The analysis

The data were analyzed quantitatively using version 2S of the VARBRUL program. The effect on vowel shortening of eight different factors was measured. The factors were chosen on the basis of previous sociolinguistic variation studies, hypotheses in the literature about the typewriter effect, and the researchers' knowledge of and intuitions about the Hungarian language. The following factors were used:

Dependent variable:

short vs. long, as the high vowel was spoken on tape. The realizations of the high vowels were coded as short or long by BSI transcribers, all of them trained linguists; one of the transcribers, Anna Borbély, is an author of this paper. Although Fónagy (1956) and Magdics (1960) found that a large number of vowel realizations were phonetically half-long, this intermediary category was not used in the BSI transcription because of categorical speech perception, i.e. the tendency of our perception system, which is determined by our phonological system, to categorize all intermediary realizations as one or the other of the phonological categories.

Independent variables:

We coded four linguistic variables (type of vowel, tempo, following sounds, and position within the word) and four extralinguistic variables (how the vowel was typed, speaker, socio-economic status, and sex). These variables are described in more detail below. At least two other factors which may have a significant effect on vowel shortening - age and dialect background of the speaker - were not immediately available, and therefore were not included in this study.

How the vowel is typed:

short vs. long. For a high vowel that according to standard orthography should be long, this variable indicated whether it was typed short or long. The effect of this variable, of course, is what we have been calling the typewriter effect.

Type of vowel:

í, ú, û. The three long high vowels differ with regard to roundness, backness, frequency, and typical intra-word positions. In a frequency count of 500,000 running words of contemporary Hungarian fiction, Füredi and Kelemen (1989: 430) found that the grapheme í occurred 12,622 times, ú 10,630 times, and û 4,393 times. (The frequencies for the corresponding short graphemes are: i 104,597; u 26,306; and ü 16,461.) The high frequency of í is at least partly due to the existence of the causative suffix -ít. In word-final position, long í is extremely rare: there are only 16 lexemes with the final unround vowel in the best dictionary of Standard Hungarian (cf. Papp 1969: 156), and most of those are non-lexical words. In contrast, there are 325 lexemes ending in -ú, and 601 ending in -û. In an early analysis of Hungarian spontaneous conversation, Szende (1973: 28) found the following frequencies in 18,000 running words of speech: of all the phonemes /i/ occurred 3721 times (4.69%) but /i:/ only 379 times (.48%); /ü/ occurred 311 times but /ü:/ only 175 times; /u/ 766 times but /u:/ only 207 times.

Tempo:

normal vs. fast. Ács and Siptár (1994: 555) classify vowel shortening as one of the lenition processes characteristic of fast speech. Kontra (1995: 15-16) found that in normal reading tempo, five out of ten teachers pronounced the final vowel in fésû 'comb' long, but in fast reading only two teachers pronounced it long. However, tempo had no effect on the vocational trainees' reading. Kassai (1991: 97) explains this in the following way: at normal rate, vocational trainees pronounce nearly twice as many short vowels for standard long vowels as do teachers, therefore "in vocational trainees' fast reading there is hardly any long vowel left for fast rate to shorten."

Following sounds:

no following sound, one vowel, one consonant, two identical consonants, two different consonants. The sounds that followed the vowel being investigated were coded as vowels or consonants, with a distinction made between one or two following consonants; in the latter case, the identity or lack of identity of the consonants was also coded. Note that in the case of a following vowel, that vowel was always different from the vowel being investigated; in other words, vowels were never doubled. As we will discuss below, we measured the effect of the following sounds within three linguistic units: the morpheme, the word, and the intonation unit (i.e. stress group). We restricted the data coded for this variable to sounds that belonged to the same intonation unit as the vowel being investigated: we assumed that the intonation unit was spoken without an interior pause, and therefore that the realization of the vowel in question could have been influenced by the following sounds within that unit. Therefore, vowels occurring at the end of an intonation unit were not included in the analysis for this variable, because it would have been impossible to establish the presence or absence of a pause after the vowel without listening to the recordings once again. This restriction reduced the data for this variable by a small amount (see note 4.4): in two cases, the vowel in question occurred at an intonation unit (stress group) boundary in the two passages (fölmerült a gyanu, hogy... 'The suspicion arose that...' in passage 1, and Felmerült a gyanú, hogy... in passage 5).

Position of the vowel within the word:

wordfinal vs. non-wordfinal.

Speaker.

There were 17 speakers, and it was assumed that the variation could be speaker-dependent.

Socio-economic status.

Speakers belong to one of five SES's: teachers, university students, sales clerks, blue-collar workers, and vocational trainees.

Sex:

male vs. female.