A phrase is a sequence of one or more words (in some theories two or more) built around a head lexical item and working as a unit within a sentence.
15 of the most commonly used tests are listed next: 1) coordination (conjunction), 2) pro-form substitution (replacement), 3) topicalization (fronting), 4) do-so-substitution, 5) one-substitution, 6) answer ellipsis (question test), 7) clefting, 8) VP-ellipsis, 9) pseudoclefting, 10) passivization, 11) omission (deletion), 12) intrusion, 13) wh-fronting, 14) general substitution, 15) right node raising (RNR).
A general word of caution is warranted when employing these tests, since they often deliver contradictory results.
The tests are merely rough-and-ready tools that grammarians employ to reveal clues about syntactic structure.
It is best to apply as many tests as possible to a given string in order to prove or to rule out its status as a constituent.
The analyses in these two tree diagrams provide orientation for the discussion of tests for constituents that now follows.
Based on these data, one might assume that drunks, could, put off, and customers are constituents in the test sentence because these strings can be coordinated with bums, would, drive away, and neighbors, respectively.
However, additional data are problematic, since they suggest that certain strings are also constituents even though most theories of syntax do not acknowledge them as such, e.g.
Data such as (h-j) are sometimes addressed in terms of the right node raising (RNR) mechanism.
The problem for the coordination test represented by examples (h-j) is compounded when one looks beyond the test sentence, for one quickly finds that coordination suggests that a wide range of strings are constituents that most theories of syntax do not acknowledge as such, e.g.
Due to the difficulties suggested with examples (h-m), many grammarians view coordination skeptically regarding its value as a test for constituents.
If such a change yields a grammatical sentence where the general structure has not been altered, then the test string is likely a constituent:[6] These examples suggest that Drunks, the customers, and put off the customers in the test sentence are constituents.
An important aspect of the proform test is the fact that it fails to identify most subphrasal strings as constituents, e.g.
These examples suggest that the individual words could, put, off, and customers should not be viewed as constituents.
This suggestion is of course controversial, since most theories of syntax assume that individual words are constituents by default.
[8] The test is limited in its applicability, though, precisely because it is only applicable to strings containing verbs: The 'a' example suggests that put off the customers is a constituent in the test sentence, whereas the b example fails to suggest that could put off the customers is a constituent, for do so cannot include the meaning of the modal verb could.
There is a problem with this sort of reasoning, however, as the next example illustrates: In this case, did so appears to stand in for the discontinuous word combination consisting of met them and because we had time.
The answer fragment test involves forming a question that contains a single wh-word (e.g. who, what, where, etc.).
It is apparently often impossible to form the question in a way that could successfully elicit the indicated strings as answer fragments.
Example c is of dubious acceptability, suggesting that put off the customers may not be constituent in the test string.
The VP-ellipsis test checks to see which strings containing one or more predicative elements (usually verbs) can be elided from a sentence.
Another problem is that the test can at times suggest that a discontinuous word combination is a constituent, e.g.: In this case, it appears as though the elided material corresponds to the discontinuous word combination including help and in the office.
The passivization test used in this manner is only capable of identifying subject and object words, phrases, and clauses as constituents.
Omission checks whether the target string can be omitted without influencing the grammaticality of the sentence.
In most cases, local and temporal adverbials, attributive modifiers, and optional complements can be safely omitted and thus qualify as constituents.
Omission used in this manner is of limited applicability, since it is incapable of identifying any constituent that appears obligatorily.
These examples suggest that the strings Drunks, the customers, and could are constituents in the test sentence.
The first is that it is limited in its applicability, since it is only capable of identifying strings as constituents if they appear on the right side of the test sentence.
Constituent structure analyses of sentences are a central concern for theories of syntax.
Observe in this regard that strictly binary branching analyses increase the number of (overt) constituents to what is possible.