6.1 The matching strategy

set MatchingStrategy (shortest | standard | longest);
In shortest mode, ?, * and + operators match the smallest number of tokens possible (refers to regular expressions at token level)
$\Rightarrow$ finds shortest sequence matching query,
$\Rightarrow$ optional elements at the start or end of the query will never be included
In longest mode, ?, * and + operators match as many tokens as possible;
In the default standard mode, CQP uses an “early match” strategy: optional elements at the start of the query are included, while those at the end are not.
The somewhat inconsistent matching strategy of earlier CQP versions is still available in the traditional mode, and can sometimes be useful (e.g. to extract cooccurrences between multiple adjectives in a noun phrase and the head noun). The example below only gives the intended frequency counts in traditional mode.
> [pos="JJ"]+ [pos="NNS?"];
> group Last matchend lemma by match lemma;
Figure 4 shows examples of all four matching strategies for the CQP query
> [pos="DT"]? [pos="JJ.*"]* [pos="NNS?"]
( [pos="IN|TO"] [pos="DT"]? [pos="JJ.*"]* [pos="NNS?"] )*

Figure 4: CQP matching strategies.
$\begin{figure}\begin{center}\footnotesize \begin{tabular}{clllllllll} \multico... ...rhd$\ & & & & & & & & & room \\ \\ \end{tabular} \end{center} \end{figure}$
Standard queries in CQP are executed by first looking up potential starting points with the help of available index data structures, and then scanning for a match of the query from each candidate starting point (by non-deterministic simulation of a finite-state automaton (FSA), as is commonly done for regular languages). If the query contains a top-level disjunction of multiple alternatives and/or there are optional elements in query initial position (as is the case in Fig. 4), the simulation is carried out in multiple passes, one for each query element that could be a starting point (DET?, ADJ* and NN in Fig. 4).
During a FSA simulation run for a given start position, all matching strategies except longest will stop as soon as a successful match has been found, while longest continues until there are no active states left or the search boundary has been reached. This makes longest matching potentially expensive, especially for unspecific queries involving []*.
The other three matching strategies only differ in how nested matches from multiple passes are integrated into the final query result:
- longest discards any match that is fully contained in a longer match;
- shortest discards any match that fully contains a shorter match;
- standard picks the shortest one of multiple matches with the same start point, and then the longest one of multiple matches with the same end point;¹⁴
- traditional simply keeps all nested matches from the individual passes.
All matching strategies remove exact duplicates from different passes, even if they have different target or keyword anchors. However, overlapping matches (that do not nest) are never discarded.
new in CQP v3.4.12: The matching strategy can be set temporarily with an embedded modifier at the start of a CQP query, e.g.
> (?longest) [pos = "NP.*"]+;
Currently, only these four modifiers are supported: (?shortest), (?standard), (?longest) and (?traditional). Embedded modifiers are particularly useful for Web interfaces that do not give users direct control over the matching strategy. Since they are part of the CQP query syntax, no modifications to existing Web interfaces are required.
The matching strategy only applies to standard queries, not to TAB queries (Sec. 8.5) or MU queries (Sec. 8.4).