So far, whenever I have wanted to express the fact that the ‘POST-OSL’ equivalents of words such as EME maken were words such as ma:ke, I have been talking in terms of ‘vowels’ and their counterparts, without giving due consideration to metalanguage. Of course, such terminological carelessness has its drawbacks. Take, for instance, the very term Open Syllable Lengthening itself. Though basic to my considerations, it is not nearly as straightforward as it might seem. Thus, it can be read both as ‘lengthening of something (in our case: vowels) in open syllables’ or as ‘lengthening of open syllables’. Now, this terminological ambivalence reflects that phenomena such as OSL might not only be conceived as lengthenings of vowel segments (implying that all the higher constituents in which a vowel figures get lengthened with it), but equally well as changes of syllable quantity that just happen to show in their nuclei, so that the vowel lengthenings could be regarded as sheer epiphenomena. A further possibility of viewing OSL is suggested by Minkova, whose version of OSL clearly implies that the change might be understood as a foot restructuring in which a foot of two light syllables came to be replaced by a foot of one heavy syllable. One aspect of these alternative interpretations is that both the syllable-based view of OSL and the foot-based one make sense only within a phonological theory that recognizes other descriptive levels than that of linear segmental organization.

The investigations are based on what could be called the ‘Minkova corpus of potential OSL candidates’ (see appendix 1). Basically, it is a list of Early Middle English words whose stressed vowels were short and non-high, and which are attested in Modern English. The list was partly drawn up by Minkova herself (the words of Old English origin), and partly taken over from Bliss (1952/3) (Anglo-Norman loans). I have myself added a few random examples, which I happened to come across during my own studies. All in all, the corpus contains 428 items. Thus, although the corpus does not really exhaust the set of potential OSL candidates, it can probably be regarded as a representative sample, as it can be estimated to contain far more than 50 per cent of those OSL candidates with non-high vowels that are still attested in Modern English.

First, the items in the corpus were categorized with respect to the phonological parameters of vowel quality, the consonant following the vowel and the syllable following the vowel. Furthermore, the parameters word class and etymological origin were considered. I trusted that they included most of the qualities that might be relevant with regard to vowel lengthening. The phonological parameters include practically the complete environment of the vowels in question. Only word-initial consonants were not considered, and this was because even a first look at my data confirmed the well-established knowledge that syllable onsets do not systematically influence the length of following vowels.

In this final Chapter I first examine the role of the universal syntactic relations S, A and O in discourse structure, perceiving that S and A are linked in one respect and S and O in another. This helps to explain why there are more examples of morphological ergativity than there are of syntactic ergativity. It also adds to the explanations given earlier for types of ergativity split.

I then focus on what it means for a language to be ergative – whether this carries any implications about the intellectual status or world-view of its speakers (I argue that it does not), or any implications concerning other grammatical parameters. The reasons why some languages are more ergative than others may relate to such things as narrative style, and accidents of historical development. §8.3 summarises some of the main conclusions of this work, and then §8.4 asks what lessons can be drawn, from this survey of ergativity, for our understanding of how language works, and what a linguistic theory should include.

The discourse basis

Categories like ‘ergative’ and ‘accusative’ belong to grammar, that aspect of language which involves definite structures and rules. Recently, attention has been paid to the organisation of discourse. This is a field in which statistical tendencies can be noticed, and quantified, but in which there are no definite constraints.

I never intended to work on ergativity. The topic more or less crept up on me, embraced me, and has never really relaxed its hold. Not that I am complaining – working on ergativity has provided the most intense intellectual satisfaction.

When I first went out to Australia to study an indigenous language, in 1963, the word ‘ergative’ wasn't in my linguistic vocabulary. The Australian Institute of Aboriginal Studies suggested working in the Cairns Rain Forest region and I chose Dyirbal for my major focus of study simply because it was the language with the most speakers (perhaps 100 fluent speakers, at that time). When I returned to London and explained the structure of Dyirbal to M. A. K. Halliday he told me that these unusual-looking grammatical patterns I had uncovered were ‘ergative’. It was only on a return field trip in 1967 that it was brought home to me how thoroughly ergative Dyirbal is at the syntactic level (as well as being split-ergative at the morphological level).

John Lyons, external examiner for my PhD thesis on Dyirbal (submitted in December 1967), was at that time editor of the Journal of Linguistics and he invited me to submit a paper on ergativity in Dyirbal. I planned to do so, ahead of publishing the full grammar, since it seemed to me that people were more likely to take note of a short article in a leading journal than of something hidden away in a long monograph.

Many languages mix nominative–accusative and absolutive–ergative types of intra–clausal marking. This chapter surveys the kinds of factor that condition these splits. They can relate to the semantic nature of the main verb (§4.1), to the semantic nature of the core NPs (§4.2), to the tense or aspect or mood of the clause (§4.3), or to the grammatical status of a clause, whether it is main or subordinate, etc. (§4.4). Some languages show just one conditioning factor while others combine two or more of the parameters (§4.5).

Split conditioned by the semantic nature of the verb

There must be contrastive marking for A and O (if a transitive clause is not to be ambiguous). In §3.3 we discussed ways in which S is like A, and other ways in which S is like O, in terms of universal semantic and discourse features and universal grammatical properties. This appears to be the major explanation for the rarity of tripartite systems, where S is marked differently from both A and O. There are pressures to identify S with A (as in an accusative language) or S with O (as in an ergative language). And some languages pursue a middle course, marking some S like A and some like O; such languages fall into two kinds, ‘split-S’ and ‘fluid-S’.

Before venturing into a detailed examination of kinds of ergative and accusative grammatical patterning, we must distinguish between two different kinds of strategy that languages employ for marking ‘who is doing what to who’. These can be called (1) the syntactically based (or ‘prototypical’) alternative, and (2) the semantically based (or ‘direct’) alternative. We shall see that labels such as nominative, accusative, absolutive and ergative are only properly applicable to languages of the first type.

For languages of the first type, each verb has a prototypical meaning, and grammatical marking is applied to the verb's arguments on the basis of their function in the prototypical instance. English basically follows this approach. The prototypical meaning of hit is that in He hit me (implied: with his hand) or He hit me with a stick. The agent (who propels the implement) is marked as transitive subject (A), being placed before the verb (and being in nominative case if a pronoun). The target, which the implement comes into contact with, is marked as transitive object (O), and placed after the verb (being in oblique form if a pronoun).

When the verb is used with a non-prototypical meaning the same grammatical marking of arguments applies. Hit is categorised as a transitive verb and so there must be a transitive subject stated. In The falling branch hit me, the noun phrase the falling branch is treated as being in A function, although it is not an agent propelling an implement (nor an implement propelled by an agent).

There has been a vogue during recent decades for the formulation of theoretical models in linguistics. These are sometimes suggested on the basis of data in a very limited set of languages, but are then put forth as general accounts of how all human languages operate. When unexpected data from new languages come to notice there can be a number of reactions: ignore it; reinterpret the data so that it fits the theory; revise the theory so that it does explain the data; acknowledge that the theory cannot explain the data and as a consequence abandon it.

In this short Appendix I shall comment on some of the ways in which some theoretical models have approached ergativity. My treatment is partial and selective; a full discussion of this topic would require a book in itself.

Foley and Van Valin's (1984) Role and Reference Grammar (RRG) is one of the few theoretical models to have been formulated with full knowledge of a range of ergative phenomena. Their discussion of syntactic ergativity, pivots, passives and antipassives is informed and useful. My one reservation is that they do not always make a sufficiently clear distinction between syntax and semantics, sometimes talking about syntactic operations applying to semantic categories (I prefer to specify that syntactic operations apply to syntactic categories and then to discuss the semantic correlates of both the categories and the operations.) Lexical Functional Grammar (LFG) has also paid some attention to the various kinds of ergativity; see, for instance Kroeger (1991a, b) for an LFG treatment of ergativity in Tagalog.

Particular linguistic changes generally proceed in a single direction. For example, a preposition or postposition may develop into an affix, and a velar or labial stop may lenite to the semi-vowel w (changes in the opposite directions are either unknown or extraordinarily rare). Through a combination of specific changes a language can shift from one typological profile to another. There is nothing unidirectional about changes at this level (see Croft 1990: 229). Any type of language – in terms of any typological parameter – can change into another type, and back again. (If this were not the case all languages would be inexorably moving in a single direction, towards some ‘ultimate’ language typé)

We have written records going back at the most five thousand years (and then for very few languages). For a number of language families, scholars have reconstructed aspects of a proto-language and suggested an approximate date for this – all of these dates are within the last ten thousand years. Yet language is presumed to have been spoken by Homo sapiens during many tens (perhaps hundreds) of millennia. We thus have available for study only a fraction of the history of human language. It is, however, enough to perceive a clearly cyclic pattern of change.

Turning our attention now to syntax, we can first of all note the confusion concerning the identity of the ‘subject’ in ergative languages. This confusion results simply from the fact that linguistic theory evolved in the context of the better-known languages of Europe, which have a predominantly accusative character at every level. For languages of this type, certain semantic and grammatical properties coincide to give a two-sided definition of subject. The ‘subject’ of a sentence is that NP whose referent could be the ‘agent’ that initiates and controls an activity; the subject NP is normally obligatory in a sentence, receives the unmarked case, may be cross-referenced in the verb, and is the pivot for operations of coordination and subordination.

For ergative languages, these semantic and grammatical criteria for ‘subject’ do not coincide; to employ the notion of subject in such languages, one must decide, in effect, which of the two kinds of criteria should take precedence. Some linguists emphasise semantic criteria, but encounter severe difficulties in explaining all types of grammatical processes in terms of semantically defined ‘subject’ for ergative languages. (In the Appendix, I describe difficulties which Relational Grammar has had in accounting for antipassive derivations.) Other linguists take syntactic/morphological criteria as basic; this facilitates statements of grammatical derivation, but is bound to complicate any attempt to provide semantic interpretation for the grammar.

Chapters 3 and 4 dealt with intra-clausal or morphological ergativity, relating to ways in which S and O are marked in the same manner, and A in a different manner, within a single clause. It is now time to consider what happens when two clauses are linked together in a coordinate or subordinate construction. In some languages there are syntactic constraints on clause combination, or on the omission of coreferential constituents in clause combinations. If these constraints treat S and O in the same way and A differently, then the language is said to be ‘syntactically ergative’, with an S/O pivot; if they treat S and A in the same way and O differently, then it is said to be ‘syntactically accusative’, with an S/A pivot. (In some languages the syntactic pivot may have a further function, relating to syntactic processes within a clause, such as the questioning of an NP; this will be discussed in §6.3.)

Preliminary exemplification was given in §1.2 for Dyirbal, which is ergative at the inter-clausal syntactic level and works in terms of an S/O pivot. Two clauses may only be coordinated in Dyirbal if they have a common NP which is in S or O function in each clause; the occurrence of this NP in the second clause is then generally omitted. We also mentioned that English works in terms of an S/A pivot – if two clauses that are coordinated have a common NP this can only be omitted from the second clause if it is in S or A function in each of the clauses (we can say Father returned and saw mother but not *Father returned and mother saw).

Every language has intransitive clauses, with a predicate and a single core argument (that we call S) and transitive clauses, with a predicate and two core arguments (A and O). There should always be the means to distinguish A and O. Some languages do this by constituent order (e.g. English), some use cases, particles or adpositions, and some employ pronominal cross-referencing on the verb (many languages employ a combination of these strategies). The marking of core syntactic relations – A, S and O – is generally referred to as ‘morphological ergativity’ or ‘morphological accusativity’ since this is generally shown by case inflections or verbal cross-referencing affixes. A more exact label would be ‘intra-clausal ergativity/accusativity’, since particles and adpositions make use of a syntactic – not a morphological – mechanism, and constituent order is without doubt a matter of syntax.

There must be some means of distinguishing A and O for a transitive clause. The marking of S in an intransitive clause can be the same as A, or the same as O, or different from both. There are thus three basic possibilities:

1. S = O (absolutive), A different (ergative) – an ergative system

2. S = A (nominative), O different (accusative) – an accusative system

3. A, S and O all different – this is a ‘three-way’ or ‘tripartite’ system.

The accusative pattern is, of course, commonest among the languages of the world. The ergative pattern, with which this book is concerned, is by no means uncommon.