8.1 Zero-width assertions

• constraints involving labels have to be tested either in the global constraint or in one of the token patterns; this means that macros cannot easily specify constraints on the labels they define: such a macro would have to be interpolated in two separate places (in the sequence of token patterns as well as in the global constraint)

• zero-width assertions allow constraints to be tested during query evaluation, i.e. at a specific point in the sequence of token patterns; an assertion uses the same Boolean expression syntax as a pattern, but is delimited by [: ... :] rather than simple square brackets ([...]); unlike an ordinary pattern, an assertion does not “consume” a token when it is matched; it can be thought of as a part of the global constraint that is tested in between two tokens

• with the help of assertions, NPs with agreement checks can be encapsulated in a macro

  DEFINE MACRO np_agr(0) 
    a:[pos="ART"] 
    b:[pos="ADJA"]*
    c:[pos="NN"]
    [: ambiguity(/unify[agr, a,b,c]) >= 1 :]
  ;
  

  (in this simple case, the constraint could also have been added to the last pattern)

• when the this label (_) is used in an assertion, it refers to the corpus position of the following token; the same holds for direct references to attributes

• in this way, assertions can be used as look-ahead constraints, e.g. to match maximal sequences of tokens without activating longest match strategy

  > [pos = "NNS?"]{2,} [:pos != "NNS?":];

• assertions also allow the independent combination of multiple constraints that are applied to a single token; for instance, the region(5) macro from Section 6.5 could also have been defined as

  MACRO region($0=Att $1=Key1 $2=Val1 $3=Key2 $4=Val2)
    <$0> [: _.$0_$1="$2" :] [: _.$0_$3="$4" :] []* </$0>
  ;
  

• like the matchall pattern [], the matchall assertion [::] is always satisfied; since it does not “consume” a token either, it is a no-op (an operation that does nothing) that can freely be inserted at any point in a query expression; in this way, a label or target marker can be added to positions which are otherwise not accessible, e.g. an XML tag or the start/end position of a disjunction

  > ... @[::] /region[np] ... ;
> ... a:[::] ( ... | ... | ... ) b:[::] ...;

  starting a query with a matchall assertion is extremely inefficient: use the match anchor or the implicit match label instead