# What is wrong with this way of looking at conjunctions?

What is wrong with the way of looking at conjunctions that I have tried to illustrate in the diagram below, in which a girl is simultaneously the object of saw and greeted, and the function of and is to convey the parallel relationship of Frank saw and Bill greeted that makes this possible?

Is this the same as the "large conjunct approach"?

Related question (actually it's more like a related answer, by Tim Osborne). The answer seems to be based on the idea that this structure would have to expand out to Frank saw a girl and Bill greeted a girl, which is a way of saying that the parallel relationship in which the object is shared is just not possible - but why not?

• With independently conjoined NPs and VPs, there's a lot of room for ambiguity. A lot depends on whether one cares about that, and if so, how one wants one's theory to account for it. Commented Mar 28, 2020 at 20:40
• why do questions about this very same structure appear every month and again? it is difficult to search for, I guess. I'd welcome if these were closed as duplicates, unless I'm missing the finer details Commented Mar 29, 2020 at 14:23

I may have misunderstood the earlier comments (by Rchivers). The type of approach to coordination described with the diagram in the question is indeed how I prefer to view coordination; the diagram indicates what I construe to be a 'small conjunct' approach. The coordinated strings are parallel and the coordination helps mark which strings are interpreted as parallel. Crucially, however, the coordinated strings in this case are NOT constituents, which is a problem for most theories of syntax.

The large conjunct approach, which stands in opposition to such a small conjunct approach, attempts to address such data in terms full underlying clauses. For instance:

(1) [Frank saw {a girl}] and [Bill greeted a girl].

The pointy brackets { } indicate material that has been elided or deleted and is therefore not actually present on the surface. It should be apparent why such an analysis can be called a 'large conjunct' approach. Exactly what the underlying mechanism is called that elides or deletes missing material varies. Some might call it Conjunction Reduction or Right Node Raising or...

The problem with the large conjunct approach is that it suggests a meaning should be present that is often not available. Consider the contrast in meaning across the next two sentences:

(2) [Larry scrounged together] and [Sam borrowed] a total of 1000 dollars.

(3) [Larry scrounged together {a total of 1000 dollars}] and [Sam borrowed a total of 1000 dollars].

The analysis indicated in (3) assumes large conjuncts, that is, the conjuncts are complete clauses underlyingly and an ellipsis or deletion mechanism reduces the first clause down to its surface appearance. In contrast, the analysis indicated in (2) assumes small conjuncts, which means no appeal to ellipsis or deletion is made. It should be apparent that only the small conjunct approach captures the intended meaning, which is such that there is 1000 dollars altogether, not 2000 dollars.

Given that the small conjunct approach comes out ahead in this area, it is nevertheless challenged in major ways. Above all, it has to account for the nature of the non-constituent strings that can and cannot be coordinated.

Finally, note that most phrase structure grammars in general are challenged in major ways by such data, since the implication is that the phrase/constituent is not the crucial unit of analysis that one needs to establish a tractable approach to coordination.

• Good example there with the "total of 1000 dollars". It's twisting my mind how a trace-based generative syntax would account for it! Commented Mar 29, 2020 at 3:06
• Large Conjunct analysis requires that the elided material be definite (it is the same girl). This is difficult to paraphrase since the elided item precedes the remaining one -- so it isn't simply a case of elision.
– amI
Commented Mar 29, 2020 at 6:41
• Thanks for that comprehensive answer. Do you see a fundamental difference between this analysis of conjunctions and the analysis of constructions like washer-dryer and paperweight-cum-ashtray you put forward in a recent thread, or do you think they are cases of the same mechanism operating at word level? Commented Mar 30, 2020 at 10:51
• @rchivers I think its the same mechanism working within complex words, as you suggest. I have not, however, produced anything in writing about the mechanism inside words, though. Perhaps a project for the future. Commented Mar 31, 2020 at 5:48

I think every linguist should know about Gerald Gazdar's and Categorial Grammar's solution to this problem, which is also used in Montague Grammar. It adopts a proposal of Ajdukiewicz (which see) to admit syntactic categories which are missing a constituent. A notation for this is to use a slash to show the missing category.

For the example, this gives S/NP, meaning "sentence missing a NP", as the category of both "Frank saw" and "Bill greeted", so the fact that these two can be coordinated with "and" is simply a consequence of the general fact that two constituents of the same category can be coordinated.

• What if the coordinated strings are missing more than just one NP constituent, e.g. Frank gave no girl [flowers today] and [chocolates yesterday] ? Commented Mar 31, 2020 at 12:13
• @TimOsborne, In general, in GPSG, it is not possible for a consitituent to be missing more than one constituent. That, details aside, is why the CNPC (the Complex NP Constraint of Ross) prohibits extraction from a relative clause -- a relative clause is already missing one thing, so it can not be missing another. I don't now see how your exam[le could be derived (nor do I think it is grammatical). Commented Mar 31, 2020 at 20:47
• What's the objection to the grammar? Are the coordinated strings in she gave him an iPad for his birthday and a bike for Christmas also missing more than one constituent, or have I misunderstood? Commented Apr 1, 2020 at 10:47
• @rchivers "What did she give him? An ipad for his birthday." What is missing? Apparently, nothing, much less than two things. So I am not following your question. Commented Apr 2, 2020 at 2:59
• @TimOsborne, Please let me amend my answer. I don't see how it makes a difference here, but there is a way to describe constituents missing more than one thing in GPSG: parasitic gaps are additional coreferential gaps. Commented Apr 2, 2020 at 3:07