Testing the Tolerance Principle: Children form productive rules when it is more computationally efficient

Abstract

During language acquisition, children must learn when to generalize a pattern – applying it broadly and to new words (‘add –ed’ in English) – and when to restrict generalization, storing the pattern only with specific lexical items. But what governs when children will form productive rules during language acquisition? How do they determine when a pattern is widespread enough to generalize to novel words, and when a pattern should not extend beyond the cases they have observed in their input? One effort to quantify the conditions for generalization, the Tolerance Principle (Yang, 2016), has been shown to accurately predict children’s generalization behavior in dozens of corpus-based studies. The Tolerance Principle hypothesizes that a general rule will be formed when it is computationally more efficient than storing lexical forms individually. Here we test the Tolerance Principle in two artificial language experiments with children. In both experiments, we exposed children to a language with 9 novel nouns, some of which followed a regular pattern to form the plural (-ka) and some of which were exceptions to this rule. As predicted by the Tolerance Principle, in Experiment 1 we found that children exposed to 5 regular forms and 4 exceptions generalized, applying the regular form to 100% of novel test words. Children exposed to 3 regular forms and 6 exceptions did not extend the rule, even though the regular form was still the majority token in this condition. In Experiment 2, we found that children continued to behave categorically: either forming a productive rule (applying the regular form on all test trials) or using the regular form no more than predicted by chance. We found that the Tolerance Principle can be used to predict whether children will form a productive generalization or not based on each child’s individual vocabulary size. The Tolerance Principle appears to capture something fundamental about the way in which children form productive generalizations during language acquisition.

Leaving Obligations Behind: Epistemic Incrementation in Preschool English

“…Artificial language studies show that child learners extract regular patterns from inconsistent input (Hudson Kam & Newport, 2009). A current line of work investigates how much noise can be tolerated before disrupting rule generalization (Hendricks et al, 2018;Schuler et al, 2016;Yang, 2016). In natural contexts, work on Nicaraguan Sign Language explores how younger learners introduce language systematicity (Cohort 2 and later; e.g., Senghas & Coppola, 2001).…”

Infinite use of finite means? Evaluating the generalization of center embedding learned from an artificial grammar

“…While we have designed our stimuli to make those heuristics unhelpful, it is still worth noting that here, as elsewhere, converging evidence is needed to convincingly determine what biases learners bring to language acquisition. In past work, ALL has corroborated or enhanced insights from natural language acquisition (Wonnacott, Newport, & Tanenhaus, 2008), language typology (Culbertson, Smolensky, & Legendre, 2012), and computational modeling (Schuler, Yang, & Newport, 2016), so we conclude that ALL can-and doesplay an important role in piecing together our understanding of learning biases. See Culbertson and Schuler (2019) and Morgan and Newport (1981) for further discussion of what ALL can tell us about language acquisition.…”

Theory and predictions for the development of morphology and syntax: A Universal Grammar + statistics approach

“…Together, these two principles have been used for investigating a rule, generalization, or default pattern for a variety of linguistic knowledge types (Yang, 2005;Legate & Yang, 2013;Yang, 2015;Schuler, Yang & Newport, 2016;Yang, 2016;Pearl, Lu & Haghighi, 2017;Yang, 2017;Irani, 2019;Pearl & Sprouse, 2019a). However, there isn't yet much evidence that children are capable of using the Tolerance and Sufficiency Principlesthe main support comes from the study by Schuler et al (2016), which demonstrates that 5-to 8-year-old behavior is consistent with children using these principles. Still, these principles seem like a promising statistical learning mechanism for UG+stats proposals, given their current success at predicting child behavior (more on this in the subsection on learning morphology in highly-inflected languages).…”