Preposition meaning: syntactic reflex and semantic relation coverage

The Oxford Dictionary of English defines preposition as “a word governing, and usually preceding, a noun or pronoun and expressing a relation to another word or element in the clause”.  A few years ago, Bill Dolan, quoting Lucy Vanderwende (both of the Microsoft NLP group), suggested that

prepositions are essentially just syntactic reflexes that have no real meaning of their own, only taking on meaning in the context of a larger syntactic pattern

More recently, Ed Hovy cited Jerry Hobbs that developing a hierarchy of semantic relations is

a fool’s errand; that each relationship between two different concepts is unique, and that one can produce some generalizations and a relation hierarchy, but you could never cover everything

Based on some recent work, I would like to put these opinions into some deeper context.

The first preposition listed in The Preposition Project Online Dictionary is a cut above, with the meaning “noticeably superior to”.  An example of its use is

This test is a cut above most of the silly self-evaluation tests one finds on the Web.

In this case, the preposition relates two elements, This test and most of the silly self-evaluation tests one finds on the Web. Without the preposition, however, the nature of this relation would not be present. In this case, the preposition can be said to trigger the FrameNet frame Surpassing, with the preposition object filling the Standard_item frame element and the subject of the sentence filling the Profiled_item frame element. Lest one attribute the triggering to the special case of an idiomatic, phrasal preposition, note that above and over have similar senses that would likewise trigger this frame, without any sufficient context where this could be called a syntactic reflex.

However, the syntactic reflex of preposition meaning is also clearly present. The UMLS Specialist Lexicon includes a vast number of prepositions that are closely related to specific verbs, nouns, and adjectives (almost 28,000). There are 30 instances of between shown as a preposition complement of verbs such as alternate, choose, distinguish, oscillate, and waver. Now, what is interesting is to look at the definitions of these verbs.  The verb alternate has one definition “change repeatedly between two contrasting conditions” and oscillate has a definition “vary or fluctuate between two states, limits, opinions, etc.” Notice that the definitions contain the preposition between with an object that serves as a placeholder. These senses, therefore, seem to demand the use of this preposition in their use. The Oxford sentence dictionary shows 16 of 20 examples for alternate and 17 of 20 examples for oscillate using between. (An open question is whether these verb senses dictate which of the 9 senses of between should be selected.) Similarly, a search of verb definitions in the Oxford dictionary shows a total of 66 cases where between appears in a similar form, of a slot with a placeholder object. Thus, the UMLS set (constructed by hand) could easily be extended to include these other verbs that may be said to take between as a syntactic reflex.

O’Hara and Wiebe (2009) spend considerable effort in attempting to piece together a workable semantic relations inventory. In addition to FrameNet, they examine semantic relations from the Penn Treebank, the Factotum
knowledge base, Cyc, and Conceptual Graphs. While they eventually develop an inventory using primarily FrameNet, the kinds of semantic relations introduced by the other sources seem to be not relevant to prepositions. Semantic relations used as the backbone of WordNet, such as synonymy, antonymy, and meronymy, are also relatively rare as meanings of prepositions. (The core WordNet relation of hypernymy may be found in one sense of of.) However, in general, the presence of these relations as meanings of prepositions is rare.

I suggest that the main reason that many semantic relations are not to be found among preposition meanings is that prepositions only encode a small subset of all the semantic relations. I suggest that the richness of the English language is embodied in the verbs, each of which also constitutes a semantic relation between a subject and an object. This seems to me to be the main reason why Hobbs’ observation is true.

The conclusion I reach from all of this is that there is meaning to be contributed to an utterance by prepositions. Frequently, this meaning is strongly tied to the meanings of other words. And finally, we cannot expect prepositions to cover all the semantic relations. So, we analyze what we have and try to make sense of the contributions from prepositions.