Bruce Hayes: A Phonologists View of the Past Tense Controversy

Value of the Information & Strength of the Argument

The talk provides valuable insights into computational morphology and the past tense debate. Hayes presents a novel algorithm that combines rule-based and statistical approaches, addressing limitations of both connectionist and pure rule-based models. The argumentation is solid, grounded in empirical data and statistical validation. He carefully explains the methodology and justifies design choices, such as using wug tests over corpus splitting. The presentation is nuanced, acknowledging the model’s limitations and the need for further refinement. The value lies in its contribution to understanding how speakers generalize morphological patterns, with implications for language acquisition and cognitive science.

Scientific Rigor, Source Quality, Title Accuracy

Hayes demonstrates scientific rigor by referencing key works (Rumelhart & McClelland, Pinker & Prince, Prasada & Pinker) and describing his methodology in detail. The sources are appropriate and well-integrated. The title accurately reflects the content, focusing on the past tense controversy from a phonologist’s perspective. The talk is well-structured and the claims are supported by empirical results. However, the presentation is from 2000, so some references may be outdated, but the core arguments remain relevant. The adéquation between title and content is strong.

The title accurately reflects the content: a phonologist's perspective on the past tense debate, focusing on a computational model of morphological learning.

8/10

The talk is by a leading phonologist, presents a well-argued computational model, and includes empirical validation with human subjects. However, it is a dated presentation (2000) and lacks recent references.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction by Luigi Rizzi, presenting Bruce Hayes.
• 02:00Hayes acknowledges collaborators and introduces the past tense controversy.
• 05:00Overview of connectionist model by Rumelhart & McClelland and Pinker & Prince's critique.
• 10:00Description of the dual-mechanism model and connectionist responses.
• 15:00Introduction of Hayes and Albright's approach: minimal generalization and rule learning.
• 20:00Explanation of micro-rules and generalization process.
• 25:00Discussion of rule evaluation: raw reliability and adjusted reliability.
• 30:00How the model takes a wug test and compares to human judgments.
• 35:00Application to English past tense: correlation with Prasada & Pinker's data.
• 40:00Application to Italian verb conjugation and cluster analysis.

• Rumelhart & McClelland (1986) - On Learning the Past Tenses of English Verbs — Connectionist simulation of English past tense.
• Pinker & Prince (1988) - On Language and Connectionism: Analysis of a Parallel Distributed Processing Model of Language Acquisition — Critique of connectionist model and proposal of dual-mechanism.
• Prasada & Pinker (1993) - Generalisation of regular and irregular morphological patterns — Wug test data used for comparison.
• Albright & Hayes (2003) - Rules vs. Analogy in English Past Tenses: A Computational/Experimental Study — Related work by the authors.

• Albright & Hayes (2003) - Rules vs. Analogy in English Past Tenses: A Computational/Experimental Study — Published version of the model and results.
• Prasada & Pinker (1993) - Generalisation of regular and irregular morphological patterns — Human data used for comparison.

• Rumelhart & McClelland (1986) - On Learning the Past Tenses of English Verbs — Connectionist model that Hayes critiques for lacking symbolic rules.