📄 English Nasal-Final Prefixes and Uniform Exponence

Summary

The paper examines why three English prefixes ending in a nasal behave differently before different stem onsets. /in-/ adapts its nasal to a following obstruent (i[m]probable) and loses it before sonorants (illegible, irradiant). /un-/ stays unchanged everywhere. /en-/ assimilates before obstruents (embroil) but stays intact before sonorants (enrich). Earlier accounts relied on lexical level ordering (Kiparsky 1982) or on distinct underlying segments for each prefix (Borowsky 1986). The author argues that neither works well, since /en-/ fits neither level cleanly and Borowsky's rules have internal problems. Working in Optimality Theory with input–output correspondence, the paper adopts Kenstowicz's (1995) Uniform Exponence constraint, which penalizes variation in how a morpheme is realized. It indexes this constraint to each prefix and ranks the copies at different heights relative to nasal place agreement, phonotactic constraints, and faithfulness constraints. A single ranking then derives all three patterns. The analysis needs no level ordering, special underlying forms, or extrinsic rule order. It suggests that apparent morphological classes can reduce to how strongly each morpheme resists alternation.

Key points

• The data come largely from Borowsky (1986): /in-/ assimilates fully and degeminates, /un-/ never alternates obligatorily, and /en-/ shows a mixed pattern.
• The final hierarchy is: Uniform Exponence-/un/ >> NasPlace >> Uniform Exponence-/en/ >> *[+nasal][+sonorant], No-Geminate >> MAX-IO >> Uniform Exponence-/in/, IDENT-IO(place).
• The paper raises three objections to Borowsky:
  • She posits three underlying forms for the same nasal.
  • Her assimilation rule both fills features and changes them, depending on the level.
  • Her default "sonorant = nasal" would produce ill-formed matrices for /l/ and /r/.
• Optional fast-speech assimilation of /un-/ (u[m]bind) is deliberately excluded; only obligatory assimilation is modeled.
• Morpheme-internal nasal+sonorant clusters (e.g., "only") are tolerated because Uniform Exponence of the stem outranks the phonotactic ban.

Interesting ideas

• I1. Instead of assigning morphemes to discrete classes or levels, give each morpheme its own copy of a "stay the same" constraint and let its ranking position decide how much it bends. This turns a categorical split into a graded, inspectable parameter.
• I2. Ranked, violable constraints with strict domination give an explainable decision: each losing candidate can be reported with the single highest constraint it fails.
• I3. A hybrid case (/en-/) that fits neither existing class is used as the decisive test against a theory. Probing a model with such boundary items is a sharp evaluation tactic.
• I4. A higher-ranked faithfulness constraint on the stem explains why a phonotactic ban is ignored inside words. Local exceptions fall out of the ranking rather than needing an exception list.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (tokens): The en morphology could map surface variants such as il-/ir-/im- back to the lemma prefix in-, while recording per prefix whether alternation is expected, so un- stems are never wrongly split.
• I2 (back): During generation, allomorph or surface-form choice could be made by a small ranked-constraint evaluator, with a printable "why this form" trace instead of a lookup table.
• I3 (learn): Evaluation suites for the morphology and tokenizer could include deliberately mixed-behavior items (en- words, fast-speech forms) as hard controls that expose rules overfit to clean classes.
• I4 (tokens): Lemma and segmentation rules could prefer stem integrity over a general pattern, so words like "only" or "enrich" are not mis-split by prefix-stripping heuristics.