Seminar Work 15 Feb 2006 Presentation 4 J. Ramanand ramanandit.iitb.ac.in

1
Seminar Work 15 Feb 2006Presentation 4 J.
Ramanand (ramanand_at_it.iitb.ac.in)
2
HowNet Recap

• Chinese Research Project
• Semantic Network combining lexical and conceptual
  entries
• Relations Inter-conceptual and Inter-attribute
• Database of entries that is machine usable

3
Conceptual basis of HowNet

• Constructive Meaning Representation A set of
  sememes that make up the foundations of HowNet
  can be used to construct concepts
• Sememe A basic unit of meaning which cannot be
  further decomposed
• HowNet used more for constructing meaning and
  concepts than differentiating concepts
• Meaning Representation achieved using sememes and
  relations between sememes to build more concepts
• Takes an ontological view of the objective
  world

4
Concept in HowNet example

• Concept to be represented Teacher
• In HowNet, concept expressed as a combination of
  the sememes for human (entity), teach (event)
  and education (entity)
• The HowNet record for teacher will have
• Its hypernym human
• Attribute(s) education
• Agent relation to teach
• Its part of speech Noun

5
HowNet Internals

• Construction
• A basic set of sememes were extracted from
  characters from the Chinese alphabet
• This takes advantage of the fact that these
  characters represent a concept by themselves
• Combining these characters further generates
  newer concepts. So HowNet records can be used to
  construct these newer concepts
• HowNet shaped as a knowledge system made up of a
  set of databases holding information about
  events, entities, lexical entities etc.

6
HowNet Concept Record

• Each Concept is represented by the following
  record structure
• NO Concept number e.g. 023249
• W_X Word or phrase e.g. doctor
• G_X Syntactic class (noun, verb etc.) - e.g. N
• E_X Usage example gloss e.g. Doctor serves in
  hospital
• DEF Concept Definition e.g. humanHostOfOcc
  upation,domainmedical,doctoragent

Note Each HowNet Concept Record contains Chinese
words, but they have been omitted in this slide
7
Definition of Concept

• This field indicates relations inter-entity and
  to attributes
• In the previous slide, doctor's DEF
• humanHostOfOccupation,domainmedical,doct
  oragent
• The indicates the following relations
• Hypernymy to human
• Human to Doctor exhibits occupation as a
  relation
• Relation to the medical domain
• '' self-reference from the verb to doctor to
  the noun doctor

8
Applications

• Given a record, a gloss can be generated. e.g.
  using the doctor record, a sentence a human
  can have an occupation of doctor in the medical
  domain who performs doctoral activities
• Uses
• Word Sense Disambiguation
• Machine Translation (especially between English
  Chinese)
• Analogical Reasoning

9
Issues and Software

• Issues
• How to identify newer concepts
• How are relations and entities identified
  maintained
• Can it be extended to other languages as it seems
  to be based on the Chinese alphabet system
• Software
• Mini HowNet available for download
• Only provides lexicon from A-D
• Dictionary and Taxonomy provided in Mini HowNet
• Concept Similarity Tool

10
References

• HowNet http//www.keenage.com/html/e_index.html
• Introduction to HowNet http//www.keenage.com/The
  ory20and20practice20of20HowNet/04.pdf
• Knowledge-based Sense Pruning using the HowNet
  an Alternative to Word Sense Disambiguation
  http//www.keenage.com/papers/sensepruning.doc
• Analogical Reasoning with a Synergy of WordNet
  and HowNet http//afflatus.ucd.ie/papers/GWC2006a
  .pdf
• Mini HowNet Download http//www.keenage.com/downl
  oad/e_download_tk.html