Showing papers on "Naturalness published in 1986"

On the naturalness of stop consonant voicing

Abstract: A long recognized problem for linguistic theory has been to explain why certain sounds, sound oppositions, and sound sequences are statistically preferred over others among languages of the world. The formal theory of markedness, developed by Trubetzkoy and Jakobson in the early 1930's, and extended by Chomsky and Halle (1968), represents an attempt to deal with this problem. It is at least implicit in that theory that sounds are rare when (and because) they are marked, and common when (and because) they are not. Whether sounds are marked or unmarked depends – in the latter version of the theory, particularly – upon the ‘intrinsic content’ of acoustic and articulatory features which define them. There was, however, no substantive attempt among early proponents of the theory to show what it was about the content of particular features and feature combinations that caused them to be marked, and others not.

Implementation of and experiments with a variable precision logic inference system

Abstract: A system capable of performing approximate inferences under time constraints is presented. Censored production rules are used to represent both domain and control information. These are given a probabilistic semantics and reasoning is performed using a scheme based on Dempster-Shafer theory. Examples show the naturalness of the representation and the flexibility of the system. Suggestions for further research are offered.

Some aspects of naturalness and abstractness in phonology

Abstract: generative phonologists are wrong when they regard French orthography as rational or quasi-optimal (see Dell (1973a: 193)), I cannot follow Martinet: generative phonology is not a 'pretext'. This type of criticism is certainly questionable because it is only a posteriori that analysts discover some regularities in the correlation between underlying representations and conventional spelling. According to Martinet (1969b: 105), many verbs of the second and the third conjugations (that is, verbs whose infinitives do not end in -er) have two stem allomorphs: (i) 'Un th4»e plein en consonne': ils 6crivent /il z ekRiv/ ('they write') ( i i ) 'Un ihdie 6court6 de U consonne finale du pr6c6dent'; j ' 6 c r i s e k R i / ('1 w r i t e ' ) Martinet gives the proper morpho-syntactic contexts in which these allomorphs appear. Final schwa is absent from the phonemic representation of the long allomorph (this detail is crucial because other phonologists will posit a word-final schwa in the phonemic representation of ils 6crivent, for instance, or of all items whose final segment in phonetic representations is a consonant: see §V.2.2 and §V.3). In most cases, in fact, phonemic representations (in structuralist works) are similar to phonetic representations, but without redundant features, such as vowel length. In French, nasal vowels, for instance, are phonemic and contrast with sequences of oral vowel plus nasal consonant, or with oral vowels: bon /bo/ ('good', masculine), beau /bo/ ('beautiful'), bonne /ban/ ('good', feminine). In Malmberg (1971b: 312), we read: 'Je continue done 4 Previous accounts 80 coepter les voye 1 les nasales pani les phondies vocaliques'. For Jakobson & Vaugh (1979: 135), 'atteipts to interpret the French nasal vowels as a eere iipleientation of a sequence -oral vowels ♦ nasal consonant — aeet with a nuaber of obstacles1. Among these 'obstacles', Jakobson & Vaugh mention the nasality contrast in liaison: bon ami /ban ami/ ('good friend'), mon ami /mo n ami/ ('my friend'). V .2.2. Gloesenatics The biphonemic interpretation of nasal vowels would entail that the phonetic sequences of oral vowel plus nasal consonant (e.g., bonne ('good', feminine)) are followed by a schwa: as Malmberg (1971b: 312) notes, this step was taken by the Danish glossematic school, led by Hjelmslev. This constitutes a remarkable exception; glossematicians (see Togeby (1951)) interpret bon tbol ('good', masculine) as /bon/, and bonne [ban! ('good', feminine) as /bona/. In the same way, final consonants are truncated: grand [gRal ('large', masculine) = /gRand/, and grande tgRShdl ('large', feminine) = /gRandd/, where the final schwa 'protects' /d/ from truncation, and is itself truncated. However, more 'orthodox' structuralists have frequently objected to this interpretation of French data: 'Un inconvenient qui est evident et qui consiste en une discordance enone entre la substance et la forme, entre le eodeie et sa manifestation physique1 (Malmberg 1971b: 312), or: 'Le principal inconvenient de ce type d'interpretation est qu'il off re une image compldtement deforce du cotporte»ent des f rangais d'aujourd'hui' (Martinet 1969a: 27). It is also true that although the system is simplified, by Previous accounts 81 dispensing with four nasal phonemes, phonemic strings are made more complex: caneton ('duckling') will be phonemicized as /kanjton/ instead of /kanto/, but as Martinet adds, 4i 1 senble que la siiplicite formaliste s'applique exdusiveient au noibre d'uniUs dans le systdee, et non au noibre d'unitts dans la chaine*. As we 6hail see in §V.3, this type of criticism applies equally to the most classical versions of generative phonology.