Colloquium: Rabbitometry vs. rabbitography: phonetic faithfulness and affix-by-affix differences in derived words

April 2, 3:30-5pm, Cobb 201
Adam Albright, MIT

It is well-known that affixes may differ in the extent to which derived forms deviate from the realization of the base in isolation: suffixes like -ation attract stress, condition vowel changes, and aspiration (distíll ~ dìstillátion, provó[k]e ~ pròvo[kʰ]átion), while affixes like -ery do not (distíll ~ distíllery/*dístillèry; cf. mónastèry). Numerous mechanisms have been proposed to derive this difference, assigning affixes different morphological levels (Siegel 1970; Allen 1978; Pesetsky 1979; Kiparsky 1982), different syntactic structures (Marvin 2003), different prosodic structures (Raffelsiefen 1998), or different faithfulness conditions (Benua 1997). However, surprisingly little attention has been paid to affixes with mixed properties, such as attracting stress but not conditioning consonant alternations.

One such case, described by Bermúdez-Otero (2008), concerns learned affixes like -ómeter and -ógraphy. These affixes attract main stress (speedómeter), but unlike other stress-attracting affixes, they do not preserve final clusters of nasal + voiced stop: swi[ŋ]ómeter/*swin[ŋg]ómeter. In this talk, I present data from American English showing that the phonological inconsistencies surrounding these affixes go well beyond nasal+stop clusters. For example, the suffix -ometer unexpectedly fails to condition aspiration (lea[p]ómeter/*lea[pʰ]ómeter) and does condition flapping (floa[ɾ]ómeter/*floa[tʰ]ómeter), much like a word boundary; yet unlike a word boundary, it blocks t-deletion (cou[nt]ómeter/*cou[n]ómeter). The suffix -ograph(y) similarly blocks aspiration for non-coronal stops (lea[p]ógraphy/*lea[pʰ]ógraphy), but for coronal stops, aspiration is preferred over flapping: floa[tʰ]ógraphy. For /nt/ clusters, where flapping is blocked, this results in a subtle aspiration contrast: curren[t]ómeter vs. curren[tʰ]ógraphry. These differences are a challenge for syntactic or prosodic accounts, which generally rely on a two-way distinction of presence or absence of a boundary or spell-out domain. They also pose a challenge for a stratal or level ordering account, since they appear to require many additional levels with no external motivation. I show that they follow straightforwardly from an account in which learners must learn different rankings of OO faithfulness for different affixes, based on the available set of data. Strikingly, these constraints must refer to non-contrastive phonetic properties of the base form, such as glottalization and strength of final release.