representations may be hierarchical, the real-time comprehension and produc-
tion of language do not resort to such abstract hierarchical structure as has
been explicated in the literature of generative grammar, because sequential
information and simple extra-linguistic heuristics are suffi cient enough for
language use. Under Frank et al’s view, which is surprisingly still prevalent
in computational cognitive science (Be rwick et al. 2011), the computational-
representational theories of mind developed by theoretical linguists cannot
be linked to language processing theories. This implies that the generative
system responsible for linguistic knowledge does not underlie language com-
prehension and production, that is, hierarchical structure is not accessed in
language processing. Similarly, language acquisition is characterized as learning
word sequences or clusters of words (“chunks” or “constructions”) on the
basis of a surface-based probabilistic tendency. This raises a renewed research
interest in the relation between hierarchical structure in linguistic knowledge
and its signifi cance in real-time language use (see Le wis and Philips 2015 for
this problem). In addition, they claim that since language comprehension
and production primarily rely on sequential information, hierarchical structure
is problematic for evolution. The next section discusses the phenomenon
called “agreement attraction,” which Fr ank et al. (2012) cite as indicating
most clearly that sequential information plays a crucial role in real-time lan-
guage use.
2 Agreement attraction
Agreement attraction is an error in subject-verb agreement as in (1), where the
copular verb were erroneously agrees with the linearly closer plural noun balls
rather than the singular head noun of the subject noun phrase coat:
(1) ∗The coat with the ripped cuffs by the orange balls were... (Fr ank et al.
2012: 3)
The phenomenon of agreement attraction is commonly observed in production
(Bo ck and Miller 1991) and often yields an illusion of grammaticality in com-
prehension (Pe arlmutter, Garnsey and Bock 1999). Agreement attraction pro-
vides us with an important source of information about how linguistic knowledge
interacts with language processing. In other words, agreement attraction enables
us to assess how grammatical information, including hierarchical structure, is
encoded in the process of building grammatical dependencies in real-time lan-
guage use.
Unfortunately, Fr ank et al. (2012) have not considered major fi ndings from
previous research on agreement attraction, which speak against their claim that
human sentence processing is insensitive to hierarchical structure and depends
on linear or sequential information. There are at least fi ve fi ndings in agreement
attraction that support hierarchical structure. First, if language processing
218 Noriaki Yusa