dstub (original) (raw)

Workshop: Voice and Grammatical Functions in Austronesian

Symmetrical Voice Systems and Precategoriality

in Philippine Languages

William A. Foley

University of Sydney

4 A comparative typological excursion: Northwest Coast languages

3.1 The lack of an noun/verb distinction in Northwest Coast languages

As strikingly unusual Philippine type languages might be, they are not completely unique. The problematic status of a noun/verb distinction was noted at least as long ago as 1911 by Boas (Boas 1966[1911]) for the Amerindian languages of the Northwest Coast of North America and has continued to be argued about by scholars working with these languages (Swadesh 1939; Sapir and Swadesh 1946; Boas 1947; Kuipers 1968; Jacobsen 1979; Kinkade 1983; Broschart 1991; Sasse 1993; Jelinek and Demers 1994). All three genetic groups in the area, Wakashan, Salish and Chimakuan, are said to exhibit a lack of a strong lexical contrast between noun and verb, as was claimed above for Tagalog roots. Consider these famous Nootka examples from Swadesh (1939: 78-9):

Note that the root mamu·k - �work� denotes an activity and would correspond to a prototypical verb, while_qu·?as_ �man� denotes an entity, a person, and as such is a prototypical member of a noun class. In (43a) everything is as we might expect: the mood marker, a prototypical verbal inflection, attaches to_mamu·k_ - �work� and the DETerminer, a prototypical modifier of nouns, attaches to_qu·?as_ - �man�. But our expectations are turned on their head in (43b): here_qu·?as_ - �man� is inflected for mood and mamu·k - �work� takes the DETerminer! These data argue that there is no sharp word class distinction between these two words; prototypical verbal and nominal inflections and modifications can apply to both. Parallel structures are found in Tagalog:

The pivot marking preposition_ang_ is the Tagalog equivalent of the Nootka DETerminer -?i. There is no mood suffix like Nootka -ma to mark predicates in Tagalog; clause initial position is sufficient. Words of a wide semantic range can function as clause initial predicates in Tagalog, with undifferentiated syntactic behavior:

Just as the case with Nootka (examples (43a, b) above and below) (Swadesh 1939):

And Straits Salish (Jelinek and Demers 1994), is another Northwest Coast language claimed to lack a lexical noun/verb distinction:

3.2 Symmetrical voice in the Northwest Coast: Kwakwala

While these data support the contention that Tagalog patterns syntactically like the Northwest Coast languages, other languages claimed to be lacking a lexical noun/verb distinction, it is important to remember the reason why this typological fact became important in the first place. This salient typological fact was claimed to be linked to that other very peculiar typological property of Philippine-type languages�their symmetrical voice structure. If roots are precategorial, i.e. neither noun nor verb, then the lack of any unmarked voice form of the verb finds a ready explanation: deriving a voice orientation of a verb requires deriving a verb as well, a process efficiently done simultaneously. It also entails a marked difference in lexical structure between asymmetrical languages like English and a symmetrical language like Tagalog. The lexeme give in English is a verb, with a corresponding argument structure <actor, undergoer, locative>. The Tagalog root bigay �give� however is precategorial; it lacks a true argument structure, but does have a precategorial semantic structure like �the giving of something to someone by someone�.

If Northwest Coast languages, like Tagalog, have precategorial roots, then we would expect to find among them symmetrical voice languages. And indeed we do, in the first of these languages to be described as lacking the noun/verb distinction, Kwakwala (Boas 1911) a language of the same family as Nootka. Kwakwala is remarkably like Tagalog in many respects, and by exploring and developing the implications of these similarities, we will attempt to delimit more precisely just what is meant by a symmetrical voice language.

Kwakwala like Tagalog is essentially predicate initial with all NPs case marked by prepositions, which along with DETerminers are phonologically encliticized to the previous full phonological word (a process, interestingly, also found in the Miaoli dialect of Atayal (Huang 1993)). The basic case markers are _i_~f PIVot, _x5_ACCusative and s INSTRumental (all homophonous with the corresponding bound verbal pronominal enclitics for third person). The PIVot is also crossreferenced by a pronominal agreement enclitic to the first element of the clause, the verb, if it occurs first; otherwise, an auxiliary (Boas 1947; Levine 1980a, b); Anderson (1984). Consider this example from Anderson (1984: 24):

(Note that the PIVot case marker and the third person enclitic pronoun are identical.) Other than the encliticization of the NP markers and the cross referencing of the PIVot, the system is much like Tagalog. Tagalog too requires clitic pronouns to precede the main verb, if there is a preceding auxiliary or adjunct phrase:

Further, Kwakwala has a system of multiple voices: an unmarked verbal form, in which the actor is PIVot; a verb form suffixed with -su/, in which the undergoer is PIVot; and a verb form suffixed with -ayu, in which OBLique NPs assume PIVot status (Levine 1980a):

Each verb form, f, -su? or -ayu, corresponds to a different PIVot choice, whose semantic role is indicated by the proper suffix. The PIVot is realized in Kwakwala by position, immediately following the main verb (if an enclitic pronoun to an AUXiliary, it will precede the verb) and the case marker -i. Like Tagalog voice affixes, these same suffixes are used derivationally to form what look like prototypical nouns; _m@x_- �strike� m@n-ayu �drumsticks�; _G@ls_- �paint� (V), G@ls-ayu �paint (N)�; _yaq�@nd_- �talk�, yaq�@nd-ayu �topic of conversation�; _laxw_- �fish� law-ayu �fish trap�; n@p-x?id throw-PUNC �throw�, n@p-x?id-ayu �a projectile�; _?n@w_- �aim�, ?no?-ayu �medicine for rubbing�; n@p-x?id- throw-PUNC �throw�, n@p-x?id-su?, �a target�;_ha?mx?id_- �eat�, ha?mi?yid-su? �food�; there is even an equivalent suffix to Tagalog -an for deriving putative nouns which describe the place where the activity or event occurs: _n@k_- �steam�, n@g-as �place for steaming�; _t@Lc�_- �warm oneself�, t@Lc�-as �place where one warms oneself�;_hanl_- �shoot�, hanl-as �place of shooting�; t�@m� �sew� t�´@?-as �place of sewing� (Boas 1947). This suffix can also be used rather like another voice marker, to indicate locational participants as PIVot, exactly parallel to Tagalog -an (Levine 1980a: 243):

And perhaps most strikingly parallel to Tagalog, any fully derived verb with its governed arguments can function as a NP simply with a preceding DETerminer and NP case marker, with no further derivation (Boas 1947; Levine 1980a):

in (52d) -@L is an allomorph of -su? used with experiential and perception verbs (Levine 1980b). In each of the examples in (52), the verb form following the DETerminer is a fully finite one that could serve as the main predication of clause if it were in clause initial position. Like Tagalog there are no nominalizing affixes used.

It is important to note the points of similarity and difference between Tagalog and Kwakwala to get a fuller typological overview of symmetrical voice systems. Note firstly that non-subcategorized OBLique arguments like locatives and instruments are promoted directly to PIVot status without going through any intermediate lexical derivation of APPLcativization. This seems to be an important defining property of symmetrical voice systems, provided, of course, the languages do not otherwise exhibit clear defining properties of asymmetrical ergative-absolutive languages which also show this property, due to their no double ABSolutive constraint. We have already argued that Tagalog is clearly not an ergative-absolute language, and Kwakwala too exhibits no clear characteristics of this language type: the presence of an overt ACCusative case preposition, for example, being a strong argument against classifying the language as an ergative one.

One difference between Kwakwala and Tagalog concerns the morphological paradigm of the voice affixes. In Tagalog all verb forms bear overt affixes; there is no unmarked verb form, a point I argued was crucial to the overall typology of symmetrical voice languages. But note in Kwakwala, the verb form which indicates the actor is PIVot bears no overt affix; does this indicate that their form is unmarked and so constitutes an argument against the symmetrical analysis which asserts there are no unmarked voice types, in contrast to the situation in asymmetrical languages? The answer is no, and it revolves crucially around making a distinction between unmarkedness and simply the lack of an overt morpheme, in essence, claiming that f is just one option in a paradigmatic cell of options and no more basic than the others. In other words we are distinguishing between the basic lexemic form and a derived form marked by no overt morpheme. There is good evidence to believe the latter is the case in Kwakwala. Note that in English we have the basic lexemic form hit, but that any derivative use of this is overtly signalled: be hit, hitting, hitter, hittee, etc. Crucially, nominalized uses like the last two, in which the argument structure of the verb had been altered, are derivationally marked. Now consider Kwakwala. Note that the same form which indicates the actor is PIVot is also the agentive nominalization form �hitter� (52a) or �thrower� (52c). And it is thefmarking which distinguishes this from the ? nominalization �hittee� (52b) or �seen� (52d). This strongly suggests that f is in paradigmatic contrast with -su?, -L, (or -ayu in still other cases). The forms are not really unmarked in the sense of being basic, underived; they are marked, but by f.

Finally, note that actors not functioning as PIVot in Tagalog are still case marked as CORE NPs with ng, but in Kwakwala they appear to be marked as OBLique with s. This may be a difference between the two languages, so that the property of non-PIVot actors remaining CORE may not be a defining property of symmetrical voice languages. However, we should inquire as to whether Kwakwala s is unambiguously an OBLique case marker. First, it does mark NP arguments of some verbs which are clearly subcategorized by the verbs in question and hence CORE participants:

Perhaps, even more problematic for a claim for the oblique status of INSTRumental NPs is the fact that the language, like Tagalog, possesses three distinct series of enclitic pronominal forms, one corresponding to the PIVot, one to the_x_5 ACC marked NPs, and one for s INSTR NPs (note that the third person enclitics are identical to the corresponding NP case markers):

| | | PIV | x ACC | s INSTR | | | --------- | -------- | ------- | --------- | ------ | | | SG | -@n( ¡) | f | -@n( ¡) | | | 1 | PL INCL | -@nts | f | -@nts | | | PL EXCL | -@nu?Xw | f | -@nu?Xw | | | 2 | | -@s | -o ¡ | -os | | 3 | | f ~ -i | -q ~ -X | -s |

Crucially, when first or second person actor does not function as _PIV_ot, it is still realized as a bound pronominal affix on the verb (Boas 1947: 286):

Data like (54) strongly argue that non-PIVot actors in Kwakwala are indeed CORE arguments. OBLique NPs are not realized by bound or enclitic pronominals crosslinguistically. And note further the behavior of the Kwakwala INSTRumental enclitic pronominals is exactly parallel to Tagalog CORE_ng_ pronominals; both encliticize to the first major constituent of the clause. Indeed as remarked earlier, Tagalog OBLique sa pronominals never encliticize. The again very striking parallelism between Tagalog and Kwakwala on this feature argues that non-PIVot actors in both languages should be assigned the same syntactic status: CORE NPs.

3.3 Summary of symmetrical voice system properties

On the basis of the pervasive similarities between Philippine languages and Kwakwala we may propose the following properties as diagnostic of symmetrical voice languages:

1. Voice forms which make nonsubcategorized arguments like locatives and instrumentals directly accessible to PIVot function without going through an intermediate APPLicative derivation. This is criterial for the symmetrical voice type provided the language is not otherwise of the ergative-absolutive type, in which case the constraint against double ABSolutive NPs may be operative.

2. No neutral voice forms of the verb, all being equally marked as derived forms, although one of the voice affixes may be a zero morpheme. A logical consequence of property (2) would be symmetrical voice languages in which semantic classes of verbs vary as to the PIVot choice signalled by zero, in contrast to Kwakwala in which zero always indicates actor as PIVot. If the zero morpheme is just a lack of overt phonological material in paradigmatic opposition with other parallel morphemes having overt phonological realizations, then it is fully to be expected that various verbal semantic classes may differ as to which voice variant bears the zero form. Such languages do in fact exist; for example Bilaan of the Southern Philippines (Abrams 1961; Rhea 1972):

(55) klang �cut�

(56) bat �throw�

(57) subè �go upstream�

3. Actor participants when they fail to appear as PIVot continue to appear as CORE NPs. The same applies to undergoer participants in voice constructions in which non-subcategorized erstwhile OBLique participants like locative or instrumentals function as PIVot (compare examples (1d) and (50c)).

4. There is a lack of a clear noun/verb contrast in the lexical roots of the language.

It is this last feature which I have identified as the lynch pin for the whole symmetrical voice type: if roots are precategorial, neither verb nor noun, they necessarily lack lexical properties of verbs, such as subcategorized participants in argument structures. Choosing a participant as PIVot requires an argument structure array or participants to choose among, and this in turn requires a verb. In symmetrical languages with precategorial roots the whole process is done simultaneously with one derivation. All words in all languages have a semantic representation of lexical conceptual structure, and let us assume these are roughly similar across languages, so that English give, Tagalog bigay and Kwakwala _c�@_- all would have a lexical conceptual structure roughly along the lines of �X causes Y to have Z�. Because English give is a verb this maps directly onto an argument structure, along the lines of (58):

Tagalog and Kwakwala, on the other hand, with precategorial roots do not have an argument structure tied to the root form; there is only the l-c structure:

Argument structures accrue to the roots when they are derived as verbs and at the same time pivot selection is made:

| | c'@- | l-c structure | X | cause | Y | to have | Z | | ------------- | --------------------- | ----- | ----- | ----- | ------- | ----- | | | | | | | | | | | | | | | | | Act | | Und | | | | | | | | | | | | | | | | | | c'u-s@w?- | arg structure | <X, | | Y, | | Z> | | | | | | | | | | | | | | | | grammatical relations | CORE | | CORE | | PIV | | | | | (s) | | (s) | | (i) |

A major typological break, right down to the structure of the lexicon, is thus claimed between symmetrical and asymmetrical languages. This obviously has important further typological and theoretical implications, what will now be addressed.

Created: 4 June 1998
Last modified: 8 June 1998

Authorised by: P. Austin, Professor and HOD, Linguistics and Applied Linguistics

Access: International
Copyright © 1998, The University of Melbourne.
Maintainer:Simon Musgrave