Handbook of Grammatical Relations

Questionnaire (September 2013)

To insure comparability of individual accounts in the volume, we suggest to use as a guideline the approach to grammatical relations outlined in Bickel (2011a) and further developed in Witzlack-Makarevich (2011). In what follows, we provide a brief overview of this approach. Please, study this section carefully before answering the questions of the questionnaire.

The view of grammatical relations adopted here is characterized by a radical shift of attention from such generalized notions as subject or pivot to single characteristics or properties of the relevant phenomena. In this way, grammatical relations are reconceptualized as equivalence sets of arguments that are treated the same way (i.e. “aligned”) by an argument selector (any morphosyntactic construction or pattern) under certain conditions. We briefly discuss these three aspects in turn.

Arguments

Arguments are complex categories defined by a generalized semantic role that may or may not be subject to lexical and referential specifications. Before discussing various argument types, it is necessary to be able to distinguish arguments from non-arguments.

A dependent expression is an argument of a predicate if its role in the situation is assigned by this predicate. This is not the case for adjuncts. Seen this way, the argument/adjunct distinction is exclusively semantic and independent of the way a dependent is expressed. Also, the argument vs. adjunct distinction is orthogonal to the question of whether a clausal dependent is syntactically obligatory or omissible. Arguments are frequently omitted in most languages.

Though the basic intuition behind the argument/adjunct distinction is relatively clear, difficulties can arise as soon as one tries to distinguish the two in individual cases. In response to this, a number of tests have been suggested in the literature to make the decision easier (for an overview, see Comrie 1993). One common test is constructions with pro-verbs, such as do so, do it or do the same thing in English (cf. Helbig and Schenkel 1991:37 on German). A clause with an adjunct can be paraphrased in such a way that an adjunct is expressed in a different clause with a pro-verb do so etc., replacing the verb together with its arguments but excluding any adjunct:

Along the same line, do so can be used anaphorically for at least a verb and its arguments, as in the following examples (for further examples and discussion see Culicover and Jackendoff 2005):

Whereas the test above might be applicable in many languages, some other tests are more language specific. For instance, in English to + NP can be both an argument and an adjunct. With verbs such as go (to X) X is an argument. This is evident from the fact that go assigns a goal role even in the absence of the preposition to, as for instance in Where did she go?, where where must be interpreted as a goal. This is different from motion verbs which take no goal argument, as for instance walk: where did she walk?, in which where can be either a location or a goal.)

Ultimately, the distinction between arguments and adjuncts requires a thorough lexical-semantic analysis of individual verbs, a task that is ultimately orthogonal to and independent from the present project. Obviously, it might be impossible to carry out such an analysis for every verb within the framework of the project. In this case we recommend that you limit your attention to the verbs you are certain about and explicitly state for which verb groups you are unable to distinguish arguments from adjuncts with reasonable certainty.

Generalized semantic roles

Once arguments have been distinguished from adjunct, one needs to further distinguish between individual arguments. Generalized semantic argument roles are identified first by numerical valence: the sole argument of one-argument predicates, the two arguments of two- argument predicates, and the three arguments of three-argument predicates. In case of the sole argument of one-argument predicates, there is no need to distinguish it from anything else; this argument is abbreviated as S.

In case of two- and three-argument predicates, arguments are distinguished on the basis of cross-linguistically viable semantic entailment properties (cf. Bickel and Nichols 2009, Bickel 2011b, Witzlack-Makarevich 2011, based on and inspired by Dowty 1991 and Primus 1999, 2006).

‘A’ stands here for the A argument of two-argument predicates only. Three-argument predicates have an “A_ditr” argument (Bickel and Nichols 2009, Bickel 2011b), and this is distinguished from T and G in the same way as A is distinguished from P arguments. Many, perhaps most, languages treat A_ditr in exactly the same was as the A argument of two-argument predicates, but this needs to be established for each language. Note, however, that the difference may be relevant only for a subclass of predicates (e.g. the A of two-argument predicates may be marked as dative under certain conditions, while this option may be absent from A_ditr; or ergatives may be compulsory on A_ditr but not on A).

Lexical specifications of arguments

Our proposal identifies generalized semantic roles like ‘S’, ‘A’, ‘P’, ‘A_ditr’, ‘T’ and ‘G’ for most lexical predicates in each language. The advantage is that these roles are not limited to what one might want to think of as (universally) “prototypical” or “canonical” meanings. But in return, it becomes necessary to state any lexical specification of the generalized roles. Chechen for example, distinguishes between several lexical types of S, A and P. These have different effects on case marking, e.g. some one-argument verbs assign the absolutive to the S argument, others assign datives to the S; some predicates assign ergatives and nominatives to A and P respectively, others datives and nominatives etc. The relevant classes of verbs are therefore close to the traditional concept of ‘valency classes’, but it is important to emphasize that classifications are defined by each argument selector separately, and that there is no a priori expectation that for example the classes relevant for case frames are identical to lexical distinctions relevant for agreement. A well-known example is the distinction between various lexical subtypes of S that determine different agreement forms in Muskogean languages but which are largely irrelevant for case marking in the same languages (cf. Broadwell 2006 on Choctaw).

Lexical specifications of S are typically discussed under the rubric of split-intransitive systems and fluid-intransitive systems, and various subtypes have been labelled ‘unergative’ and ‘unaccusative’ verbs, ‘active alignment’ or ‘stative-active’, ‘agentive’ or ‘agent-patient’, ‘semantic alignment’, and ‘split S’ (cf. Merlan 1985, Dixon 1994, Levin and Rappaport Hovav 1995, Van Valin and LaPolla 1997, Croft 1998, Donohue and Wichmann 2008).

Referential specifications of arguments

Apart from the generalized semantic role properties and lexical specifications of the predicate, argument marking or behavior then depends on referential properties in the broad sense and including such categories as definiteness, topicality, specificity, animacy, part-of-speech properties, etc. The resulting situation has been investigated under a variety of labels. The most common general terms include ‘split’ (Silverstein 1976) or, more specifically, ‘split conditioned by semantics of NPs’ (Dixon 1994), ‘differential marking’ (Comrie 1989) or ‘differential argument marking’ (Baerman 2008).

Among specific manifestations of splits, the best studied patterns are splits in the marking of the P argument commonly referred to as ‘differential object marking’ Bossong (Bossong 1982, 1985, 1998) and of the A argument called ‘split ergativity’ (Comrie 1978, Dixon 1979) or ‘differential subject marking’ (de Hoop and de Swart 2008). The effects of referential properties on argument marking are then summarized in terms of referential hierarchies of various shapes, also known as ‘agency’, ‘animacy’, ‘empathy’, ‘indexability’, ‘ontological salience’, ‘person’ and ‘prominence’ hierarchy (cf. Silverstein 1976, Comrie 1979, 1989, Bossong 1985, Croft 1990, Lazard 1998, Aissen 2003).

Argument selectors

For the purposes of this questionnaire, argument selectors refer to any morphosyntactic structure, process, rule, constraint or construction that selects a subset of arguments (and possibly non-arguments) and treats them differently from other arguments (or non-arguments) of the clause. In order to qualify as an argument selector a particular morphosyntactic structure, process or rule must display a specific constraint as to which arguments it applies to, e.g. only to A, or to S, A, and A_ditr, or only to S, A, P, A_ditr, T, and G, but not adjuncts. To illustrate the difference between a genuine argument selector and a construction that might look like an argument selector but does not qualify as one, consider the following examples (based on Comrie 1988 and LaPolla 1993; also cf. Bickel 2011b):

In all these examples, the second coordinate clause lacks an overt argument. In all cases, the silent argument is either the S argument or the A argument. These silent controllees are obligatorily interpreted as being coreferential with either the S or A argument of the first clause. The interpretational constraint can even override pragmatic plausibility, as in the fourth example, where the second clause can only be interpreted as referring to the situation where Bob burst, however unplausible this is in the real world.

Constructions like these constitute argument selectors because they impose a strict constraint on arguments. It is important not to confuse such selectors with similarly-looking phenomena which do not impose any syntactic constraints on obligatory coreference and whose interpretation relies wholly on previous discourse and our knowledge of the world. The following examples from Mandarin Chinese illustrate this. The deleted argument is interpreted as referring to the watermelon, as this is the most plausible scenario based on our world knowledge. The same is true for the second example, where the silent argument of the second clause is interpreted as referring anaphorically to the man, because watermelons normally do not get flustered (see Bickel and Yādava 2000 for examples from other languages):

Crucially, the Chinese constructions illustrated by these data are not argument selectors.

Clause-level conditions

Whether a certain argument is selected by a particular selector is not only determined by the nature of the argument and its lexical or referential specifications. A number of other clause-level properties can influence the inclusion or exclusion of the argument as well, resulting in various additional splits. The conditions on splits can be of a number of types. The most wide-spread conditions include the following:

Most conditions are well-established in the literature (see e.g. Dixon 1994, Bickel 2011b). What is less well-known is scenario conditions, and we illustrate this briefly here. In some traditions, scenario conditions are treated under the rubric of ‘hierarchical alignment’ (cf. Mallinson and Blake 1981, Nichols 1992, Siewierska 1998). The basic phenomenon is that argument selectors include information about more than one participant. This can bring with it a notion of an ontological hierarchy of participants that compete for a specific selector (e.g. whichever participants is higher on the animacy hierarchy triggers agreement), but the phenomenon of scenario conditions is more general and refers to any kind of condition that makes reference to the whole constellation of arguments (‘who is acting on whom’), not only those which can be stated in terms of competition along a hierarchy.

Questionnaire

The questionnaire is organized in such a way as to encourage systematical collection of information about any argument selector of a language, i.e. about any rule, construction or restriction that shows sensitivity to grammatical relations. However, we do not want to reduce this questionnaire to a check-list of known argument selectors (e.g. case, agreement, or conjunction reduction) and known conditions on split alignment (e.g. aspect), so contributors should feel free to extend the list of argument types, argument selectors and conditions with any information that they find of relevance for grammatical relations in your language.

Please try to provide an example of each argument selector and each condition discussed. Also include examples of three-argument predicates if the language has any. When giving examples, please use the Leipzig glossing rules for glossing conventions.

For every argument selector present in the language you describe please provide at the mini- mum an explicit description of

As we are interested in any restrictive mechanisms of a language, the list of possible selectors can be very long and often includes among others the following:

For discussion and examples of these and other argument selectors please consult Dixon (1994), Van Valin (2001), Bickel (2011a), Van Valin (2001), Bickel (2011a), Witzlack-Makarevich (2011).