Conceptual Aspects: How Landmarks Can Be Described in Data Models - Landmarks: GIScience for Intelligent Services - page 109

larger number of database instances, and this ambiguity needs to be resolved.
A reference to an instance by name may be ambiguous as well (how many
instances of “State Library” are out there?), but also many names are not
gazetteered, or stored in spatial databases. Addressing ambiguity requires two
skills that are both not well developed in spatial databases, context-awareness and
qualitative spatial reasoning. Context-awareness involves reasoning of the kind
that a person in Melbourne, mentioning “State Library”, most likely refers to
the State Library of Victoria, Australia. Qualitative spatial reasoning, in contrast,
involves mechanisms to interpret the qualitative spatial relationships used in the
human-generated message to resolve ambiguity, such as in “The café opposite
the State Library”.
However, where needed for disambiguation, people also refer to properties
of objects. Landmark references such as “the yellow building” or “the white-
steepled church” pose additional challenges to spatial databases that typically do
not capture a broad range of perceptual properties with the stored features.
￿For generating human-like messages the system is still completely handicapped
as it does not have a notion of landmarkness. It may, however, maintain exten-
sional lists of points of interest or similar surrogates for landmarks, in which case
it is capable of generating messages that look like spatially intelligent messages.
However, the elements may miss the discriminatory power, the identifiability, or
the relevance in a particular context we would now expect from an intelligent
system.
Thus it is essential to introduce a notion of landmarkness for simulating human
communication behavior. This notion will be made context-aware. Yet the issue
of capturing context and modelling context awareness is not well understood and
needs further research [ 9 ] . Furthermore, abilities of qualitative spatial reasoning are
essential for an intelligent system. But this is another active area of research [ 4 , 7 ,
47 , 62 ] .
In summary, we postulate that the notion of landmarkness should cover types as
well as individual entities in spatial databases, and should characterize a collective
expectation for embodied or mediated experiences of the environment. This addition
to spatial databases must allow an internal reasoning and external communica-
tion behavior simulating some of the properties of mental spatial representations
and abilities in working memory, such as choosing relevant reference objects,
constructing hierarchical localization (e.g., [ 57 ] ), switching between survey and
route perspective, and appropriate qualification of quantified delimiters. A machine
with these capabilities will not only be able to interpret human-generated spatial
expressions but also will have the capacity to produce human-like spatial expres-
sions. Further conversational capabilities in case of insufficient information are
advantageous [ 48 ] and appear in regard to intelligent behavior even essential.