MT and Overcoming Language Barrier

Brief History of Machine Translation

Anusaaraka: a tool to reduce language barrier

Anusaaraka: A better approach to Machine Translation

Issues in Machine Translation (MT)

MT systems are necessarily fragile   WHY ?

• Languages code information only partially, and
• languages are incommensurate among themselves

The incommensurability may surface at different levels such as

Grammar level

• Differences in coding scheme:
  • English does not have equivalent morphems for
    • Hindi 'ko{2}'
      • codes information in SUBJ position!
    • Hindi 'kyA{yes-no}'
      • Uses auxiliary inversion mechanism to code the information!
  • English typically uses same verb form for transitive as well as intransitive meanings.
    • It can't drop the object

• Labeling and Packaging of concepts:
  • English: chairs
  • Hindi_gloss: kUrsI+{ba.}/adhyakshatA_nibhAnA+tA_hE{sg.}
     
  • English : As
  • Hindi_gloss : sAmya{tulanA}{rUpa/vyavahAra/desha/kAla[>kAraNa/ke~bAvajUda]}
   

Solution: ANUSAARAKA

Provides a fall-back mechanism which includes

• graceful failure, and
• a safety net

Anusaaraka Guide lines for developing MT systems:

• Make complete information available to the user  but, do not clutter the scene.

                    ---Hiding Mechanism/User Interface
                    ----- Reversibility

• Separate the resources that can be made, in principle, reliable from those that are, inherently unreliable. Mention the degree of reliability explicitly.

                    ----- Change the Ordering of Operations

• provide alternative means to get the information

                      ----Online help
                  ----Human readable algorithms

Practical Implications:

• Reverse the order of operations, and
• Develop a Graphical User Interface (GUI) to present the right amount of information at right time