Nyelvtudományi Közlemények 91. kötet (1990)

Tanulmányok - Kornai, András: The sonority hierarchy in Hungarian 139

THE SONORITY HIERARCHY IN HUNGARIAN 143 Since the counterarguments based on distribution and speech errors are equally valid in the case of Hungarian, I will not adopt Selkirk's solution here. But as these cases (e.g. Danish fisk 'fish' vs. fiks 'fix') constitute the only serious counterexample to (1) and (3) in a number of languages, Selkirk is obviously right in trying to explain them by some special principle that leaves the larger generalization intact. The sum up what we have so far, the mirror image of the onset structure of Hungarian syllables is, by and large, properly included in the set of attested coda combinations that make up the coda structure. As Algeo (1978) notes, it is possible to make the system look a great deal more elegant by being more selective with the data we include. For instance, the exclusion of inflected forms would go a long way in eliminating the "irregular" entries that appear under the diagonal. The 2nd.sg.imp.indef suffix j is the only source of codas such as döfj 'pierce!', lopj 'steal!', bukj 'fall!', szívj 'suck', dobd 'throw!', or vágj 'cut!', and the 2nd.sg.imp.def suffix d is the only source of codas such as dobd 'throw!', szívd 'suck!', or tanítsd 'teach!'. Of the remaining 10 entries under the diagonal, two (NOSZF and MAFC) could be excluded on the basis that they are acronyms (these were the only acronyms included in the corpus, because the "spelling pronunciations" *enóeszef and *emaefcé are never heard) and two (Ape and Szakcs) because they are proper names. One (borscs 'kind of soup') is the single example of its type, and another one (szomj 'thirst') can be argued to end in |h| rather than |j| on the surface. But no matter how hard one tries, there seems to be no way to exclude füst 'smoke', szaft 'gravy', recept 'prescription' or akt 'nude'. As the 'hard' counterexamples all end in i, one might try to reshuffle the matrix in (6) so that only "easy" items appear under the diagonal. To see what is involved here, let us take a look at the high end of the hierarchy. The ordering m<l<n<j<r appears to be extremely well motivated: the codas in film 'id.', slejm 'phlegm', farm 'id.', ajánl 'recommend', fájl 'file', görl 'girl', kombájn 'combine-harvester', modern 'id.', and férj 'husband' are all decreasing in sonority. Moreover, every pair of decreasing sonority corresponds to an attested coda, with the exception of n > m, but even this is attested (in the reversed order) as onset. Yet it is possible to rearrange the sequence (e.g. to m<n<l<r<j) by the data-manipulation techniques discussed above* and the reader is invited to try it. For those who prefer to work with a larger set of data, I list here the "best" example of each onset and coda type not mentioned so far: srác 'kid', sors 'fate', mersz 'daring', francia 'french', turf 'id.', próba 'trial', szörp 'juice', krém 'cream', park 'id.', bérc 'peak', trágya 'manure', szirt 'cliff', tekercs 'scroll', korty 'gulp' érv 'argument', bróm 'bromide', szerb Nyelvtudományi Közlemények 91. 1990.