Crossing dependencies are known to be difficult to process in a
head-initial language such as English (Levy et al., 2012). Since dependency
length minimization has been claimed to be a crosslinguistic processing
constraint (Futrell et al., 2015; Liu et al., 2017), the length of crossing
dependencies is also expected to be minimized. This implies that the length
of the intervening
phrases (IL) that leads to crossing should be short. Contrary to the
dependency minimization hypothesis, considerable research (e.g., Levy and
Keller, 2013) has shown that in head-final languages, the head-dependent
length can be long (and lead to processing facilitation). This points to
processing adaptability based on certain typological properties (cf.
Vasishth et al., 2011, Levy
et al., 2013). We carry out a corpus study to investigate the influence of
head directionality on crossing dependencies. We choose 12 languages
(Head-final:6, Head-initial:6) annotated with Universal Dependencies (Nivre
et al., 2016). We classify crossing dependencies as (a) Head-F: head
follows the dependent, and (b) Head-P: head precedes the dependent. The
results indicate that head directionality of a language has a role to play
in determining both the linear and hierarchical distance in a costly
structure such as crossing dependency. Crossing dependencies can have
higher relative depth in head-final languages, and allow longer IL in a
configuration that aligns with the default head directionality of its
language. Under the assumption that corpus data reflects processing
constraints, our work points to processing adaptability with respect to
crossing dependencies based on certain typological properties of a language
(cf. Vasishth et al., 2011; Levy et al., 2013).