The Power of Ignoring: Filtering Input for Argument Structure Acquisition

Area of Science:

• Developmental linguistics
• Computational linguistics
• Language acquisition

Background:

• Language learning relies on accurate data representation.
• Children's developing linguistic knowledge can lead to incomplete or inaccurate parsing of speech input.
• Misparsed input, such as non-basic clauses, poses a challenge for learning verb argument structure.

Purpose of the Study:

• To investigate how children succeed in language learning despite potentially misleading input data.
• To computationally demonstrate a novel approach for learners to infer input filters without prior knowledge of problematic data.
• To address the challenge of learning verb transitivity from immature input representations.

Main Methods:

• Computational modeling of a language learner.
• Instantiating a learner that considers potential parsing errors.
• Developing a method for learners to infer an input filter without identifying specific non-basic clauses in advance.
• Testing the model on 50 frequent verbs in child-directed speech.

Main Results:

• The computational model successfully inferred a filter on input data.
• The learner avoided drawing faulty inferences by filtering out parsing errors.
• Accurate inference of verb transitivity was achieved for the majority of tested verbs.
• Demonstrated that learners can filter input without pre-identifying non-basic clauses.

Conclusions:

• Learners can overcome challenges posed by immature input representations by inferring filters.
• This approach allows language acquisition to succeed even with incomplete or inaccurate parsing.
• The study provides a novel computational solution to the problem of learning from noisy linguistic data.