The Acquisition of Evidentiality and Source Monitoring

ABSTRACT

Evidentiality in language marks how information contained in a sentence was acquired. For instance, Turkish has two past-tense morphemes that mark whether access to information was direct (typically, perception) or indirect (hearsay/inference). Full acquisition of evidential systems appears to be a late achievement cross-linguistically. Currently, there are two distinct hypotheses about why this is so. According to the first hypothesis, the acquisition of evidentiality is delayed by conceptual factors related to source monitoring (the process of identifying and evaluating information sources). According to a different hypothesis, a substantial part of the learning difficulty comes from mapping evidential markers onto the underlying source concepts (even if these concepts are already available to the child), most likely because source concepts do not correspond to observable referents in the world. Here we tested these two hypotheses in a series of experiments comparing the acquisition of evidential morphology (Experiments 1–3) and the development of source monitoring (Experiments 4–6) in the same group of Turkish-speaking children. We found that the semantics and pragmatics of evidential morphology in Turkish are not acquired until age 6 or 7. A comparison between linguistic evidentiality and source monitoring experiments revealed that conceptual understanding of information access develops before the corresponding concepts are linked to evidential morphemes in Turkish, thereby demonstrating that mapping difficulties underlie the late acquisition of evidentiality in Turkish. Nevertheless, our data also suggest that conceptual limitations play an important role in the acquisition of evidentiality, since in both language and source monitoring direct evidence seems to be privileged compared to indirect evidence. This work has implications for the acquisition of mental-state language and the relation between children’s linguistic and conceptual development.

Acknowledgments

The authors thank Louis A. Arena, Roberta Golinkoff, Gabriella Hermon, Paul Quinn, and Satoshi Tomioka for comments on this research. They are also grateful to Gyn. Op. Dr. Zübeyde Öztürk, Özden Lülü, and Ömer Öz Lülü, Business Manager of Terakki Vakfi Schools in Istanbul, Turkey, for giving the authors the opportunity to conduct their experiments. The first author is now at Columbia University.

Funding

Anna Papafragou acknowledges support from NSF grant BCS-0749870.

Notes

^1. In general, the indirect marker can be used to communicate the speaker’s dissociation from the event, and its extended uses include surprise, distance, irony, story-telling (Johanson, 2003) and “lack of preparedness” for new information (Slobin & Aksu, 1982). There is considerable discussion about how best to capture these dimensions of evidentiality in Turkish (see Slobin & Aksu, 1982, among others).

^2. The past-tense evidential -mIş should be distinguished from its nonevidential homonym -mIş used in participial adjectives (e.g., kır -ıl- mış tabak break-Caus-Perf plate ‘broken plate’; Johanson, 2003). The indirect evidential should also be distinguished from the copula particle imiş, which cannot be added to verb stems but may follow nominals/nominalized verb stems and has indirect-evidential meaning (e.g., gül ‘rose’ + imiş = gülmüş ‘it is/was evidently a rose’; Johanson, 2003).

^3. Formal linguistic treatments of evidentiality are only beginning to emerge (see McCready, 2008, for an overview), and recent theoretical treatments of evidentiality have used different mechanisms to account for the behavior of evidential devices in different languages (e.g., presupposition: McCready & Asher, 2006; speech-act operators: Faller, 2007; probabilistic modality: McCready & Ogata, 2007). Our experimental efforts are neutral with respect to specific theories of linguistic evidentiality. Future empirical work will be needed to ascertain whether specific properties of evidential mechanisms (e.g., whether an evidential morpheme operates on speech acts or on the basic propositional content) raise additional complications for the learner mapping evidential meanings onto the corresponding linguistic forms. Future work would also have to address the possibility that evidential source might not be a semantic primitive (see Speas, 2010).

^4. Aksu-Koç (1988) reports longitudinal data where children younger than 3 appear to be producing evidentials correctly in spontaneous speech. However, given the subtlety of the distinctions involved, these data do not reveal the extent to which evidentials are truly understood (e.g., they might be interpreted as confidence markers; see de Villiers et al., 2009 on Tibetan; Robinson, 2009). Indeed, in Aksu-Koç’s elicited production data, 3-year-olds (alongside much older children) overextended the direct marker to cases of indirectly acquired information (also Papafragou et al., 2003, for a similar point on spontaneous production data from Korean reported in Choi, 1995).

^5. Ten monolingual Turkish-speaking adults also participated as controls. Adult results were at ceiling in all experiments (i.e., adults produced and comprehended evidential morphology correctly and reasoned about their own and others’ knowledge sources appropriately close to 100% of the time), so adult data will not be reported here.

^6. Inspection of the data in Figures 1 and 2 suggests that Turkish children’s productive command of evidentiality is better compared to their comprehension of evidentials. This was confirmed in an ANOVA comparing accuracy in Experiments 1 and 2: the analysis revealed an effect of Experiment (F (1, 93) = 16.93, partial η² = .03, p=.00), an effect of Age (F (2, 93) = 4.66, partial η² = .01, p = .01), and no significant interaction between Experiment and Age (F (2, 93) = 1.036, partial η² = .002, p = .36). The production-comprehension asymmetry runs contrary to the general observation that production follows comprehension in language development (also Papafragou et al., 2007, on Korean). One possible explanation is that the specific memory and metalinguistic requirements of the comprehension task make it more demanding than the production task (see also de Villiers & Johnson, 2005; de Villiers, Cahillane, & Altreuter, 2006). Alternatively, it might be that the present comprehension task places more stringent criteria on what counts as knowing evidential morphology (including evaluating evidentials from the perspective of another cognizer). Further research is required to adjudicate between these explanations.

^7. Our source monitoring experiments involved verbal instructions (as is typical in this literature, see Introduction) but did not rest on knowledge of evidential morphology. We piloted nonlinguistic versions of these experiments in which children had to point to a card with an eye or an ear to identify knowledge source. These versions proved less natural and more difficult to interpret than experiments involving verbal instructions.

^8. Since children would probably not understand the verb “infer” in Turkish, we used a complex phrase including the verb anlamak ‘understand.’ We decided to embed the verb in a phrase because our prior pilot work (Ozturk & Papafragou, 2007) showed that stand-alone use of the verb was not helpful to children.

^9. Children’s low success with the Hear items may be at least partly due to the choice of verb for reporting information acquired via conversation (dinlemek ‘listen, make a conscious effort to hear’; see Ozturk & Papafragou, 2007, for higher-accuracy data using a different verb in an earlier version of this task).

^10. It is an open question whether more implicit tasks might also reveal earlier sensitivity to the meaning of linguistic evidentiality, even though the issue arises only for the present comprehension task. We are pursuing this possibility in ongoing work.

Additional information

Funding

Anna Papafragou acknowledges support from NSF grant BCS-0749870.