As for parts of speech, I am quite sure it is not the case.

For instance, some languages are problematic in separating clearly verbs from adjectives like Japanese and Korean, some native American languages seem to have much less categories than Indo-European. Since each part of speech corresponds to a phrase, the phrases can't be universal either.

Now grammatical relations are like roles certain phrases can take, and it strikes me intuitively that these roles are limited by logic and can't allow for idiosyncrasies like parts of speech can.

But is it really so, are not even some sentence parts universal among languages but each language has the same set of them?


1 Answer 1


I assume, based on the your posts elsewhere, that by 'sentence parts', you are referring to grammatical relations (GRs) like subject, object, etc. In the future, it would be clearer for you to call them that, as this is the standard terminology in English.

I think you are confused about argument structures and grammatical relations. Argument structures specific semantic roles like agent, patient, etc. These may be universal, but they do not have bijective maps with grammatical relations. Indeed, what motivates the use of GRs in linguistic analysis is the lack of one-to-one correpsondence between semantic roles and certain grammaticalised aspects of language (e.g. nominative case). Moreover, although argument structure can be easily expressed in terms of traditional predicate logic, grammar is not bound by the rules of any system of logic, so this is not a good argument for the universality of GRs.

With that said, I'd like to split the question into three different ones:

1. Do there exist necessary and sufficient conditions that allow us to conclusively identify GRs in any language?

The answer to this is definitely no. There are no objective criteria to help us pick out subjects, objects, indirect objects, oblique arguments, etc. in any language, in the same way that we can identify chemicals by spectroscopy. A set of criteria that works for English is not going to work for Chinese, and a set of criteria that works for Lakhota is not going to work for Swahili.

In this sense, GRs are not universal. We cannot give a formal characterisation of GRs that work for any language.

2. Is it possible to identify GRs in any language which are subject-like, object-like, etc.?

By this I mean: given a set of prototype characteristics of subjects and objects, can we identify subjects and objects in any language? The answer to this is uncertain. Certain languages, such as Archi and Acehnese, have been claimed by some linguists to lack GRs, and the role usually played by GRs can be replaced entirely be semantics and pragmatics.

However, in most languages it is possible to identify certain elements as subjects, objects, etc. based on a) overt coding (word order, case, etc.) and b) behaviour-and-control properties (pivots, control, raising, etc.) But it must be noted that none of these are necessary-and-sufficient conditions for GRs; GRs are a fuzzy concept here. A classic example is Keenan (1976), which gave a list of 30 characteristics of subjects. So, under this characterisation of universality, yes, all languages could have the same set of GRs; but this set is one of fuzzy categories.

3. Are there linguists working with the assumption that GRs are universal?

Yes; GRs are central to lexical-functional grammar (LFG) and relational grammar (RG); in these theories, GRs are assumed to be theoretical 'primitives'. In RG, grammatical relations are called 1, 2, 3; LFG has a fancier hierarchy of GRs: SUBJ > OBJ > OBJ(theta) > OBL(theta) > COMP > XCOMP > ADJ > XADJ.

Moreover, in typological work it is often necessary to assume the notion of grammatical relations, or it would impossible to make word order generalisations; however, this is purely methodological and does not involve 'reifying' GRs or assuming their universality.

  • Wonderful clarification Commented Mar 17, 2017 at 19:29
  • 1
    Acehnese doesn't lack GRs; it simply uses a different set. Indonesian languages generally have neither an accusative ("subject/object") nor ergative ("agent/patient") GR system; in Ac, they use three GRs instead of two: Transitive Subject, Intransitive Subject, Direct Object.
    – jlawler
    Commented Mar 17, 2017 at 19:49
  • 1
    @jlawler Sorry if I was unclear - to clarify, I was just citing the opinions of certain linguists holding the opinion that Acehnese lacks GRs (Van Valin and LaPolla 1997:255-261; they cited Durie's grammar there). I'm aware that you're an expert on the language, have authored several classic papers on it, and did identify GRs there (though you rejected the notion that the usual subject exists in Acehnese - which still means the answer to the question is negative)! Commented Mar 17, 2017 at 20:25
  • 1
    Like phonemes and syntactic rules, grammatical relations are unique to each language, once you look at enough of the details. There are a lot of contributing phenomena. For Austronesian languages, the reason for the odd GRs is that relative clauses are constrained very tightly, so there have to be a lot of ways for a given NP to be made some kind of subject so that relativization can work. This leads to multiple phenomena that get labelled "passive"; in Malagasy Keenan found seven different kinds, iirc.
    – jlawler
    Commented Mar 17, 2017 at 20:53
  • @jlawler I definitely agree with that. BTW, could you suggest any good references for the Austronesian voice system? I've been wanting to get into Austronesian syntax for a while. Thanks! Commented Mar 18, 2017 at 4:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.