Marc van Oostendorp
Schwa in phonological theory


Schwa: an overview of its phonological and phonetic properties; underspecification theory and phonological emptiness; projection theory.

Phonological properties

The ideal theory of phonology should eventually be able to account for the following facts.

  1. Schwa iff stressless:
    • Underlying schwa avoids stress
      • Stress is on the antipenultimate syllable in Indonesian, except if this syllable contains schwa:
        cári `search for'
        bicára `speak'

        b@rí `give'
        gám@lan `Indonesian orchestra'
      • Stress is always on the vowel immediately preceding (underlying) schwa in Dutch/German. (Stress can be on any of the last three syllables in a word, but not if the final syllable contains schwa):
        páanama `Panama'
        pyjaacute;ma `pyjamas'
        chocolá `chocolate'

        ballád@ `ballad'
      • Schwa is unstressed in most dialects of Yupik. E.g. in General Central Yupik, stress is generally on the second syllable of the word. If this syllable contains schwa, this vowel is deleted:
        /qayapigkani/ [qayáápixkani] `his own future kayak'

        /qan@qa/ [qánqa] `my mouth'
    • Stressless vowels get reduced to schwa
      • English
        /fonoloji/ [f@nól@Ji]
    • Words with schwa do not satisfy minimality requirements (Indonesian, French, Germanic).
  2. Schwa -> simple syllable structure:
    • In certain languages, schwa cannot occur in a closed syllable:
      • French
        /aS@te/ [a.S@.te] `to buy'
        /aS@te/ [a.SEt] `(I) buy'
      • Bodø Norwegian:
        /Enkl/ [EnkEl] `simple' (cf. Oslo [Enkl]; southern dialects [Enk@l])
        /lakn/ [lakEn] `sheet' (cf. Oslo [lakn]; southern dialects [lak@n])
    • In certain languages, schwa does not allow a complex onset:
      • Dutch:
        *[papavr@] [papav@r] (cf. French [papavr(@)]
        */kadr@/ [kad@r] (cf. French [kadr(@)])
    • Unsolved question: are there languages which allow for CCVC, but have CV syllables only?
    • No cases are known of languages where vowels turn to schwa because they are in an open syllable.
  3. Schwa cannot occur after a void
    • In many languages, schwa cannot occur at the beginning of a word
      • French, Dutch, German, Norwegian, Indonesian, Breton.
      • In Dutch, underlying schwa cannot occur at the beginning of the word, and unstressed vowels cannot get reduced to schwa
        /plezir/ [pl@zír] `fun'

        /eGal/ [egál] (*[@Gal]) `even'
    • Schwa cannot occur after /h/
      • In Dutch, schwa after /h/ behaves in exactly the same way as schwa at the beginning of a word
        /plezir/ [pl@zír] `fun'

        /helas/ [helas] (*[h@las]) `alas'
  4. Schwa is a vowel without features
    • Schwa (the epenthetic vowel) is very sensitive to vowel harmony
      • In Turkish, the 'high schwa' [I] is the epenthetic vowel. Turkish has many roots that behave as exceptional with respect to vowel harmony; but [I] is always harmonic.
        *** *** `'

        /prEns/ [pirEns] `prince'
      • In Winnebago, the epenthetic vowel seems to be the only vowel that assimilates to neighboring vowels (for that reason, we cannot actually be sure about the identity of the vowel; schwa never surfaces as schwa):
        *** *** `'

        /prEns/ [pirEns] `prince'
    • In elision contexts, if one of the two vowels disappears, it is usually schwa:
      • Dutch
        /elit@/+/Er/ [elitEr] `snobbish'
        /d@/+/Arm@/ [dArm@] (poetic) `the poor'
        /sla/+/@n/ [slan] `to beat'

Phonetic properties

Understanding the basic facts about the phonetics of schwa might help us understand the phonology better.
  1. Schwa is central
  2. Schwa is 'targetless'
    • Vi__Vi contexts
  3. Schwa is extremely short

Underspecification theory

Ways of representing an empty vowel:

I. Empty prosodic position a. Empty mora
b. Empty X-slot
c. Empty nucleus
II. Empty segmental position Empty root node ([-cons])
III. Neutral position 'Relaxation position'

Projection theory

In my view, prosodic structure is a projection of segmental structure. I use X-bar theory to represent this notion of projection, but that is actually not essential. It is more important that I use constraints of the following type:

PROJECT(F,PP): (F a feature, PP a prosodic position)
If segment S dominates F, S should be the head of a PP
PROJECT(PP,F): (F a feature, PP a prosodic position)
If segment S is the head of a PP, it should dominate the feature F

On the one hand, these constraints are supposed to imply that (underlying) schwa cannot occur in a stressed position: it does not dominate any feature, therefore according to PROJECT(PP,F), it cannot be the head of any type of projection. On the other hand, if a full vowel is in a non-head position of a PP (say, a foot) it is in conflict with PROJECT(F,Ft).

PROJECT(PP,F) and PROJECT(F,PP) can be seen as constraint schemes, defining families of constraints, just like ALIGN(V,X,Y,Z) or the MAX and DEP families of Correspondence Theory.

In order to understand how this works, here's a miniature analysis of