26
Examples: III
Errors:
non-genuine à genuine
| (0.5) |
|
| Sample 1: row
consistency. In fact one could argue that this is in fact a genuine table,
with implicit row headings of: Title, Name, Affiliation, phone number and
email address. This example demonstrate one inherent difficulty in this work,
namely whether or not a table is a genuine table is some times ambiguous even
to humans. |
|
| Sample 2:
reasonable consistency along the columns. This is a case where current
features are not adequate. In order to correctly classify a table like this,
one needs to have more advanced language analysis to discover the fact that
there is no logical relation/semantic consistency among the cells – and that,
of course, is a very difficult task. |