ESD Test Suite Examples
The garden dog barked. The Abrams picture arrived. Kim Browne arrived. Professor Browne arrived. The tobacco-happy dog barked. The state and local dogs barked.
Compounding comprises a variety of (semantic) head–modifier structures that can often be paraphrased using overt prepositions. But in instances of compounding there is a syntactic construction contributing an underspecified two-place relation. Medicine delivery, for example, could be paraphrased as delivery of medicine, whereas home delivery might be paraphrased as delivery at home. The phenomenon is characterized by the underspecified compound relation, whose shape and use are parallel to the semantics contributed by regular prepositions. While prototypical heads in compounds are nominals (introducing an instance), there can be exceptions: tobacco-happy is analyzed in terms of an event head. Compounding can interact with coordinate structures (e.g. in state and local dogs), when modifiers of the same head are conjoined.
The phenomenon can be sub-divided according to the (semantic) types of the head and modifier. The most basic instances, arguably, involve non-quantified or quantified modifiers to a nominal head, e.g. garden dog and Abrams picture, respectively.
h0:compound[ARG1 x1, ARG2 x2] h0:[ARG0 x1] [ARG0 x2]
With event heads, the structure is very similar, except for the type of the ARG1 in the compound relation, e.g. for tobacco-happy:
h0:compound[ARG1 e1, ARG2 x1] h0:[ARG0 e1] [ARG0 x1]
When conjoining different types of modifiers (to a nominal head), compounding and coordination interact in syntactically subtle ways, but result in a comparatively straightfoward semantics, e.g. for state and local dogs:
h0:compound[ARG1 x1, ARG2 x2] [ARG0 x2] h1:[ARG0 e1, ARG1 x1] h2:_and_c[L-HNDL h0, R-HNDL h1] h2:[ARG0 x1]
Up until its 1214 release, the ERG made a distinction between compound and compound_name relations, but seeing that proper name heads in a compound construction can be unambiguously identified by their (proper_q) quantifier, this distinction was judged redundant.
The compound relation is also used in what Bender, et al. (2011) call the N-ed construction, as for example in rabbit-eared dog. Here, there are in fact two instances of the two-place compound relation, one relating the two elements of the N-ed construction (ear and rabbit), and another one connecting the construction as a whole to its head noun (i.e. dog). As the N-ed construction allows predicative use, this is a rare case where compound can appear with its ARG1 realized as a syntactic subject. A deeper understanding of which nouns should be treated as relational may lead to revisions of this analysis. (For a dissenting view, arguing that no nouns are relational, see Payne et al 2013.)
The N-ed construction is not the only construction where the 'compound' relation can appear in predicative position, though it is the only one where the subject of the predication is an argument of the 'compound' relation. In a sentence like The mountain is snow-covered, the relationship between '_cover_v' and '_snow_n' is expressed as a 'compound', to express the underspecified relation between the event and the instance (here most naturally paraphrased by the preposition with). The intent of the 'compound' with N-ed (used attributively or predicatively) is to provide the necessary underspecified relation that holds between the instance of the noun suffixed with '-ed' and the instance of the noun that is either modified or the subject of predication. Those two instances have to be linked together somehow, and the only clear alternative would be to have every N-ed noun introduce a two-place relation (as is the case in the ERG for kinship terms, for example), where the instance of the noun modified or predicated of would be the ARG1 of the N-ed noun's relation. This might be no more than a notational variant of the current 'compound' representation, since there would be a simple equivalence rule that converts from the one to the other.
Bender, E. M., Flickinger, D., Oepen, S., & Zhang, Y. (2011, July). Parser evaluation over local and non-local deep dependencies in a large corpus. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (pp. 397-408). Association for Computational Linguistics.
Payne, J. & Pullum, G. K. & Scholz, B. C. & Berlage, E.(2013). Anaphoric one and its implications. Language 89(4), 794-829. Linguistic Society of America. Retrieved April 22, 2014, from Project MUSE database.