The modal auxiliary should, when it is used to express ‘probability’, is generally considered as a weaker equivalent of must (see for instance Leech & Svartvik, 1975: 131 and Hornby, 1962: 223). Although this informal assumption contains a fair degree of truth, this paper will show first, that it requires qualification and secondly, that it gains strength once it has been qualified. In other words, there is more truth in the assumption than is suspected by its casual users.

The method I will use is to consider the compatibilities between must and should and their environment. Though the relevant kind of environment cannot be defined in a purely syntactic way, it can be accurately specified in simple semantic terms, many of which have syntactic correlates. Verbal contexts are generally available but situational contexts will occasionally have to be used. It will then appear that the diversity of situational contexts can be reduced to very few elementary relations.

I. A pervasive, though not always explicit assumption of transformational grammar has been that syntactic and phonological rules form separate, discrete units in the organization of grammar; that phonological rules are not ordered among syntactic transformations, but rather apply in a block at the end of the syntactic derivation.1 Chomsky and Halle (1968) go even further in this separation, arguing that the surface structure is not the immediate input to the phonological component, but first undergoes certain ‘readjustment rules’, such as insertion of word boundaries. I have argued elsewhere (Kaisse, 1977) that this added step is unnecessary, and that there are in fact certain inter-word phonological processes that require the full information present in the labeled bracketing of the surface structure. However, my proposal is not a radical departure from the standard theory, in that it remains an INTERFACE model, in line with that proposed in Pullum and Zwicky (to appear). The claim remains that all syntactic rules apply before all phonological rules. In the case of word-internal phonological rules this is perhaps not crucial, but for rules of external sandhi, it is very important, both empirically and theoretically. For if we give up the requirement that no phonological rule apply during a syntactic derivation, we greatly increase the power of our theory of grammar and give up the ability to predict on a universal basis, the order of application of two rules, one phonological and the other syntactic.

Derbyshire and Pullum (1981) report on the evidence they have accumulated during the past two years showing the likely existence of twelve languages with object-initial basic order (OVS or OSV). Such languages are contrary to what had been generally predicted in the literature on word order typology until 1977. All twelve languages are found in what might broadly be termed the Amazon basin of South America. Seven of the eight OVS languages belong to the Carib family: Apalai, Arekuna/Taulipang, Bacairi, Hianacoto-Umaua, Hixkaryana, Makúsi, and Panare. There is not much descriptive linguistic material available for any of these languages, other than Hixkaryana. What there is can be found listed in Derbyshire and Pullum (1979).

I. Like so many linguistic terms, the term syllable has been used in many different senses, and theoretical distinctions have been made, aimed at avoiding confusion. There is hardly any need to remind the reader of Pike's distinction between PHONETIC and PHONEMIC syllables: ‘the phonetic syllables must be analyzed into the structural phonemic syllables’ (Pike, 1947: 90). Although obviously beneficial, this division still leaves considerable room for confusion, both on the phonetic and on the phonological side. On the phonological side the syllable can be thought of as a unit of organization of smaller phonological units (cf. e.g. Bell (1977), Bell & Hooper (1978)) or as a unit defining the scope or environment for phonological processes, or as accounting for regularities of quantity and tonality (cf. e.g. Pike (1947: 90), and Hooper (1972)). On the phonetic side there are at least two approaches to the syllable. It has been looked on as defining units in speech production (cf. Stetson (1951), Catford (1977: 89–90)) or as a means of describing sonority variation in the phonetic signal (cf. e.g. Jespersen (1949/1934: 118–121)). This latter aspect relates to the perception of the linguistic signal by speakers and to the acoustic buildup of the phonetic medium. Finally, there are those who deny the usefulness of the concept altogether.

The present paper1 will be concerned with some descriptive and theoretical issues connected with Glide Formation (GF) and Stress Placement (StrP) as well as with the interplay of these rules in the ‘phonological component’ of Modern Greek (MG). These problems have recently been discussed by Irene Warburton whose article (1976) will serve as a point of reference.

Warburton's paper consists of two major parts of rather unequal strength. The first section (‘The relationship between stress and vowel length’, 260–268) makes a good case against attempts to condition the MG stress rule(s) phonologically by setting up never-surfacing distinctions of vowel length. It is also meritorious plea for the view that morphological conditioning ‘does not necessarily introduce irregularity in the language’ (Warburton, 259), as far as the rule makes reference to ‘a natural class of items… that can be subsumed wihin a single morphological feature (e.g. Past, Plural, etc.) that is independently available’ (l.c.).

1. Efforts to limit the power of transformations, and thereby reduce the class of possible grammars, have been concentrated in two areas: the conditions which restrict the application of the transformational rules, and the examination of the consequences of placing the burden of description on the non-transformational components of the grammar. Recent work on the non-transformational components has led to some interesting conclusions about how this power can be reduced. An article by Bresnan (1978) is a good example of this line of research. She cites a number of cases where transferring the burden of description from the transformational component to the lexicon and semantic interpretive components not only leads to a simplification in the from of the transformational rules, but also captures generalizations not adequately explained by transformations.

Chomsky (1977) argues that all constructions that apparently involve unbounded deletion rules obeying island constraints in fact involve successive cyclic wh-movement followed by deletion in COMP. Among the relevant constructions are English equatives. There is no direct evidence that they involve wh-movement. Polish equatives, however, often show quite clear evidence of the operation of wh-movement. It might be thought, then, that they provide some evidence for Chomsky's position. In this paper, I will argue that this is not in fact the case. I will argue that Polish equatives often involve not only wh-movement but deletion in S as well. I will then argue that they sometimes involve just deletion. Finally, I will argue that this deletion is unbounded and subject to island constraints. Clearly, if this is correct, Polish equatives, far from providing support for Chomsky' position, in fact provide important evidence against it.

In recently proposed analyses of the causee's case (Aissen, 1974; Cole, 1976; Comrie, 1976) it has been generally assumed that: (a) causatives are underlyingly bisentential with a structure such as s[X cause s[Y to do something (run, hit, jog, … ]s]s; (b) the causative merger with its complement verb erases the grammatical relation of the complement subject; and (c) the case of the derived NP (i.e. the causee) emerging from the complement subject is determined by some universal grammatical principles of higher order.1 On the basis of a cross-language study, Comrie (1976a), for example, has suggested that the case of the causee follows from two main syntactic principles: the Doubling constraint and the Case Hierarchy (CH). The doubling constraint prevents the primary subject and object relations from duplicating in simple structures. The case hierarchy, which feeds on the doubling constraint, demotes the causee to Direct, Indirect, or Oblique Object (DO, IO, OO) position with each increase in the number of complement arguments of the causative construction. Thus if the complement clause is without an object, the causee positions itself as DO, the highest unoccupied position on the hierarchy. If the complement DO is already present the causee moves to IO, and if both DO and IO positions are filled, the causee demotes to OO NP whose nature under CH is unspecified. Turkish supposedly represents a paradigm of this operation.

Most phonologists regard the syllable as a unit of language. In recent years, partly in reaction to Chomsky's and Halle's neglect of it in The sound pattern of English, much work has been done to incorporate the syllable into phonological theory (Anderson & Jones, 1974; Bailey, 1978; Hoard, 1971; Hooper, 1972, 1974, 1976, 1978; Kahn, 1976; Pulgram, 1970; Rudes, 1977; Vennemann, 1972).

Syllable theories have been based on evidence from phonetics, phonological processes, prosody, language change, child language acquisition, and language universals. The purpose of this experimental study is (a) to contribute empirical evidence about the nature of the syllable from native speakers' actual syllabifications of words and (b) to determine how this evidence reflects on the syllable theories already proposed.

Syllable studies have focused on two major questions: (1) what is the structure of the syllable and (2) how are words divided into syllables (syllabified). These questions are obviously related, and the answers to each have implications for the other. This study was designed to elicit data directly addressing both questions.

It is usually assumed that English has two aspects, the perfect and the progressive. I would like to suggest, however, that there are in English not two but three aspects – the perfect, the progressive and the passive.

The most common description of the (actional) passive is one which derives sentences containing be+ past participle from an underlying transitive active structure. In a description of this kind there are several defects of a theoretical nature which reveal themselves in three main areas. The first problem area is the set of verbs which are apparently transitive but which nevertheless do not passivize, for example in English have, suit, resemble, in German haben ‘have’, überkommen ‘come over’, erfordern ‘require’ (see Steube & Walther, 1972: 20–22).

With friends like Edmondson (1979), the descriptivist needs no enemies. The descriptivist's dilemma which Harris (1978) drew attention to was a dilemma about the assignment of truth values to performative and corresponding non-performative utterances. Edmondson's appeal to an obscurely drawn distinction between semantics and pragmatics is simply irrelevant to this issue. The semantic theory with which Edmondson saddles the unfortunate descriptivist makes his position, if anything, worse.

At a time when linguistic theory is becoming increasingly fragmented and decreasingly dominated by a single orthodoxy, it has been an encouraging exercise to compile a list of statements about language which are likely to be accepted by virtually all linguists, irrespective of what they think about all the many issues on which linguists disagree. The following list contains no fewer than 83 claims which have been accepted by a wide range of British linguists, and there is no reason to believe that the sample of linguists who have helped me in compiling the list is particularly biased. It seems reasonable to claim that other linguists are LIKELY to accept these statements, although I certainly cannot claim that every linguist accepts every one of them. So far as I know, no attempt has ever been made before to find out what linguists at large actually believe, although any writer of an introductory text-book hopes that he is expounding a widely held set of views. Considered as a piece of research, this investigation seems to me to have produced at least one interesting result: linguistics really is making some progress, in a cumulative way, and we are not just lurching from one 'paradigm' to another, as some of us sometimes suspect in our gloomier moments. Moreover, it raises the interesting question what other statements could be added to the list given here, which certainly is not meant to be exhaustive. I hope that other linguists with more imagination than me can bring the list into the hundreds, as should surely be possible.

Pillinger (1980) argues that the Latin accusative and infinitive construction provides evidence against some of the current assumptions of Relational Grammar (RG) and, 1to a lesser extent, the Extended Standard Theory (EST). I argue that, even if one accepts Pillinger's analysis of this construction – as I would, perhaps with less diffidence than Pillinger himself – then this analysis is completely neutral with regard to deciding between RG and EST and compatible with either.

The first sentence of David Lightfoot’s review (1981) of my Liberty and language (1980) refers to the recent emergence of ‘the ̏nouveaux philosophes̋ and ̏the New Right̋ ’. Later in his first paragraph he characterizes this group as people who ‘believe in the predominance of heredity over education’. His second paragraph begins: ‘Liberty and language is a contribution to this literature.’