Children’s difficulty with raising: A performance account

ABSTRACT

This article explores English-speaking children’s acquisition of raising structures with an experiencer (e.g., John seems to Mary to be happy). We review and address previously unnoticed issues in the methodologies of existing studies testing the acquisition of raising, thus providing a more reliable picture of children’s abilities with respect to raising. We then present three experimental studies, which reveal that children’s purported difficulty with raising is significantly reduced when the experiencer is fronted to the beginning of the sentence (e.g., To Mary, John seems to be happy) and when the experiencer is pronominal (e.g., John seems to her to be happy). While the acquisition of this construction has been argued to be delayed due to grammar-internal reason, we propose an alternative account that attributes the difficulty to a performance limitation, responsible for intervention effects observed in a variety of other structures.

Funding

The research was supported by the NWO Spinoza grant “Native and Non-native Listening” awarded to Anne Cutler. Additional funding was provided by an NWO (Dutch NSF) Rubicon grant to Suzanne V. H. van der Feest and SSHRC and NSERC Grants to Elizabeth K. Johnson.

Notes

¹ Although the majority of raising predicates (e.g., tend, used to, be about, be going) do not permit an experiencer argument, a subset of raising predicates, including seem and appear, do allow such an argument (Mary in (ia)) to occur between the raising predicate and the embedded clause ([ia], with [ib] its unraised counterpart).

Table

While there is no clear syntactic evidence for or against the idea that the experiencer argument is obligatory or optional, Orfitelli (2012:111) posits the existence of an implicit experiencer argument even when it is unpronounced—based on “the native speaker intuition that a proposition cannot ‘seem’ without there being a ‘seem-ee’ to experience it.” We will proceed on this assumption without making any commitment, as the issue is not directly relevant to the present article. It should be noted, however, that patterns like (ia), where raising takes place across the experiencer, are cross-linguistically rare. In some languages, such as Icelandic, Italian, and Spanish, raising across an experiencer NP is simply forbidden (e.g., Boeckx 1999, 2008).

² This study is also presented in Hirsch (2011) as Study 5, with further analyses and expanded discussion.

³ As an aside, it appears from Hirsch’s (2011) detailed description of Hirsch, Orfitelli & Wexler (2008) that each child saw the exact same scenario eight times, each time paired with a different test sentence. This unusual experimental design raises concerns that children’s performance was affected by repeated exposure to the exact same scenario—concerns that we are unable to address conclusively, as noted by an anonymous reviewer.

⁴ This research also appears in Hirsch (2011) as Study 1, with new individual subject analyses and more discussion.

⁵ This study is also presented in Hirsch (2011) as Study 4, with further analyses and expanded discussion.

⁶ Yet, this literal seem-as-think analysis is later dismissed in Hirsch (2011), since it entails that children are not paying attention to the nonfiniteness of the embedded clause in the raised sentences, despite previous findings that suggest sensitivity to the finite/nonfinite distinction from around 18 months of age (Pierce 1992; Poeppel & Wexler 1993). Instead, Hirsch (2011) proposes an imagine-analysis where children interpret the raising sentences (iia) as (iib), by analyzing the verb seem as a raising-to-object verb, like imagine. The following examples are taken from Hirsch (2011:110).

Table

As with the think-analysis, the imagine-analysis should lead to below-chance performance on the raised condition with DR foils, since the DR foil matches this incorrect interpretation.

⁷ This point has been raised by Choe (2012:21) as well as Becker (2014:198). The latter says “the fact that thought-bubbles were used in all of the drawings, and this depiction was explained to children as indicating what a character was thinking about, could easily have biased children to a seem=think kind of interpretation of the raised seem sentences.”

⁸ There may be additional reasons for why children were ignoring the fronted experiencer in raised sentences, since those same children had no problems interpreting the fronted experiencer with unraised sentences. Hirsch (2011) does not provide an explanation for this unexpected pattern of results.

⁹ A phase is a subdomain of the syntactic derivations that proceed incrementally. Both vP and CP are identified as phases (N. Chomsky 2001).

¹⁰ Note that the UPR also predicts the delayed acquisition of verbal passives and unaccusatives, since all three of these structures purportedly involve deficient phases. Yet, as pointed out by an anonymous reviewer, there have been a number of studies that show that children comprehend and produce such structures early and easily (for passives, see Fox & Grodzinsky 1998; O’Brien, Grolla & Lillo-Martin 2006; Messenger et al. 2012; for unaccusatives, see Costa & Friedmann 2012; Friedmann 2007; Snyder, Hyams & Crisma 1994).

¹¹ Relativized Minimality (RM): In the configuration … X … Z … Y …, Y cannot be related to X if Z intervenes and Z has certain characteristics in common with X (Rizzi 1990).

¹² Minimal Link Condition (MLC): K attracts α only if there is no β, β closer to K than α, such that K attracts β (N. Chomsky 1995:311).

¹³ It should be noted that there are different accounts of intervention effects, discussion of which can be found later in this article.

¹⁴ Match items are those where the target response is “true”; mismatch items are those where the target response is “false.”

¹⁵ Although see Mitchener and Becker (2011), who find the raising predicate going (to) to be much more frequent than seem in the input to children.

¹⁶ As correctly noted by an anonymous reviewer, no significant effect of the factors considered is likely to be due to the small number of items and children per age group. For further data presentation, see Table C1 in Appendix C, which shows the results in terms of the number of children per age group in each condition.

¹⁷ We also examined how the excluded 10 children (who scored 50% overall on the Unraised condition) comprehended raising sentences. Their score on the Raised condition was 60%, which is higher than 40.9% from the earlier 11 children, but this difference did not yield significance by a mixed-effects logistic regression analysis (β = 0.35, Z value = 0.53, p = .6). There is thus no reason to think that the excluded children might have acquired the raising pattern before the nonraising pattern.

¹⁸ Fronting the experiencer phrase (thereby topicalizing it) was motivated and natural, as each story contained another potential experiencer that contrasted with the real experiencer in the test sentence.

¹⁹ For further data presentation, see Table C2 in Appendix C, which shows the results in terms of the number of children per age group in each condition.

²⁰ The overall scores of the nine excluded children were 44.4% on the Unraised condition and 66.7% on the Raised condition.

²¹ As for cognitive control, the Unraised condition was deemed to be sufficient.

²² For further data presentation, see Table C3 in Appendix C, which shows the results in terms of the number of children per age group in each condition.

²³ The purpose of this analysis was to see if the children had difficulty finding the correct antecedent of the pronoun when the gender of the pronoun was matched with the gender of the referent of the matrix subject (gender match) as in (iii).

Table

As the governing category for the pronoun him in sentence (iii) is the matrix clause, Principle B rules out a referential dependency between him and Donald. However, prior acquisition studies show that children around age 5 often violate Principle B, allowing the ungrammatical interpretation (e.g., Chien & Wexler 1990; Grodzinsky & Reinhart 1993). Because the children in the current study are in a similar age range, it is crucial to establish that their responses on the gender-match items did not result from a difficulty with Principle B. Thus, the children’s scores on gender-match items were compared with their scores on gender-mismatch items, where the gender cue prevents the pronoun from being coreferential with the local subject (e.g., Bart seems to her to be studying). No difference was observed between gender-match and gender-mismatch items on any condition, indicating that the children’s responses were not confounded with any effect related to Principle B.

²⁴ The overall scores of the five excluded children were 73.3% on the Unraised condition and 60.0% on the Raised condition.

²⁵ An anonymous reviewer (correctly) points out that Relativized Minimality or other similar frameworks allow for the possibility of pronouns and nouns having different feature sets and thus not counting as like elements. As currently stated, the AIH does not invoke such a possibility and simply states that movement may not occur over an intervening argument (nouns and pronouns alike). Were such a system to be adopted by the AIH, it might prove fruitful to explain some of our results.

²⁶ It should be noted that different theories have been put forward to explain the exact nature of intervention effects in language acquisition and processing and that not all accounts are processing-based explanations like the Dependency Locality Theory (DLT). For example, some (Belletti & Rizzi 2010; Friedmann, Belletti & Rizzi 2009) suggest that the observed intervention effects can be explained in terms of Relativized Minimality (RM; Rizzi 1990). This locality constraint states that in the configuration, … X … Z … Y …, Y cannot be related to X if Z intervenes and Z bears a certain similarity to X. Extending this idea to language acquisition, Friedmann, Belletti & Rizzi (2009) argue that children require the relevant features of X (the target) and Z (the intervener) to be entirely distinct in order to move an element over an intervener. In other words, the child’s principle allows only patterns in which there is no overlapping feature between the moved element and the intervener. Thus, extraction in object relatives like (v) is disallowed in the child system, as the relative head the boy and the intervener the girl share the same “+NP” feature, since both have a lexical head noun.

Table

In turn, the RM account predicts that if the relative head and the intervener are made sufficiently different (for example, if one is a lexical NP and the other is a pronoun), then extraction should be possible even in the child system. As such, we do not necessarily deny the possibility that the findings of the current article, namely the pronoun advantage in raising acquisition, may also be explained by a grammar-based approach such as the RM account. However, further evidence for a performance effect comes from Choe, Deen & O’Grady (2014), who found an asymmetry in the effect of NP type on children’s comprehension of raising sentences: An improvement in performance was observed when a lexical NP was raised across a pronominal experiencer (e.g., Bart seems to her to be studying) but not when a pronoun was raised across a lexical NP experiencer (e.g., He seems to Lisa to be studying). Such results are not expected under the RM account but are predicted by the DLT-based account of intervention effects.

Additional information

Funding

The research was supported by the NWO Spinoza grant “Native and Non-native Listening” awarded to Anne Cutler. Additional funding was provided by an NWO (Dutch NSF) Rubicon grant to Suzanne V. H. van der Feest and SSHRC and NSERC Grants to Elizabeth K. Johnson.