Locality and expectation effects in Hindi preverbal constituent ordering

Area of Science:

• Psycholinguistics
• Computational Linguistics
• Linguistic Theory

Background:

• Hindi, a predominantly SOV language, exhibits flexible word order influenced by discourse information structure.
• Cognitively grounded factors affecting Hindi constituent ordering remain underexplored.
• Existing research highlights information structure's role, but the impact of cognitive factors like dependency length and surprisal is less understood.

Purpose of the Study:

• To investigate the relative impact of Dependency Locality Theory (DLT) and Surprisal Theory on preverbal constituent ordering in Hindi.
• To test if dependency length minimization predicts syntactic choice beyond information status and surprisal.
• To determine the primary driving force behind preverbal constituent ordering in Hindi.

Main Methods:

• Generated meaning-equivalent grammatical variants of Hindi sentences by linearizing preverbal constituents from projective dependency trees in the Hindi-Urdu Treebank (HUTB) corpus.
• Utilized machine learning models incorporating information status and surprisal measures (trigram and parser-based).
• Compared the predictive power of dependency length, information status, and surprisal measures on constituent ordering.

Main Results:

• Dependency length showed a weak effect in predicting sentence variants compared to other predictors.
• Trigram surprisal significantly outperformed both dependency length and parser surprisal.
• Maximizing lexical predictability emerged as the primary factor in Hindi preverbal constituent ordering choices.
• Dependency length minimization predicted non-canonical orders in specific cases where surprisal estimates failed due to frequency bias.

Conclusions:

• Lexical predictability, particularly as captured by trigram surprisal, is the dominant factor in Hindi preverbal constituent ordering.
• While dependency length minimization plays a role, its effect is weaker than surprisal, especially in predicting non-canonical word orders.
• The presence of case markers in Hindi might override the pressure for dependency length minimization.
• Findings have implications for the information locality hypothesis and theories of language production, emphasizing local statistical biases and accessibility.