• Keine Ergebnisse gefunden

Analyses of Swedish Accent Assignment .1 Complexity of Accent Assignment

PROCESSING AND REPRESENTATION OF TONES IN SWEDISH

4.1 Introduction to Tone in Scandinavian Languages

4.1.3 Analyses of Swedish Accent Assignment .1 Complexity of Accent Assignment

The phonological and morphological structure of Swedish words alone is not sufficient to predict accent correctly. This may present a challenge to language learners but even more so it challenges linguists whose goal it is to infer the underlying mental representation of these tones that allows correct accent assignment in logical and parsimonious ways. Before dealing with these approaches to accent assignment, the complexity of today’s tonal system of Central Swedish is outlined in a few examples (taken from Lahiri, Wetterlin, Jönsson-Steiner, 2005; Bruce & Hermans 1999).

As already mentioned above, monosyllabic words, and polysyllabic words with final stress are always ACC1. They do not possess the trochaic structure necessary for ACC2 to occur, as shown in (1) and (2).

1. Monosyllabic words are always ACC1:

a) 'skriv1 write – IMPERATIVE (IMP.)

b) 'bil1 car – NOMINATIVE (NOM.)

2. Words with final stress are always ACC1:

a) pi 'lot1 pilot – NOM.

Polysyllabic words with a trochaic foot can take either ACC1 or ACC2, as illustrated in (3). As explained above, a trochee consists of a stressed syllable followed by an unstressed one.

3. Trochaic stems are either ACC1 orACC2:

a) 'taxi1 taxi – NOM.

b) 'taxa2 tax – NOM.

Morphological processes can affect the tonal contours in different ways. Some processes, like the addition of the present tense suffix {-r} (+ epenthetic vowel) (as shown in (4)) or the definite article clitic (as shown in (5)), do not lead to accent changes as compared to the stem.

4. Present tense suffix {-r} does not lead to accent changes:

a) 'läs1 > 'läs-er1 read – IMP.; read – PRESENT TENSE (PRES.) b) 'tala2 > 'tala-r2 speak – IMP.; speak -PRES.

5. Definite article clitic does not lead to accent changes:

a) 'bil1 > 'bil-en1 car – NOM., DEFINITE ARTICLE (DEF. ART.) b) 'termos1> 'termos-en1 thermos – NOM., DEF. ART.

c) 'flicka2 > 'flicka-n2 girl – NOM., DEF. ART.

Other processes appear to cause accent changes in a consistent manner. For instance, ACC1 is caused by affixing a verb with an unstressed prefix (as in (7)), while ACC2 is triggered by the verb infinitive ending (see (6)), the addition of a stressed verbal prefix (see (8)) or compound formation (see (9)).

6. Verbs take ACC2 in the infinitive form. The infinitive is marked with the suffix {-a} (compare to German {-en}):

a) 'tala2-a > 'tala2 talk – INFINITIVE (INF.) b) 'kom1-a > 'komma2 come – INF.

c) 'skriv1-a > 'skriva2 write – INF.

7. Verbs with unstressed prefixes have ACC1:

a) för-'tala2-a > för-tala1 slander – INF. b) be-'tala2-a > be-'tala1 pay – INF. c) be-'kom1-a > be-'komma1 receive – INF.

8. Verbs with stressed prefixes have ACC2:

a) 'ut-'tala2 -a > 'ut-'tala2 pronounce – INF. b) 'ut-be'tala1 -a > 'ut-be'tala2 pay out – INF.

9. Compounds are always ACC25:

a) 'sko1 + 'kräm1 > 'skokräm2 shoe cream – NOM. b) 'kyrke2 + 'orgel2 > 'kyrkorgel2 church organ – NOM. c) 'aktie1 + 'marknad2 > 'aktiemarknad2 stock market – NOM.

Finally, some morphological processes, like adding the indefinite plural suffix {-Vr} (10) sometimes do and sometimes do not lead to changes in accent.

10. The indefinite plural suffix is either ACC1 orACC2:

a) 'bil1 -ar > 'bilar2 car – INDEF. PLURAL (IND. PLUR.) b) 'termos1-ar > 'termosar1 thermos – IND. PLUR.

c) 'hummer1-ar > 'humrar2 lobster – IND. PLUR. d) 'flicka2 -or > 'flickor2 girl – IND. PLUR.

Different approaches have been undertaken to handle this seemingly arbitrary behaviour of ACC1 and ACC2. Of the privative theories mentioned before, all (except Lahiri, Wetterlin & Joensson-Steiner, 2005) assumed that ACC2 is the marked member, particularly because it is the more complex tone having two peaks (though not in all dialects). This assumption, however, cannot describe the language facts without implementing additional rules, as we show below on the basis of Riad’s analysis.

4.1.3.2 Analysis by Riad (1998, 2003): Accent 2 is Lexically Marked

According to Riad’s (1998) analysis of the diachronic development of Scandinavian tone, ACC2 became the lexically marked tone because of the extra H that remained in the words after the resolution of stress clash in Old Norse, while ACC1 reflects pure intonation. Both, stems and affixes can be lexically marked for ACC2. If a word is marked, the lexical H tone will align with the stressed syllable, displacing the focus tone LH to the right (See Figure 4.2 above). In the case of compounds, the constituents need not be specified for ACC2. Compounds are

5This is only true for Central Swedish, not for Norwegian or Southern Swedish. In Norwegian, a compound is ACC2 if its first constituent is an ACC2 word. If the first word is ACC1, the compound is either ACC1 or ACC2. See Lahiri, Wetterlin & Jönsson-Steiner, 2004.

always assigned ACC2 because there are two stressed syllables in a word. The lexical H of the contour is aligned with the primary stress in the first constituent;

the second (focal) peak is associated with the last stress of the compound word.

The assumption that lexical ACC2 dominates word accent can account for a large part of the data. Nevertheless, certain phenomena cannot be easily explained.

For instance, Riad specifies the infinitive marker for ACC2 because words carry ACC2 in their infinitive form, as we can see in (6). However, in (7) one can observe that an infinitive like tala2 (7a,b), which Riad assumes to be lexically marked for ACC2, takes ACC1 when preceded by an unstressed prefix, yielding för-'tala1 / be-'tala1. In order to explain this, Riad needs a constraint, which he calls the ‘Two-Morpheme-Constraint’, stating that the lexical ACC2 information must occur within the first two morphemes of a structure in order to become a property of the whole structure. In (10) one can see another troublesome case when assuming specified ACC2: The plural ending {-ar} is assumed to carry lexical ACC2.

However, when adding {-ar} to a trochaic ACC1 word, as in (10b) 'termos1 >

'termos-ar1, ACC2 is not assigned, but rather ACC1 prevails. This problem he solves with a ‘Locality-Constraint’, postulating that lexical ACC2 can only be assigned from an inflection if it is immediately adjacent to main stress. The same problem, yet, does not apply to a case like (10c) 'hummer1> 'humr-ar >'humrar2. This is an example of a monosyllabic stem ('humr) that turns into a disyllable when a suffix is added and consequently an epenthetic vowel is inserted ('hummer1).

Epenthetic insertion of a vowel means that, as a word is uttered, the speaker inserts a vowel (usually a schwa) between two adjacent consonants. In Swedish, this happens to words ending in CONSONANT + SONORANT CONSONANT clusters (/Cl/, /Cr/, /Cn/), like 'humr, which are produced with a schwa between the word final sonorant consonant and the preceding consonant, resulting in hummer1. In the mental lexicon, these words are stored without the schwa, i.e. in the Swedish case as monosyllabic words. Their monosyllabic underlying form humr- is visible in suffixed forms like the plural humr-ar, but never in isolation. Since epenthesis is a post-lexical process and takes place after accent assignment, ACC1 in 'hummer1 is assigned to a monosyllable. Similarly, the plural suffix {-ar} docks onto a monosyllable, thus not calling the ‘Locality-Constraint’ into action.

In sum, although Riad’s analysis of lexically marked ACC2 can account for a large part of the data, he still needs to assume additional constraints in order to prevent ACC2 carrying affixes like {-a} INFINITIVE and {-ar} PLURAL from incorrectly assigning their accent to forms like be'tala1 and'termosar1.

4.1.3.3 Analysis by Bruce and Gussenhoven (1977, 1999, 2005): Both Accents are Lexically Marked

The equipollent analysis by Bruce and Gussenhoven (Bruce, 1977; Bruce &

Hermans, 1999; Gussenhoven & Bruce, 1999; Gussenhoven, 2005) assumes that a lexical HL tone is associated with both ACC1 and ACC2 words. In ACC1 words the L aligns with the stressed syllable, in ACC2 the H. Still, the factors determining the choice of accent need to be explained. According to Bruce (Bruce & Hermans, 1999), stress placement and the morphological make-up of a word are the main determiners of accent. A word with final stress is always ACC1 (e.g. pi'lot, ‘pilot’), nouns with pre-stress syllables and stem-stress on the penult (e.g. prin'sessa,

‘princess’) are – with few exceptions –ACC2. In words with stress on the first syllable and attached unstressed suffixes, the accent depends on the kind of suffix.

Some suffixes, like the plural or the infinitive suffix, are assumed to be ACC 2-inducing (see (10) and (6), respectively), others ACC1-inducing, like the definite article (see (5)) or the present tense marker (see (4)). If a post-stress syllable of a stem ends in {-e} or {-a}, it will be ACC2 (e.g. taxa2), else often ACC1 (e.g. taxi1).

Bruce (1977) tried to combine these regular patterns in a diagram for predicting word accent. It is displayed below in Figure 4.4. He has to note, however, that “the predictability of word accent by this diagram is (...) far from complete (...)(and that) the rules have to count a fair sized number of exceptions”

(p. 19). Comparing this tree to the examples above, which posed problems in Riad’s theory, this diagram correctly classifies verbs with unstressed prefixes as ACC1 ('tala2 > för-'tala1), and also correctly assigns ACC1 to the plural forms of trochaic ACC1 stems ('termos1 > 'termos-ar1).

As can be seen in the diagram, prosodic (Where is the stress placed? How many syllables are there?), morphological (Is there a suffix? What kind of suffix is it?), lexical (Is the word a noun?) and segmental (Does the word end with {-e} or {-a}?) information has to be considered in order to arrive at the correct accent. An example for an exception to Bruce’s diagram is the ACC1 noun be’talning1 (payment). It is a noun with a post- and a pre-stress syllable and stress on the penult and thus would be assigned ACC2 according to the diagram (leftmost column).

Figure 4.4: “Diagram for predicting the word accent of a Swedish non-compound word” by Bruce (1977, p. 18). The examples in italics are not given in the original literature but correspond to examples used in this text.

4.1.3.4 Analysis by Lahiri et al. (2005): Accent 1 is Lexically Specified As outlined above, theories assuming the lexical specification of ACC2 have to accept a number of exceptions to their rules and need to implement additional constraints to keep lexical accent from applying to certain word forms. In an attempt to resolve these problems, Lahiri et al. (Lahiri, Wetterlin & Joensson-Steiner, 2005, henceforth LWJ) questioned the undisputed assumption that ACC2 is lexically specified.

Simply being a complex tone does not justify or necessitate lexical specification in a phonological analysis. In their privative analysis, LWJ assume that stems and affixes can only be lexically specified for ACC1. The default accent assignment is that a word with a trochaic structure receives ACC2, while a word with no post-stress syllable can only have ACC1. The only exception to this rule are trochaic words, i.e. words that provide the necessary structure for ACC 2-assignment, but nevertheless take ACC1. These words or their affixes have to be lexically specified for ACC1. This logic of accent assignment is outlined below in Figure 4.5.

Figure 4.5: Diagram of word accent assignment according to Lahiri et al. (2005).

The examples given in italics correspond to the cases mentioned above.

In this way, all the above-mentioned examples can be accounted for. Monosyllabic words (see (1)) and words with final stress (see (2)) are assigned ACC1 by default.

They do not provide the structural requirements for default ACC2 to apply.

Therefore, they are not assumed to be lexically specified for ACC1.

Polysyllabic stems with a trochaic foot as given in (3) can be either ACC1 or ACC2. According to LWJ, these words will be assigned ACC2 by default if they are unspecified for word accent, while those specified receive ACC1. The fact that the present tense suffix {-r} (see (4)) and the definite article clitic (see (5)) do not change the stress pattern of the form they attach to is easily accounted for: They do not change the syllable structure of the words at the level of accent assignment. In the case of 'läs1 > 'läs-r1 > 'läser1 the stem is ACC1 because it is monosyllabic.

Attaching the present tense suffix {-r} leads to the still monosyllabic form 'läsr1. Only after accent assignment has taken place, the schwa is inserted by epenthesis and leads to the disyllabic form 'läser1 on the surface. No lexical specification for ACC1 is assumed in this case. Similarly, the definite article does not change accent because it is a clitic and therefore also added after accent assignment has taken place.

Further, virtually all other theories assume the infinitive suffix -a to be marked for ACC2 (see (6)). This is not necessary here. The infinitive suffix turns a monosyllabic stem into a disyllabic word as in 'kom1-a > 'komma2, thereby establishing the required trochee for ACC2 to be assigned by default. An unstressed

prefix (see (7)) always turns a verb into ACC1, as in be-'komma2 > be'komma1. Therefore, it is necessary to lexically specify unstressed verbal prefixes for ACC1.

If there is lexical specification in a word, this will dominate word accent and cannot be overridden by anything else. That is, the lexically specified prefix be1- turns the infinitive 'komma2 (which is ACC2 by default) into a lexically specified ACC1 word, be'komma1. Notice that for such cases, Riad (1998, 2003) needs to assume a repair strategy and has to accept that a marked accent does not apply in certain cases. According to LWJ (2005) there is no need for additional constraints or repair strategies and lexically specified ACC1 always prevails in non-compound words.

A final case of non-compound accent assignment to be discussed here is that of the indefinite plural suffix (see (10)). As outlined above, Riad assumes the plural suffix {-ar} to be marked for ACC2 to explain cases such as 'bil1 > 'bil-ar2, and needs to employ the ‘Locality-Constraint’ to avoid ACC2-assignment in 'termos1 > 'termos-ar1. When assuming ACC1 to be specified for trochees, then the 'bil1 > 'bil-ar2 example is again a case of an unspecified monosyllabic word becoming disyllabic by affixation and therefore being assigned default ACC2. A word like 'termos1 on the other hand is assumed to be lexically specified for ACC1 and hence remains ACC1 in the singular and plural. No specification of the plural suffix needs to be assumed. Note that trochaic ACC1 stems are assumed to be lexically specified only if they are disyllabic in their underlying lexical representation6. As explained above, words like 'hummer1 are underlyingly monosyllabic, i.e. 'humr, therefore ACC1 by default, and only surface as disyllables due to post-lexical epenthesis. As the plural suffix is added, a trochaic form is established and ACC2 is assigned by default, leading to 'humr-ar2.

Finally, we will briefly consider Swedish compounds. In Standard Swedish, compounds always receive ACC2, independent of the accents of its constituents (see (9)). This is also the explanation for ACC2 in verbs with stressed prefixes (see (8)). The stressed prefix is treated as a prosodic word, and therefore the compound rule assigns ACC2, as in 'ut-tala2 or 'ut-betala2. This analysis of stressed prefixes is agreed on by all three theories mentioned here.

Summing up, by specifying only ACC1 for stems and affixes, accent assignment in Scandinavian can be explained in a parsimonious way, without needing further rules and constraints.

6 There is one exception to the rule that anything specified for ACC1 does not become ACC2 by any process. Verbs like bok'sera1 need to be specified forACC1 because they contain a trochee. However, they turn into ACC2 nouns as the suffix {-are} is added, leading to bok'serare2. (Riad, personal communication).

4.1.4 Danish Stød

We now briefly turn to the Danish stød. Stød is not considered a tonal phenomenon. Its production is characterized by a sharp fall in pitch accompanied by a creaky sound. Nevertheless, it is a suprasegmental feature, i.e. it is not part of any particular segment. Interestingly, the words that take stød in Danish correspond to ACC1 words in Swedish and Norwegian, while words without stød largely correspond to ACC2 words. There are a few minor differences in stød-assignment in Danish as compared to accent-assignment in Swedish. Monosyllabic words can take either stød or non- stød. Whether stød can occur or not depends on the segmental makeup of the rhyme of the stressed syllable. For stød to show up it must either contain a long vowel, in which case stød is expressed on this vowel, or a vowel followed by a sonorant consonant, in which case stød is realized on the consonant. Stød can appear on syllables with primary or secondary stress. In compounds, constituents usually maintain the same pattern as in isolation, except for monosyllables, which tend to lose stød in compounds (Bruce & Hermans, 1999).

Crucially, stød is generally assumed to be the marked member of a privative opposition. Recall that words with stød correspond to words with ACC1 in Swedish and Norwegian and that stød and the tonal accents have the same historical origin. Theories assuming ACC2 to be the marked member (Riad, 1998, 2003; Bruce, 1977; Bruce & Hermans, 1999; Gussenhoven & Bruce, 1999;

Gussenhoven, 2005) would lexically specify one set of words (with stød) in Danish, but another set of words (with ACC2) in Swedish and Norwegian. On the contrary, when specifying ACC1 (Lahiri, Wetterlin & Joensson-Steiner, 2005), the same set of words is assumed to be exceptional and therefore lexically specified in Danish (stød) as well as in Swedish and Norwegian (ACC1).

4.2 Psycholinguistic Experiments on Suprasegmental Information