Page 70 - My FlipBook
P. 70
工
智
慧
計
畫
Arti cial Intelligence Projects Also in 2019, we developed and released E-HowNet 2.0 – a representation (see Figure 3). It has the following
new entity-relation commonsense representation model improvements relative to its previous version:
with semantic composition capability for knowledge
a. Reorganization of the hierarchical structure of primitive c. Development of a new automatic ontology reconstruction
and basic concepts: We have extended a large set system: In cases where lexical sense expressions or
of basic concepts, generating a deeper hierarchical nodes of conceptual hierarchy are revised, the ontology
structure and more precise semantic branching. This reconstruction system may re-attach each lexical entry
work has also resulted in lexical senses expressed from to appropriate ontological nodes, resulting in a new
basic concepts becoming more precise and readable. ontology.
We have reformed the ontology structure into two
parts. The first part represents a hierarchy for entities d. Improvements to sense de nitions via basic concepts:
and the second part is a hierarchy for relations, i.e., Many word sense definitions can be revised and
semantic roles. Furthermore, Attribute and Value types rendered more precise and readable by using basic
have also been reorganized accordingly. concepts in their sense expressions. More semantic links
are established due to the shared semantic features
b. Rich lexical information: In addition to sense de nition, as well as explicit relation links, such as antonyms,
each entry of lexical sense may include operational attribute values, and entailment.
expressions as well as semantic functions, facilitating
future semantic composition processes. Event frames,
i.e., argument structures, for event type primitives are
also provided.
Figure 3 : Hierarchical structure including primitives and basic concepts in E-HowNet 2.0, with
the example of the word " 蜻蜓 (dragon y)".
68
智
慧
計
畫
Arti cial Intelligence Projects Also in 2019, we developed and released E-HowNet 2.0 – a representation (see Figure 3). It has the following
new entity-relation commonsense representation model improvements relative to its previous version:
with semantic composition capability for knowledge
a. Reorganization of the hierarchical structure of primitive c. Development of a new automatic ontology reconstruction
and basic concepts: We have extended a large set system: In cases where lexical sense expressions or
of basic concepts, generating a deeper hierarchical nodes of conceptual hierarchy are revised, the ontology
structure and more precise semantic branching. This reconstruction system may re-attach each lexical entry
work has also resulted in lexical senses expressed from to appropriate ontological nodes, resulting in a new
basic concepts becoming more precise and readable. ontology.
We have reformed the ontology structure into two
parts. The first part represents a hierarchy for entities d. Improvements to sense de nitions via basic concepts:
and the second part is a hierarchy for relations, i.e., Many word sense definitions can be revised and
semantic roles. Furthermore, Attribute and Value types rendered more precise and readable by using basic
have also been reorganized accordingly. concepts in their sense expressions. More semantic links
are established due to the shared semantic features
b. Rich lexical information: In addition to sense de nition, as well as explicit relation links, such as antonyms,
each entry of lexical sense may include operational attribute values, and entailment.
expressions as well as semantic functions, facilitating
future semantic composition processes. Event frames,
i.e., argument structures, for event type primitives are
also provided.
Figure 3 : Hierarchical structure including primitives and basic concepts in E-HowNet 2.0, with
the example of the word " 蜻蜓 (dragon y)".
68
