Difficulties in MT

Questions are often raised regarding why there has been only limited success in Machine Translation efforts thus far?


Machine Translation efforts are aiming at reducing the language barrier to ZERO! This is almost like trying to build a machine that can walk like a human! This is rather ambitious if not impossible
.

The following are examples of some key problems faced in machine translation:

Ambiguity
E.g.:1: राम फल खाता है l (rāma phala khātā hai.)

Challenge: Machine has difficulty in identifying कर्ता and कर्म.
Solution: Human beings use world knowledge to resolve this.

E.g.:2: Time flies like an arrow.


Possible parses: (Note: Parsing is a term used in Linguistics and implies the identification of the grammatical role of each word in the sentence. In the following sentences: N refers to Noun; V to Verb; Prep is used for preposition; Det for determiner.)

a) Time     flies     like     an     arrow
        N            V      Prep   Det     N

b) Time     flies     like     an     arrow
       
N            N      Prep   Det     N

c) Time     flies     like     an     arrow
       
V            N      Prep   Det     N

(Interpreted as: flies are like an arrow)

d) Time     flies     like     an     arrow
        N            V      Prep   Det     N
(Emphasis on: manner of timing)

Challenge: Different interpretations are possible.
Solution:
World knowledge and context.

E.g.:3: He saw a man on the hill with a telescope.

Possible readings:
                    saw with a telescope
                    man with a telescope
                    hill with a telescope

Challenge: Different interpretations are possible.
Solution:
World knowledge and context.

E.g.:4: Patient had a stiff neck and fever

Patient had stiff neck and fever
Patient had stiff neck and fever

Challenge:
Different interpretations are possible.
Solution:
General awareness.

E.g.:5: 'Table the resolution'

U.S.A.: To postpone
U.K.: To present before the committee

Challenge: Different interpretations in different countries.
Solution: Knowledge of British vs. American usage.

Dereferencing

E.g.: 1:
The farmer's wife sold the cow because she needed money.
The farmer's wife sold the
cow because she was not giving milk.

We know that 'she' in the above sentences can refer to wither the 'wife' or the 'cow' giving different interpretations.

E.g.: 2:
The mother with babies under four ...
The mother with babies under forty ...

Clue: Common Sense.

E.g.: 3:
William James cited Mozart's discussion on his composition.
Marvin Minsky cited McCorduck's discussion on his research.

Clue: Knowledge of historical background is important.

Incomplete/ungrammatical sentences

Doctors' diagnosis notes:
                        • stiff neck and fever.
                              • brain scan negative.

As one would note from the examples above – interpretation depends on common sense and knowledge of the world which the machine would not have. Thus, the task of Machine translation becomes challenging. Interestingly, even for human interpretation, if two languages are similar, then the challenge for the user to interpret is not very great.

However, if the incompatibility increases (as in the case of English and Hindi), the task of interpretation (even for a human being) becomes more complex. Here we are attempting to get the machine to do the same task – thus, the magnitude of the challenge increases manifold.

 

form input

Root/Word