Ordering Restrictions between Affixes

Abstract

Abstract A language possesses a limited set of affixes; and, of all possible combinations of those affixes, only a relatively small number exists. This gives rise to questions about the nature of the restrictions on affix combinability. In the literature, grammatical and extragrammatical restrictions have been reported. The former refer to the levels of grammar (phonology, morphology, semantics, syntax), the latter concern statistical or psycholinguistic information (or both). Phonological restrictions rely on regular phonology. Morphological restrictions use morphological information and, as a rule, produce fixed combinations of two or three affixes. Semantic restrictions are cognitive in nature and refer to scopal relationships or compositionality of meaning (or both). Syntactic restrictions are due to mirroring of syntactic derivations. Two types of morphological organization, layered and template morphology, provide the canvas for affix order analyses. Layered morphology operates with hierarchically organized layers and derives morphological structure step by step, which implies syntactic ordering and semantic compositionality. Layered morphology may produce variable orders of the same morphemes. Template morphology is flat and the order of affixes in a template is stipulated. Templates derive fixed combinations of affixes and are incompatible with variation. Although a clear distinction between layered and template morphology is made at a theoretical level, there is hardly a language that falls under only one of the two types. All grammatical restrictions on affix ordering are compatible with both layered and template morphology, and no language's affixation system can be explained with a single affix‐ordering principle. As regards the length of the affix sequences examined in different studies, affix order analyses are usually based on combinations of two and three affixes. Bigrams and trigrams appear to be the most appropriate sequences for learning affix ordering patterns, both for humans and for machines.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 59 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.