WEBVTT

00:00:06.677 --> 00:00:11.306
How is it that so many 
intergalactic species in movies and TV

00:00:11.306 --> 00:00:14.483
just happen to speak perfect English?

00:00:14.483 --> 00:00:17.886
The short answer is that no one
wants to watch a starship crew

00:00:17.886 --> 00:00:21.774
spend years compiling an alien dictionary.

00:00:21.774 --> 00:00:23.392
But to keep things consistent,

00:00:23.392 --> 00:00:26.789
the creators of Star Trek
and other science-fiction worlds

00:00:26.789 --> 00:00:30.514
have introduced the concept
of a universal translator,

00:00:30.514 --> 00:00:35.012
a portable device that can instantly
translate between any languages.

00:00:35.012 --> 00:00:38.539
So is a universal translator 
possible in real life?

00:00:38.539 --> 00:00:42.137
We already have many programs
that claim to do just that,

00:00:42.137 --> 00:00:45.954
taking a word, sentence, 
or entire book in one language

00:00:45.954 --> 00:00:49.004
and translating it into almost any other,

00:00:49.004 --> 00:00:52.337
whether it's modern English
or Ancient Sanskrit.

00:00:52.337 --> 00:00:55.913
And if translation were just a matter
of looking up words in a dictionary,

00:00:55.913 --> 00:00:59.825
these programs would run circles
around humans.

00:00:59.825 --> 00:01:03.299
The reality, however, 
is a bit more complicated.

00:01:03.299 --> 00:01:07.349
A rule-based translation program
uses a lexical database,

00:01:07.349 --> 00:01:10.302
which includes all the words 
you'd find in a dictionary

00:01:10.302 --> 00:01:13.283
and all grammatical forms they can take,

00:01:13.283 --> 00:01:18.925
and set of rules to recognize the basic
linguistic elements in the input language.

00:01:18.925 --> 00:01:22.396
For a seemingly simple sentence like,
"The children eat the muffins,"

00:01:22.396 --> 00:01:27.050
the program first parses its syntax,
or grammatical structure,

00:01:27.050 --> 00:01:29.587
by identifying the children 
as the subject,

00:01:29.587 --> 00:01:32.317
and the rest of the sentence 
as the predicate

00:01:32.317 --> 00:01:34.368
consisting of a verb "eat,"

00:01:34.368 --> 00:01:37.422
and a direct object "the muffins."

00:01:37.422 --> 00:01:40.249
It then needs to recognize
English morphology,

00:01:40.249 --> 00:01:44.681
or how the language can be broken down
into its smallest meaningful units,

00:01:44.681 --> 00:01:46.124
such as the word muffin

00:01:46.124 --> 00:01:49.755
and the suffix "s," 
used to indicate plural.

00:01:49.755 --> 00:01:52.449
Finally, it needs to understand 
the semantics,

00:01:52.449 --> 00:01:56.178
what the different parts of the sentence
actually mean.

00:01:56.178 --> 00:01:58.074
To translate this sentence properly,

00:01:58.074 --> 00:02:01.982
the program would refer to a different set
of vocabulary and rules

00:02:01.982 --> 00:02:05.166
for each element of the target language.

00:02:05.166 --> 00:02:07.020
But this is where it gets tricky.

00:02:07.020 --> 00:02:11.820
The syntax of some languages
allows words to be arranged in any order,

00:02:11.820 --> 00:02:16.954
while in others, doing so could make
the muffin eat the child.

00:02:16.954 --> 00:02:19.647
Morphology can also pose a problem.

00:02:19.647 --> 00:02:23.243
Slovene distinguishes between
two children and three or more

00:02:23.243 --> 00:02:27.097
using a dual suffix absent 
in many other languages,

00:02:27.097 --> 00:02:30.532
while Russian's lack of definite articles
might leave you wondering

00:02:30.532 --> 00:02:33.575
whether the children are eating 
some particular muffins,

00:02:33.575 --> 00:02:36.719
or just eat muffins in general.

00:02:36.719 --> 00:02:39.708
Finally, even when the semantics
are technically correct,

00:02:39.708 --> 00:02:42.757
the program might miss their finer points,

00:02:42.757 --> 00:02:45.809
such as whether the children 
"mangiano" the muffins,

00:02:45.809 --> 00:02:47.794
or "divorano" them.

00:02:47.794 --> 00:02:51.558
Another method is 
statistical machine translation,

00:02:51.558 --> 00:02:55.762
which analyzes a database 
of books, articles, and documents

00:02:55.762 --> 00:02:59.488
that have already 
been translated by humans.

00:02:59.488 --> 00:03:02.959
By finding matches between source
and translated text

00:03:02.959 --> 00:03:05.393
that are unlikely to occur by chance,

00:03:05.393 --> 00:03:09.345
the program can identify corresponding
phrases and patterns,

00:03:09.345 --> 00:03:12.429
and use them for future translations.

00:03:12.429 --> 00:03:14.969
However, the quality 
of this type of translation

00:03:14.969 --> 00:03:17.690
depends on the size 
of the initial database

00:03:17.690 --> 00:03:21.357
and the availability of samples 
for certain languages

00:03:21.357 --> 00:03:23.383
or styles of writing.

00:03:23.383 --> 00:03:27.140
The difficulty that computers have
with the exceptions, irregularities

00:03:27.140 --> 00:03:30.994
and shades of meaning
that seem to come instinctively to humans

00:03:30.994 --> 00:03:35.045
has led some researchers to believe
that our understanding of language

00:03:35.045 --> 00:03:39.251
is a unique product 
of our biological brain structure.

00:03:39.251 --> 00:03:43.101
In fact, one of the most famous
fictional universal translators,

00:03:43.101 --> 00:03:46.439
the Babel fish from 
"The Hitchhiker's Guide to the Galaxy",

00:03:46.439 --> 00:03:49.726
is not a machine at all
but a small creature

00:03:49.726 --> 00:03:54.210
that translates the brain waves
and nerve signals of sentient species

00:03:54.210 --> 00:03:57.005
through a form of telepathy.

00:03:57.005 --> 00:03:59.726
For now, learning a language
the old fashioned way

00:03:59.726 --> 00:04:05.106
will still give you better results than
any currently available computer program.

00:04:05.106 --> 00:04:06.749
But this is no easy task,

00:04:06.749 --> 00:04:09.014
and the sheer number 
of languages in the world,

00:04:09.014 --> 00:04:12.989
as well as the increasing interaction
between the people who speak them,

00:04:12.989 --> 00:04:18.004
will only continue to spur greater
advances in automatic translation.

00:04:18.004 --> 00:04:21.409
Perhaps by the time we encounter
intergalactic life forms,

00:04:21.409 --> 00:04:24.660
we'll be able to communicate with them
through a tiny gizmo,

00:04:24.660 --> 00:04:29.026
or we might have to start compiling
that dictionary, after all.