructions are infrequent and (usually) ambiguous. We propose that in this case children can rely on more general evidence about the syntax of English goal-PP constructions, together with knowledge of a basic semantics for again, to deduce the restitutive reading.

Acknowledgments

For helpful feedback and discussion, we would like to thank Diane Lillo-Martin, Jon Gajewski, Jonathan Bobaljik, Magdalena Kaufmann, Kate Davidson, Jon Sprouse, Jeff Lidz, graduate students at UConn and audiences at UConn UMass Smith Language Acquisition Workshop (April 2013), LSA 2014 and PLC38. We also received useful comments from three anonymous reviewers and the editors at Language Acquisition, which helped to improve the form of the paper. We are also grateful to UConn Child Development Labs, Mansfield Discovery Depot, all the adult participants, all the families, and children for participating. Last but not the least, we would like to thank Corina Goodwin for recording the stimuli, Lee Prunier for helping to create a preliminary version of the experimental material, and Kate Guerrera for helping with recruitment. All errors remain our own.

Notes

¹ Speakers report that the restitutive reading is more salient if (1) contains the adverb back. It has been observed that many of the most natural examples of restitutive again in English involve adverbs like back and up, which (for unexplained reasons) render the result state of a goal-PP sentence more visible than in sentences without these adverbs (see Beck 2005 footnote 8). However, this is not a necessary condition for the restitutive reading to be available for goal-PP sentences (and as can be seen in the appendix, none of the test sentences in our experimental study contains back or up).

² As pointed out by an anonymous reviewer, speakers also allow an atelic reading for (5) in which the bottle passed under the bridge but did not end up there (it floated under the bridge and kept going). This reading is not relevant to the current study (for more details about the acquisition of atelic readings of some goal-PP constructions, see Goodrich & Snyder 2013).

³ Bilingual children may have been tested in the study, since the information about what language(s) children spoke at home was not collected. A possible improvement in future studies is to exclude bilingual children. Meanwhile, it would also be interesting to examine how bilingual children interpret again.

⁴ For restitutive again with goal-PP sentences, such as John walked to the village again (1), the sentence presupposes that John was at the village before (without necessarily having walked there). This reading was often illustrated in the literature (e.g., Beck 2005) through a context where the agent, John, was born at the destination (the village) from the beginning. Such contexts were also widely used in our experiment for the examination of children’s interpretation of restitutive again, as shown in (10). In principle, restitutive again should also be felicitous in a context where John went to the village before but through a different manner of motion than walking; for example, perhaps last time John drove to the village, but this time he walked there. In our experiment, we included one such story—see (11) in the appendix, which involves different manners of motion (flying versus swimming).

⁵ An anonymous reviewer asked whether using the word back in the stories could have cued the children to a restitutive reading before they got to the test sentence, given that back with again helps speakers find the restitutive reading more natural. We do not think this is concerning: both of the stories in (10) and (11) include, at their middle point, a sentence with back. Neither sentence uses back in the test sentence. Hence any contrast between (10) and (11) cannot be related to back.

⁶ Ideally one would place focal stress on lizard/snail in the lead-in and thereby turn the snail/lizard into a “contrastive topic”; this could highlight the issue of whether it is the snail or the lizard who better meets the description. This was not done in the present experiment, but it would be a possible improvement in future studies.

⁷ An anonymous reviewer expresses the following concern about testing participants’ interpretation of the repetitive reading: Since the repetitive reading entails the restitutive reading, how can we tell that participants are in fact interpreting test sentences like (11a) as repetitive? We believe this is the case for two reasons. First, on the plausible assumption that frequent use of an interpretation makes it easier to access, and on the further assumption that repetitive readings of again are more frequent than restitutive readings (as seems likely, given that restitutive readings are available in a much more restricted set of environments), we might expect the repetitive reading to be preferred when both are possible. Second, as will be discussed, only the repetitive reading is available when again is stressed (Fabricius-Hansen 1983; Stechow 1996; Beck 2006). In the experiment, again was not stressed in the stimuli for restitutive stories, yet it was stressed in stimuli for repetitive stories. If (child) participants are following the adult grammar, they should entertain the repetitive interpretation under a repetitive context.

⁸ However, by stressing again in the repetitive items, the restitutive reading may have been blocked there, at least for children who were following the adult grammar. In this experiment again is stressed in the test items for repetitive stories, which stacks the cards against us. This is because participants are likely to be primed for repetitive readings and may have more difficulty accessing the restitutive reading. As we will see later, however, children in our experiment still showed a high degree of facility with restitutive again despite the possible priming effect.

⁹ Both figures also indicate a drop in adults’ accuracy on true restitutive items compared with true repetitive items. Three out of 12 adults each rejected one true restitutive item, which suggests that they sometimes had a strong preference for the repetitive interpretation. This led to an accuracy of 87.5%. We speculate that the drop may stem from a processing bias based on frequency: In general, the repetitive reading is more frequent than the restitutive reading, given that restitutive readings are available in a much more restricted set of environments. On the plausible assumption that frequent uses of an interpretation make it easier to access, we speculate that the more frequent repetitive reading is more accessible than the restitutive reading. On the other hand, given that there were only two items in each condition, it is hard to make a firm conclusion based on the results here. Adding more items per condition could be another improvement of the experiment.

¹⁰ GLM modeling was performed using R (version 3.2.3; R Foundation for Statistical Computing, 2015) and the lme4 software package (Bates et al. 2015). As suggested by an anonymous reviewer, we first tried using a random-effect structure consisting of random intercepts by Item, plus random slopes (and intercepts) by Subject for the effect of Condition: (1|Item)+(Condition|Subject). Unfortunately, when Age was added as a predictor the model failed to converge. Suspecting that this was due to a random-effect structure that was too complex relative to the variance it accounted for, we tried using simple random intercepts by Item and Subject: (1|Subject)+(1|Item). This simpler structure consistently yielded convergence and was therefore adopted throughout, so as to allow direct comparison of models.We were unable to obtain convergence when we included an interaction term (Condition x Group) in the model. To get a rough estimate of the possible interaction, we submitted our data to conventional RM factorial ANOVA, using percentage of acceptance as the dependent variable. There was in fact a “significant” interaction there, although the p values are not trustworthy (given that a number of important assumptions of the ANOVA were violated by the data). When examined graphically (as in Figure 3), it appears that children were slightly less accepting than adults of the repetitive-true and restitutive-true items but were slightly more accepting than adults of the mismatch items.

¹¹ In response to a reviewer query, we also checked differences in d’ as a function of story type for children and adults, and there were no statistically reliable differences.

¹² It is difficult to set the standard for the true repetitive condition because there were only two items in this category. For rep-only knowers, we tried to stack the cards against ourselves by deciding that the child knows repetitive again if s/he gets at least one out of two true repetitive items correct. We think that this should not be too problematic, since the mismatch items also help to distinguish rep-only knowers from again droppers.

¹³ For the remaining 11 child participants categorized as “others,” their patterns were hard to identify because of the following reasons: First, there were only two true repetitive items, which makes it hard to conclude whether a child knows the repetitive reading if s/he gets one item wrong. Second, the task involved judgment on pragmatic felicity (whether the presupposition was met), which was subtle and complex, and this may have contributed noise to the data. Third, some standards of categorization may be too strict, given the complexity of the task. For instance, the standard that a child needed to get at least 9 out of 10 again items correct to be counted as adultlike excluded three children who got 8 out of 10 items correct and could be potentially adultlike.

¹⁴ The acquisition of presuppositions is an issue that goes beyond the scope of this article. As discussed in footnote 22, it is not clear what kind of evidence exactly is needed for children to acquire presuppositions. Besides, presupposition triggers of different kinds (for instance, hard versus soft presupposition triggers (Abusch 2002, among others) may have distinct learning processes and mechanisms (also see Yatsushiro’s (2007, 2008) discussion on children’s acquisition of lexical and implicated presuppositions of every and Karmiloff-Smith (1979) and Schaeffer and Matthewson (2009)’s study on children’s use of the definite determiner the).

¹⁵ An example with go to school, as shown in (i), was set aside in the analysis, because the mother (M) seemed to intend a figurative meaning of go to school, i.e., to receive education, instead of a real “goal-PP” reading.(i) Child(C)’s name: Lily; file number: 70; line 294M:you know Justine goes to school too.C:ooh.M:just like Lily and just like mommy.C:just like daddy.M:well daddy’s already done with school he doesn’t go anymore.M: it’s true it is true.M:daddy is all done with school.C:yeah.C:he goes to school more after he’s done.M:really should he go back to school again?

¹⁶ Following Bach (1986), we use eventuality as a cover term for activities, accomplishments, achievements, and states.

¹⁷ The Logical Forms presented here follow the notation of Heim & Kratzer (1998): In the logical forms an index (1 in the example) adjoins to the sister of the moved phrase, which semantically binds all variables with that index (t₁ and PRO₁ in the example). This is interpreted as predicate abstraction.

¹⁸ Without the assumption that to can be interpreted as ‘at’ in English, the restitutive reading of again for (18a) would require that there was a path of going to the village or direction toward the village that was repeated, which is incompatible with the classical context for the restitutive reading in which John was born at the village and has never left (cf. Section 1). Sufficient for our purpose in this study, (18d) is a simplification of Cresswell’s (1978) semantics of to, which involves the notion of a path derived from progress through time (see Beck 2005 for more details).

¹⁹ In (18d), g is an assignment function (see Heim & Krazter 1998) that interprets variables, in this case PRO₁.

²⁰ Crucially, Snyder (2012) assumes that GM does not give rise to a “locative” interpretation in the sense that the walking event occurs at the village. Interpretation of the locative type is done differently (see Beck 2005 for a discussion).

²¹ This explains why (7), the Spanish counterpart to float under the bridge, does not have a goal-PP reading.

²² There is a complex learnability question involved here, which goes beyond the scope of this article: How can learners tease apart the presupposed content from the asserted content of a presupposition trigger like again? This question is especially interesting given that a positive sentence containing again (iia) seems to have a similar meaning as a positive sentence containing twice (iiia). A possible answer is that children may be sensitive to the parental uses of a presupposition trigger like again in constructions where presupposition projects—for example, presupposition projects under negation, which distinguishes it from simple logical entailment. Take again as an example: If negation takes wider scope than again, (iib) presupposes that John had been late before. In contrast, (iiib) does not entail that John had been late before. For instance, it is compatible with a context in which John was not late last time but was late this time. However, (iib) is not compatible with such a context (even with negation taking narrower scope than again).(ii) a. John was late again.b. John wasn’t late again.(iii) a. John was late twice.b. John wasn’t late twice. (In fact, he was never late./In fact, he was late only once.)If children (somehow) have knowledge of this property and also pay attention to such indirect negative evidence that sentences like (iib) never occur in a context where John was not late last time but was late this time, they may be able to distinguish again and twice and deduce the presupposition component of again.