Modeling How Suffixes Are Learned in Infancy.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Linguistics

Background:

• Infants demonstrate early awareness of grammatical morphemes, such as the English suffix -s, by 6 months of age.
• Understanding the cognitive mechanisms behind early morphological acquisition is a key challenge in developmental linguistics.

Purpose of the Study:

• To computationally model and understand the process by which infants acquire knowledge of suffixes.
• To evaluate the effectiveness of different distributional learning models in replicating observed suffix acquisition orders.

Main Methods:

• Trained a set of distributional learning models on a corpus of child-directed speech.
• Developed a best-performing model that segments utterances into words to create a proto-lexicon, then identifies affixes using Terminus Frequency and Parse Reliability heuristics.
• Compared this model's performance against models that attempt affix discovery directly within utterances.

Main Results:

• The best-performing hierarchical model successfully mimicked the experimental order of suffix acquisition with appropriate parameter settings.
• Models that did not prioritize word segmentation failed to match acquisition orders and generated spurious affix candidates.
• Specifically, models without a word-discovery step extracted implausible pseudo-affixes from high-frequency words.

Conclusions:

• Affix learning in infants appears to be a hierarchical process, with prior word segmentation being essential.
• Computational modeling supports the hypothesis that word discovery precedes and facilitates affix discovery in early language acquisition.
• The findings highlight the importance of a structured, word-based approach for computational models aiming to replicate infant language learning capabilities.