When
a translator works upon a translation, a complex series
of operations belies the ostensibly simple process that
is taking place. It has long been the goal of computational
linguists to achieve fully automatic high quality machine
translation, but to date that goal is some way off, nevertheless,
with the means at our disposal today, is machine translation
a viable choice?
Before proceeding, it would be well to analyse
the translation process to gain some appreciation of the
problems to be overcome by translation software...the process
is thus:
The translator must arrive at a full understanding
of the source text, the means of accomplishing this through
a thorough understanding of the languages’:
- Grammar - The set of rules determining
the use of a language.
- Syntax - The rules that determine the
structure of sentences and, therefore, how grammatical
such sentences are.
- Idioms - Terms or phrases whose meanings
cannot be derived from their literal translation, but
only by knowledge of their local usage e.g. ‘raining cats
and dogs’.
- Semantics - The study of culturally specific
meanings or concepts within a language.
- Cultural Framework - The basis for understanding
the ‘patterns of activity’ of a discrete group of people.
Once this is accomplished, using an understanding
of the same features of the target language, a conversion
is made resulting in the finished translation.
So what technical methodology can machines
employ to duplicate the process shown above?
The first and most basic is ‘dictionary-based’
translation. This is where translation is carried out in
an analogous manner to a dictionary - word for word; however,
this can often give a somewhat meaningless output with no
connective theme. Nonetheless, it can form the ‘base’ upon
which other methodology can be added.
The next method that can be used is that
of statistical machine translation. This is a somewhat complex
translation methodology, but in its simplest terms it takes
two corpora of text and compares and matches each corpus
against the other, thus arriving at a sort of equivalence
and thus enabling it to produce a sort of statistically
modelled output.
The last method is that of Interlingual
machine translation. Interlingual machine translation takes
a source language and transforms it into a language independent
medium - a lingua (rather like the conversion of the alphabet
into binary 1’s and 0’s). The target language will then
be generated from this Interlingua. This is an important
methodology because the concept of converting a language
into an abstract representation before re-formation is a
core principle used in artificial intelligence.
So has a combination of the aforementioned
technological approaches led to machine translation being
a viable choice? Well, at the moment no...or should I say
not as a stand alone method. The trouble with language is
that although it follows general rules, actual usage in
real life is open to very wide variance, there are always
linguistic exceptions and everyone uses widely different
modes of expression, making it difficult for any mechanical
method to account for this.
This being said, machine translation can
often produce an acceptable output and can therefore be
used as a timesaving device to quickly provide a loose translation
of the mass of a corpus before it being passed to a translator
for refining. Many translations agencies, do in fact, use
such a procedure and if you were to pass a text to any translation
service in the UK you would probably find that
to be the case. This can ultimately only be for the good...as
machine translation improves, the time required to carry
out a translation will decrease and so, subsequently, will
the cost.
Lingo24 Translation
Service
