There is a period in the history of the German language of which we know relatively little, although it is much better documented than the Old High German period, where every phonetic detail has been intensively studied by several generations of scholars. I refer to the stage after the phonologizations of the umlaut variants were completed and before Luther and his precursors and contemporaries demanded a unified Schriftdialekt, the major feature of which was to be a uniform vocalism. As Fourquet has pointed out, this period was one marked by a large number of changes. To the traditional historical phoneticians these changes were ‘spontaneous’; that is to say, since they were phonetically unconditioned they were also inexplicable. Fourquet further suggests that this period may be the period needed, the period particularly favorable to testing and sharpening the methods of structural diachronic phonemics with the help of the descriptions that modern dialectology has put at our disposal. The following is an attempt at a structural history of the vocalism of Middle Bavarian during this period and up to the present time. Since the term Middle Bavarian covers a huge dialect area with many geographical variations, I must add here that the specific ‘target dialect’ of this study is that of the Salzachgau. It is spoken in a region about 80 km. east-southeast of Munich and 40 km. north-northwest of Salzburg. This area was under Austrian rule until the Treaty of Frankfurt in 1810, and has since belonged to Bavaria.

One of the most common items in the Albanian vocabulary, plak ‘old man’, pl. pleq, f. plakë ‘old woman’, has never received a conclusive etymological explanation. Tagliavini (L'albanese di Dalmazia 222–3), following Jokl, properly rejects Meyer's connection with Gk. palaiós. The latter must be related to pálai, and this in turn, together with Skt. ciráḥ and Welsh pell, to IE *kwel-; Pokorny accepts this doctrine (IEW 640). Like Jokl, Tagliavini and Pokorny (IEW 804–5) relate plak to Gk. peliós, poliós, etc., under a base *pel-. This seems like a good fit. But, as we shall see, it is not necessary to follow Jokl in considering the final -ak a suffix from Slavic. Not only does the word bear all the appearances of belonging to the old layer of the vocabulary; there is no evidence that so tenuous an element as pl- could have been segmented as a productive base to receive a borrowed suffix as late as the first millenium a.d. Furthermore, the conservative Greek enclaves show an initial cluster pl-, with no trace of an intervening vowel.

The Finnish nominal declension has fourteen eases and two numbers. There is no distinction of gender. Case and number are separate formal entities, the case markers for eleven of the fourteen cases being identical in the singular and the plural. The genitive case is one instance in which the case endings are not the same in the two numbers. The other two are the illative and nominative.

The classical definition of the phoneme runs something like this: a set of sounds in complementary (noncontrastive) distribution and phonetically similar. To this one may append Trager's statement, ‘The criterion of complementary versus contrasting distribution is first, then that of phonetic similarity.‘ While it may be generally admitted that the criterion of phonetic similarity is of secondary importance, the question can be raised: is it of any importance at all? Here, as in everything else, the answer depends upon the level of analysis. The phoneme qua phoneme exists only by virtue of its distribution on a higher level, in the morpheme. On the phonetic level the phoneme is only a metaphysical possibility, on the morphemic level it is reality. Phonemes are phonemes because of their function, their distribution, not because of their phonetic similarity. Most linguists are arbitrary and ad hoc about the physical limits of phonetic similarity.

[A problem which arises in the course of research on mechanical translation is the prediction of dictionary size. This article investigates the relation between empirical frequency laws and the function V(n)—the expected number of different words in an n-word sample of text. It is found that the probability-law proposed by Joos (1936) yields results which do not check well with experiments, and it is concluded that some modification of it is necessary for the purpose of vocabulary prediction.]