Umlaut

  German umlaut has been widely discussed and a variety of alternative accounts of it have been proposed. Historically, it is a process of vowel harmony which fronts back vowels in roots when a suffix with a front vowel is added. However, its synchronic status is somewhat different, both in its range of application and in its precise phonological realisation (Chapman 1994). In the domain of nominal inflection, it marks certain subclasses of nouns for plural, usually with an accompanying -e or -er suffix. Whilst those nouns which suffix -er in the plural always undergo umlaut as well, this is not true of the -e suffix. There is also a handful of nouns which undergo umlaut on one of their back vowels, but which have other back vowels which are unaffected. The class of nouns which undergoes umlaut cannot be determined by independent synchronic linguistic factors, be they phonological, syntactic or semantic. It is simply a matter of declensional class membership.

The realization of umlaut is more complex than a fronting of back vowels. As Wiese (1996a, 181-94, 1996b) points out, the mapping requires that the featural definitions permit /E/ to be viewed as the front variant of /a/, rather than the raised variant as one might expect from a neutral featural definition of the vowel system of German. The realisation of umlaut in the diphthongs is especially interesting. Contrary to what one might expect, it is not the case that both vowels in a diphthong are fronted when umlaut is applied. The diphthong /au/ has the umlauted version /Oy/, the second element being fronted, but the first being raised and rounded. Wiese (1996a, 182) proposes a rule of Rounding Assimilation to account for this apparent discrepancy, but this still leaves a question about the height of the vowels in question. The only other diphthong that occurs in German (with the exception of those which may appear in English loan words) is /ai/, and this is invariant. An elegant featural statement of umlaut phonology may well be possible, but it is not our concern here.

Like Trost (1990, 1991), we impose a complete separation between the phonological mapping and the morphosyntactic conditions that require the mapping to take place. Such a modular approach has the advantage that the very same phonological rule can be invoked (or not invoked) by quite different components of lexical description. Thus he notes the existence of contrasting triples such as Hand/Hände/handlich and Tag/Tage/täglich (Dieter Wunderlich has suggested to us that both handlich and täglich may be frozen forms). Trost himself deals with such phenomena monotonically by invoking an internally complex umlaut feature that contains a disjunction of feature structures where each disjunct contains a list of all the possible inflectional and derivational umlaut triggers (Trost 1993, 371-372). This analysis illustrates the rather high price that has to be paid for maintaining monotonicity in the lexicon. German verbs also undergo umlaut in the past tense of some strong verbs, and in the subjunctive forms of all verbs.

Reinhard (1990) provides an analysis of umlaut which resembles ours in many respects. Her representation of lexical entries is, like ours, expressed in terms of phonologically based syllable constituents, although she is describing the orthography, not the phonology. Her account, like ours, treats the realization of umlaut by means of a separate node defining the mappings from vowels to their umlauted counterparts, but she omits any reference to diphthongs. We shall discuss her definition of inflectional classes in the section on alternative declension hierarchies. Reinhard includes some derivational occurrences of umlaut, while we only address inflectional occurrences here.

Although umlaut occurs in both nouns and verbs in German, with identical phonological consequences, the morphosyntactic conditioning is different for the two parts of speech. For nouns we locate the statement of this conditioning at the Noun node (unsurprisingly). What we need to say can be glossed as follows: ``A peak in a root undergoes umlaut if (i) it occurs in a noun that is morphosyntactically marked for umlaut, (ii) it occurs in a plural form of the lexeme, and (iii) the peak occurs in the focussed syllable. Otherwise the peak is left unchanged''. Expressing this in DATR , we get:

Noun:
    <> == Word
    <phn $yll peak> == IF:<AND:<"<mor umlaut>"
                                EQ:<plur>
                                EQ:<$yll "<phn root focus>">>
                       THEN Umlaut:<"<phn $yll peak->">
                       ELSE         "<phn $yll peak->">.

Note the distinction between <peak-> and <peak> attributes where the former corresponds to the `underlying' segment and the latter to its surface realization as in the exactly analogous <tail-> and <tail> attribute pair which we drew attention to in our discussion of final consonant devoicing.

The Umlaut function itself can be defined thus (cf. Göhler 1995, 50):

Umlaut:
    <a>   == E
    <a:>  == E:
    <O>   == /
    <o:>  == |:
    <U>   == Y
    <u>   == Y
    <u:>  == y:
    <a u> == O y
    <a i> == a i
    <>    == IDEM.

This mapping defines the alternations for all of the vowels (and the diphthong) which undergo umlaut. The observant reader will have spotted that two diphthongs appear in the list of explicit arguments to the Umlaut function. But one of them, /ai/, is simply mapped to itself. And one might reasonably have expected such a mapping to have been taken care of by the IDEM identity mapping that is invoked for all the arguments not explicitly listed. However, removal of this apparently redundant equation would lead to /ai/ getting mapped to /E/ by the first equation given. This technical infelicity could be eliminated by recoding the function. But only at the cost of obscuring how it works. We have opted for transparency.