In this article, I want to present the minor subject I had at university: computational linguistics (CL). Many UniLangers are interested in CL, since it combines languages with information technology and mathematical methods. Because of this unusual combination, I will examine CL from three sides: for language freaks, linguists and computer programmers. A general approach written by a CL professor can be found at http://www.coli.uni-sb.de/ ~hansu/ what_is_cl.html

A brochure informing about the study programs at my university says:

Computational Linguistics examines human language. It uses formal models that can be implemented on the computer. In this way, it gains knowledge about the phonetic, syntactic and semantic structure of languages and about the way humans understand, produce and learn language.

Computational linguistics is used to create computer programs for natural language processing, natural language generation and translation.

• A natural language processing (NLP) system tries to understand spoken or written language. There are car navigation systems and word processors that react on spoken input. But the technology is still far from what you can see in Star Trek.
• The other direction is natural language generation (NLG), which deals with producing an acceptable text (or even speech) in a natural language. NLP and NLG are often combined in dialog systems, but they are two distinct modules that work in quite different ways.
• Machine translation is more than looking up words in a dictionary. It is very difficult to translate a text you can’t understand. Therefore the best machine translation systems are those that use both NLP and NLG, but computer scientists developed more primitive methods that are quite good in many cases.
• Information retrieval tries to find relevant information in texts. With the incredible size of the World Wide Web today, it is impossible for humans to read every single website when you’re looking for information. CL can help to advance search engines by making them ‘understand’ the texts on the websites.

Forget about learning foreign languages

If you are a language freak, you want to learn as many languages as possible, because you like communicating with people from all over the world. You don’t need that in CL. Most of the examples you will study will be in English and/or the language of instruction. Sometimes you will see sentences in exotic languages, but you needn’t understand them.

Computational linguists use exotic languages to show properties unusual in well-known languages like English. In a course on grammar theory, I once had an example from Warlpiri, a Pama-Nyungan language spoken in northern Australia:

You will always get those literal translations annotated with grammatical functions; no need to actually learn Warlpiri…

Fall in love with grammar

What do you love in a language? Its sound, the beauty of the writing system, proverbs, metaphors, poetry? That is mostly irrelevant to CL. You will deal with language on the following levels:

• Lexical semantics deals with word meanings, which can become difficult the deeper you get into the topic.
• Morphology deals with the way words change when they are inflected. There is little variation in English (speak, speaks, spoke, spoken), none in Chinese and lots of different forms in French (parler, parlez, parlerions, parlées, parlant, parlèrent, …).
• Syntax deals with the way words combine to form phrases and sentences. It’s about finding the subject of the sentence, the predicate, the temporal adverbial, the relative clause and so on.
• Semantics deals with the meaning of sentences. Once the computer knows the grammatical structure of an utterance like "John loves his cousin" and the meaning of the individual words, it can infer who is loved by John.
• Pragmatics deals with the meaning of a sentence in a context. If someone says ‘I am cold’, is s/he informing the listener about unpleasant room temperature, fearing s/he might get ill or requesting to close a window?

If you want to know how CL deals with those levels, please read the following section from the linguistic point of view.

A practical approach with computers

Computational linguistics combines several linguistic disciplines in a practical way. CL is no blabla and no philosophical discussions. Its aim is to make computers process or generate language, so you should know a little about computer programming.

Computers are stupid. They can only perform a task if it is sufficiently explained (either by you or the programmer). If you ask people how to say ‘apple’ in Finnish, they will either use their knowledge or a dictionary to tell you the answer. A computer would have to execute the following program:

if English-Finnish dictionary database is available then
    return dictionary_entry(database, English-Finnish, "apple")
else
    check which dictionaries are on the bookshelf
    if book_found(bookshelf, English-Finnish dictionary) then
        search_entry(that book, "apple")
        if entry found then
            return read_entry(that book, found entry)
        else
            return error_message("Word not found in dictionary")
        end if
    else
        return error_message("Dictionary not found on bookshelf")
    end if
end if

And this is only a simplified sketch; real computer programs are much more complicated!

The language of mathematics

Computational linguists use strict formalisms like mathematicians and computer scientists. The example below shows a formalization of the sentence ‘I want to put a book on the table’.

For computational linguistics, it is important to express everything related to a language (phonetics, phonology, morphology, syntax, semantics, pragmatics) in such a formal way. No philosophical reasoning about metaphors, just plain logical thinking. Why? Because CL is about teaching language to computers. And everything computers understand is formal logic.

As a programmer and computer freak, you don’t have much of a problem with the points mentioned in the two previous chapters. But wait a moment, CL still stands for computational linguistics…

CL is about human languages

There are huge differences between programming languages and real human languages. How would you react if someone told you for(i=1; i<life.length; i++) if(location==NYC) return 1; instead of plain English: "Have you ever been to New York"?

Of course you use programming languages, or at least stuff like PROLOG or LISP. But that’s not what you’ll implement. Parsing natural languages is much more difficult:

Now your parser should know that the order Aux-NP-V indicates a question in English (with that noun phrase as its subject), and that the present tense of the auxiliary ‘to have’ together with the past participle of a verb puts that verb into present perfect.

CL is a branch of linguistics

A program always has a purpose, even if it’s fun writing it. When you write natural language processing or generation programs, you need to know how natural languages work.

Do you know what an adjunct is? Ever heard of equi and raising phenomena? Are you able to recognize the finite verb in a long sentence? You should be. In my CL studies, I was confronted with tons of words like that, and all that was really explained by the teachers was unification, which is not much more than a simple intersection of two sets.

The main problem of CL is the complexity of natural language, especially ambiguities. A word can have several meanings (the bank of a river vs. the Bank of England), and so can a complete sentence ("John follows a gangster in a sports car" – Who drives the sports car?). But even if there is no ambiguity, there are difficulties. The sentence ‘I am cold’ can be an expression about temperature, an indication of a disease or a request to close the window.

Are you interested in expressing natural language in a formal way? Interested in programming software that can process and generate natural language? Why don’t you start computational linguistics? :-)

CL methods can be used in various applications such as car navigation systems, spell checkers for word processors or better search engines in the WWW.

CL is a rather new science that still has a lot to discover. Although computers are getting faster and faster, we’re far from universal translators devices like in Star Trek. Though we can do better than Altavista’s Babelfish, as the example of Verbmobil shows:

To learn more, check out The Association for Computational Linguists and The Linguist List or ask UniLangers who have experience with CL. :-)

By the way, I was told the best universities to study computational linguistics are the following three: