56
”restaurants including
t and”
”restaurants for example
t and”
”restaurants like
t and”
”restaurants such as
t and”
Table 4.3. Hyponym patterns for instance-class relation.
The learned patterns with the highest scores are given in Table 4.2. We note
that the numbers for
spr and freq are low due to the limited number of queries used.
Many of the patterns are recognizable as typical for the given relation. However,
the vast majority of the patterns has a value for
pspr = 1. Hence, the use of such a
pattern in combination with either
KFC or Burger King led only to a single country
name. These patterns may therefore be too specific.
Recognizing Instances. The country names are recognized using the collected
list of countries. We assume the country names to be unambiguous. We extract the
longest terms
t consisting of at most 4 capitalized words at the placeholder, directly
preceding or following the queried expression.
As this extraction rule is likely to not only cover restaurants but a wide range of
terms, we use a check function to filter out erroneously extracted terms. Hereto, we
use the set
H of hyponym patterns in table 4.3 in the following acceptance function
accept(
t) =
∑
Dostları ilə paylaş: