The causative project investigates the acquisition of causatives in human language and the influence of causative expressions on causal understanding in cognition in an interdisciplinary manner. In this project, we bridge corpus study and experimental work and look at the acquisition questions from a cross-linguistic perspective.
Causality is a ubiquitous feature of human cognition and communication. Causality is universally expressed in human languages but it is particularly intriguing for acquisition, because causes of events cannot be directly perceived but rather need to be inferred. From early on in ontogeny, infants focus on the causality of events. What remains unclear is how children learn about the interpretation and expression of such causal events in becoming a native speaker of their language. The goal of this project is to contribute to this question. Specifically, we ask the following:
Our corpus study is aimed at exploring the facilitative factors for the learning process to answer the first question, with a special focus on the role of child-directed speech; our experimental study looks at the second question, adopting a cross-linguistic approach by comparing Swiss German with Turkish language. The combination of these two studies allows us to not only discover the causal expressions in naturalistic speech, but also test crucial variables with controls, so as to shed light on the big picture of causal acquisition.
In corpus studies, we explore the trajectory of causative learning and ask specifically how child-directed speech impacts the learning process. The first two studies examine the acquisition of lexical causatives with English language as an example, while the third study looks at the development of both lexical and morphological causatives in Turkish language.
In study 1, we investigated the inference of lexical causatives (e.g. break, kill, raise) in child-directed speech in English. Lexical causative lack explicit markers and the casual meaning is embedded, thus making the inference task particularly difficult. While most theories stress that syntactic structure is needed to extract word meanings, recent studies have found a prominent role of the statistical patterns of raw speech stream. We therefore asked how infants can infer the causative meaning from child-directed speech, or more specifically, what cues from the linguistic context children can rely on in this process. We simulated the process of meaning extraction with a distributional learning algorithm applied to large text collections of three speech genres, namely child-directed speech, adult conversation, and as a control, written language. The results show that meaning can be successfully extracted from child-directed speech by using simple linear co-occurrences of neighboring words, without relying on any structural information. By contrast, meaning extraction in adult-directed speech fundamentally requires additional access to syntactic structure. We hence suggest that child-directed speech is ideally shaped for a learner who has not yet mastered syntactic structure.
A publication on this study can be found here: https://doi.org/10.31234/osf.io/zmktd
In study 2, we looked at the development of semantics for English children and investigated the dynamics in semantic learning. It has been evidenced that adaptation between child and child-directed speech occurs at various linguistic levels. Evidence for semantic adaptation is, however, somewhat lacking, chiefly due to the insufficient representations of semantics. We employed a similar approach as in the first study, namely a distributional algorithm, to establish semantic networks for causative meaning. We thereby examined the developmental trajectory of causative semantics in both child and child-directed speech, and asked if there is adaptation regarding semantic information. We find that both speech genres show similar developmental trends, with child-directed speech showing a short-time lag. Concretely, child-directed speech responds to the dynamics in child speech and child speech in turn benefits from the adjustments. We therefore conclude that semantic adaptation occurs between these two speech genres over the course of acquisition.
A video of the talk of this study at AMLaP2020 can be found here: https://mediaup.uni-potsdam.de/Play/Chapter/220
We switched our scope to a comparison between different causative constructions in study 3. Both lexical and morphological causatives exist with prominence in Turkish language, rendering a complex system of causative learning. Particularly, argument ellipsis, namely the omission of participants in verbal constructions, prevalently occurs in Turkish, and object ellipsis is largely displayed. This poses a great challenge for Turkish children to acquire lexical causatives, of which the learning crucially relies on argument structures. We therefore asked how ellipsis occurs in child-directed speech and whether it impacts the acquisition of causatives. We modeled the development of lexical causatives, with morphological causatives as a comparison. The results show that object ellipsis is generally more frequent for lexical causatives, though the frequency drops over time. Despite the large proportion of elliptical expressions, children are able to develop the variability of their use of lexical causatives. This resonates with our findings in the first study that statistical cues in child-directed speech are rich and can go beyond fixed frames to facilitate semantic generalization.
Combining the experimental and corpus work, one study focuses on the acquisition of the causative morphology in Turkish. First, with a language judgment task, we show that Turkish-learning children acquire a rule-based understanding of the most productive causative marker in Turkish, -DIr, only towards the end of the 4th year. Next, an examination of a small-scale corpus of naturalistic speech shows that the variation of the morphological causatives with -DIr in the child-directed speech remains low with no significant increase across the ages of 0;7-3;0. Together, they suggest a relatively delayed generalization of a rule for the morphological causatives possibly because the input does not accumulate sufficient exemplars to generalize from until about 4 years of age.
Two other experimental studies focus on the interface between causal language and causal event conceptualization by a crosslinguistic examination of Turkish and Swiss-German. These languages differ in terms of the means with which causality is expressed. Turkish has both lexical (e.g. kesmek to cut) and morphological causatives (e.g. yemek to eat vs. yeDIRmek to feed) to express causality, whereas Swiss-German has lexical causatives but lacks morphological causatives. Beside their semantics, lexical causatives provide a cue for causality only when they are used in transitive syntactic frames, whereas morphological causatives provide a local cue (i.e., the causative marker) that specifically denotes causality. A further difference is that Turkish allows pervasive argument ellipsis in contrast to Swiss-German.
One of the experimental studies looked at whether the cross-linguistic differences in the expression of causality influence the conceptualization of events as causal or non-causal by Turkish-learning and Swiss-German-learning 3-to-4-year-olds. Children were asked to map pseudo-verbs onto one of a causal (e.g., girl touches and immediately turns on a toy) or a non-causal scene (girl stops short of touching a toy which turns on after a delay). Swiss children were tested in a condition where they heard sentences with a syntactic cue, namely, transitive frame (e.g., Where did the girl gorp the toy?). Turkish children were tested in either of three conditions: i. Only syntactic cue (e.g., Where did the girl gorp the toy?), ii. Only morphological cue (e.g. Where did the girl gorp-CAUS?), or iii. Both syntactic and morphological cue (e.g., Where did the girl gorp-CAUS the toy?). Results showed that both Swiss children and Turkish children in all conditions reliably inferred causal meanings for the pseudo-verbs, and they did so to a similar extent. This suggests that children as early as the age of three are able to use the available linguistic cues in their languages to derive causal event representations with no specific effect of verbal morphology or argument ellipsis.
Here is a link to a poster presentation of this study: https://bit.ly/32tR2Cl
The other experimental study asked, first, whether the linguistic cues to causality facilitate children’s representation of events as causal for untypical causal events which would have otherwise been perceived as non-causal, and second, whether the cross-linguistic differences facilitate this causal event representation to differing degrees. 2;6-to-3;6-year-old Swiss-German-learning and Turkish-learning children were shown videos of untypical causal events where an object agent (e.g., a cube) operated on a wooden apparatus (e.g., pressed a button on a box) upon which a distant object activated (e.g., a toy car started to wiggle and make a sound). Depending on the condition, the videos were accompanied by either a non-causal sentence (e.g., Here is a cube and a car) or a causal sentence, that is, a sentence with pseudo-verbs that exemplified a different causative construction. Swiss children were tested in a non-causal condition or a lexical condition (e.g., The cube gorped the car). Turkish children were tested in either a non-causal condition, lexical condition, or a morphological condition (e.g., The cube gorp-CAUS-ed the car). At test, children were presented with the exact same apparatus and toy that they saw in the video along with a distractor toy, and verbally prompted to act on the apparatus, using the same construction of their condition (e.g., if lexical- Now you can gorp the car). Results showed that children of both language groups performed successfully (i.e., acted on the apparatus and expectantly looked at the toy) in causal condition(s) compared to the non-causal condition, confirming the first question. However, Turkish children in both of the causal conditions performed similarly to Swiss children, failing to confirm the second question. Yet, only Turkish children in the morphological condition performed successfully in more trials than expected by chance, thereby providing some evidence that the verbal morphological causality cues might provide an advantage in causal event conceptualization.
In conclusion, experimental studies 1 and 2 showed that both young Turkish-learning and Swiss-German-learning children were able to use language-specific cues to derive casual event representations. Cross-linguistic differences did not seem to influence the extent of these representations, at least at the ages of 2;6-3;11. Interestingly, Turkish children were able to make use of the verbal morphological cues, even in the absence of syntactic cues, to infer causal meanings as well as their Swiss counterparts made use of syntactic cues, as early as 3 years of age. Further, study 2 hinted for an advantage of the verbal morphological cues in representing events as causally linked. However, the joint study showed that Turkish children abstracted a rule for the morphological causatives only toward the end of age 4. Although this seems a bit contradictory, it might be that Turkish children, at age 3, already have a phonological sensitivity for the causative marker -DIr such that when they hear it already attached to a pseudo-verb they conjecture a causal meaning. However, an adult-like morphological operation that is flexible enough to expect an explicit attachment of the -DIr to a pseudo-verb when the event is causativized might require a longer developmental maturation.
Sabine Stoll, Prof. Dr. (PI)
Moritz Daum, Prof. Dr. (PI)
Ebru Ger (PhD student)
Guanghao You (PhD student)
Aylin Küntay, Prof. Dr. Koç University
Tilbe Göksun, Prof. Dr. Koç University