4.1 Using labels

• patterns can be labelled (similar to the target marker @)

  > adj:[pos = "JJ.*"] ... ;

  the label adj then refers to the corresponding token (i.e. its corpus position)

• label references are usually evaluated within the global constraint introduced by ::

  > adj:[pos = "ADJ."] :: adj < 500;
  $\to$ adjectives among the first 500 tokens

• annotations of the referenced token can be accessed as adj.word, adj.lemma, etc.

• labels are not part of the query result and must be used within the query expression (otherwise, CQP will abort with an error message)

• labels set to optional patterns may be undefined

  > [pos="DT"] a:[pos="JJ"]? [pos="NNS?"] :: a;
  $\to$ global constraint a is true iff match contains an adjective

• to avoid error messages, test whether a label is defined before accessing its attributes

  > [pos="DT"] a:[]? [pos="NNS?"] :: a -> a.pos="JJ";

  (-> is the logical implication operator $\to$, cf. Section 2.6)

• labels are used to specify additional constraints that are beyond the scope of ordinary regular expressions

  > a:[] "and" b:[] :: a.word = b.word;

• labels allow modelling of long-distance dependencies

  > a:[pos="PP"] []{0,5} b:[pos = "VB.*"]
:: b.pos = "VBZ" -> a.lemma = "he|she|it";

  (this query ensures that the pronoun preceding a 3rd-person singular verb form is he, she or it; an additional constraint could exclude these pronouns for other verb forms)

• labels can be used within patterns as well

  > a:[] [pos = a.pos]{3};
  $\to$ sequences of four identical part-of-speech tags

• however, a label cannot be used within the pattern it refers to; use the special this label represented by a single underscore (_) instead to refer to the current corpus position

  [_.pos = "NPS"] $\Longleftrightarrow$ [pos = "NPS"]

• the this label can also be used to constrain tokens to a certain range of corpus positions without explicit labels, e.g.

  > [pos = "ADJ." & _ < 500];

  such constraints are not allowed in query-initial position, so queries such as [_ >= 666]; and [_ < 500 & pos = "ADJ."]; will be rejected as invalid

• new in CQP v3.4.17: as a special case, the pattern

  > [_ = 666];

  can be used to look up a known corpus position efficiently

• the built-in functions distance() and distabs() compute the (absolute) distance between 2 tokens (referenced by labels)

  > a:[pos="DT"] [pos="JJ"]* b:[pos="NNS?"] :: distabs(a,b) >= 5;
  $\to$ simple NPs containing 6 or more tokens

• the standard anchor points (match, matchend, and target) are also available as labels (with the same names)

  > [pos="DT"] [pos="JJ"]* [pos="NNS?"] :: distabs(match, matchend) >= 5;

• various other built-in functions have been added in recent versions of CQP and can be used with label references or directly with attribute values; see Sec. 8.3 for a complete list

• new in CQP v3.4.17: use function strlen() to filter by word length, e.g. to find particularly long words:

  > [word = ".*ment" & strlen(word) >= 16];

• NB: inequality comparisons (>, >=, <, <=) are only allowed for integers (corpus positions, string lengths, etc.), but not for strings and regular expressions; CQP versions before v3.4.17 used to silently accept and misinterpret such inequality comparisons