Product- and source-oriented generalizations regulate affix distribution in Polish locative adjectives: A corpus-based analysis

• Published in: Glossa (London) Vol. 9; no. 1
• Main Author: Czaplicki, Bartłomiej
• Format: Journal Article
• Language: English
• Published: Open Library of Humanities 19.11.2024
• Subjects: Research Article; source-oriented schema; subgrammar; arbitrariness; selectional restriction; probabilistic grammar
• ISSN: 2397-1835; 2397-1835
• DOI: 10.16995/glossa.10177

A large-scale corpus-based analysis of affix distribution in Polish locative adjectives provides evidence for (i) a selectional restriction formalized as a product-oriented schema, (ii) selectional restrictions formalized as source-oriented schemas, (iii) distinct native and foreign subgrammars, (iv) phonological arbitrariness of the patterns and (v) non-uniformity of outputs. A selectional restriction, which takes the form of a product-oriented segmental requirement, regulates the distribution of the available affixes. Bases that contain a final sonorant form locative adjectives using a different morphological composition than bases that do not contain a final sonorant. As a result of the application of the product-oriented schema, locative adjectives comply with the requirement irrespective of the segmental structure of their bases. Certain consonants in the input show an idiosyncratic affinity for particular affixes. Crucially, the required generalizations refer to segmental material that is absent from the output but present in the input, which must be captured by source-oriented schemas. Further, it is shown that the preferences in affix selection are different for foreign words than for native words, supporting the view that the grammar is composed of two subgrammars with different properties. Affix distribution shows a degree of phonological arbitrariness in two senses: it does not always result in phonological optimization and the segments or groups of segments that condition it often do not form phonetically natural classes. The non-uniformity of outputs is derived using a model that assumes probabilistic weighting of competing generalizations.