Machine Translation Fact Sheet

1. Why is machine translation not perfect?
2. What are the advantages of using MT?
3. What is machine translation used for and in which areas?
4. How does machine translation work?
5. What has to be considered before translation?
6. What kinds of translation programs are available?
7. Buying tips / what kind of translation program would be suitable for me?
8. What is the difference between free online translators and PC versions?

1. Why is machine translation not perfect?

Whereas language is just a matter of course for people, the concept of language poses tremendous problems for software. The reason for this is the virtually infinite diversity of a language. A person can express facts and circumstances in various ways and it is only very rarely the case that the language is as clear and concise as a digital system that is based on ones and zeros. According to language researchers, every person is born with an inherent knowledge of grammar. Small children acquire a language by unconsciously categorising everything they hear and by structuring everything they say themselves with the help of inherent language rules. The words and rules and how they can be combined with one another sometimes differ greatly from language to language. Nevertheless all languages have common structures, so-called deep structures. Simple translation programs are based on surface structure and they translate one word after the next. This means that unintelligible nonsense is often produced, like “equal goes it loose” for “gleich geht es los”. Usable results are only produced by software that can analyse the surface structure of a sentence for each and every sentence in the source language, and then categorises them to identify their deep structures. It then transfers them into corresponding structures in the target language. Practical considerations mean that the analysis is limited to the sentence level. An analysis of larger units, ideally over and beyond sentences, is desirable but is currently only being implemented by a few translation programs. Many factors contribute to the, often incorrect, output of machine translation. Human language is complex, imprecise, idiosyncratic and unclear. Words with more than one meaning, sentences with grammatical structures having more than one meaning, the recognition of pronouns and other grammatical problems cause translation software to stumble. How can a machine know how the German word “Bank” (can mean a bank as in a financial institution but can also mean a bench) should be translated in the following sentence: “Er schrieb einen Brief an die Bank”. People know that you usually don’t write a letter to a bench so the correct translation would be “bank” in English and not “bench”. People possess practical world knowledge and machines do not have such knowledge. In good translation programs, there are settings to define specialist areas (e.g. banking) but this fine-tuning can not be done for all sentences.

Examples of several meanings:
Die Firma stellte einen neuen Mitarbeiter ein.
Der Mechaniker stellte den Vergaser ein.
Die Gäste stellten das Rauchen ein.
Ich stelle mich auf Schwierigkeiten ein.

Depending on the context, the word “einstellen” has to be translated differently in every example:

The company hired a new employee.
The mechanic adjusted the carburetor.
The guests stopped smoking.
I adapt myself to difficulties.

2. What are the advantages of using MT?

Machine translation is neither a miracle machine, nor a cost-efficient gadget without any practical use. Although questionable results are sometimes produced, machine translation has nevertheless become an essential technology. The need for automated translation is illustrated by current trends: the rapid growth of the Internet, political internationalisation, economic globalisation and rationalisation which all are increasing the need to have documents / text translated. Europe has a common currency but not a common language. Language barriers have to fall, as a barrier-free Internet alone is not a guarantee for worldwide communication. The Internet already has over 800 million sites containing material. Previously, most of the information was in English, but other languages are quickly becoming more important. Anyone that publishes their website in English can no longer assume that the majority of users will understand it. A lot of information and possibilities remain hidden from Internet users with little or no knowledge of foreign languages. Economic results and computer topics in particular are generally produced in English. User-oriented websites that help with technical questions, news or online shops are also frequently in English or in the relevant country’s language. There is only a limited number of human translators available for this enormous market. According to market studies, the demand for translation vastly outweighs the number of translators. What’s more, human translators are expensive. Good results are produced, but the process takes quite a long time. Depending on the difficulty, a translator needs about 1 hour for one page of text. Translation systems can manage up to a hundred times this volume in the same time! Time savings and productivity increases are becoming more and more important for companies. Tests carried out by the Fraunhofer Institute have shown that Personal Translator can considerably increase the productivity of PC users. These tests have also demonstrated that you can achieve up to an average time saving of 41.4% by using Personal Translator! What can be better than having the computer do the translation work for you?

3. What is machine translation used for and in which areas?

If you have large volumes of text that often need to be translated or you work with foreign-language texts that need to be understood quickly, a translation program is the practical choice. The main application areas for machine translation are:

• International communication via emails or business correspondence is faster with machine translation.
• Existing documents can be made available for branches or business partners located abroad in their native language.
• Translation of technical texts, e.g. documentation, manuals, operating instructions etc. Machine translation is especially efficient at translating standardised texts (using translation memory).
• Translation of websites to ensure the reader understands them.
• Translation of websites to make them available in several languages on the Internet.
• Use of machine translation in online chats, real-time auctions.
• Machine translation is less suited to the literary translation of novels and poems.

4. How does machine translation work?

Along with the basic grammar rules of the source and target language and the exceptions to these rules, a variety of information must be given for every word. This includes:

• the word class and the inflection form (noun, verb…)
• the meaning depending on the context (e.g. “driver” means “Fahrer” in a general context but “Treiber” in an IT context)
• and last but not least various attributes. For nouns, you have to specify if it is an animate object, an inanimate object, a place, an object or only a concept. For verbs, you have to specify if they are transitive or intransitive or what kind of object follows it.

Software translates in the following manner: It breaks down the sentences into words for the analysis and looks these up in the dictionary. The program then identifies the grammatical elements, possible translations for every word and additional knowledge. Then it performs a syntactic analysis of the sentence. It breaks down the sentence into individual components – the clauses. It determines the function of every word in the sentence such as the subject, predicate and object.

The actual translation takes place in two steps:

• The lexical transfer takes place first. Every word is given a translation valid for the context of the sentence. The program takes into account the grammatical elements and any additional information that has been saved on each word in the dictionary. The number of entries that exist for each word affects the number of decisions made by the program.
• Then the structural transfer takes place. This ensures the correct position of the word in the sentence and makes other structural adjustments. Structural adjustments include transformations such as the conversion of a German interrogative clause into an English one (in English, interrogative clauses often begin with “do …”).

The program then produces the correct form of words and the translation is given a final structure.

The translation process is illustrated again in the following diagram:

A completely new feature of Personal Translator is that, in addition to rule-based translation technology, it contains a newly developed component that simulates the thinking process of the human brain according to the principle of (Neural Transfer) and makes it usable for translation.

5. What has to be considered before translation?

Translation is a very complex task. A translation program works best if the text to be translated is grammatically correct. Slang expressions, typos, missing punctuation, complex sentences or sentences that are too long can result in an incorrect translation.Furthermore, it is important to select from the various translation options available the one most suited to the source text. In this case, the selection of suitable subject areas is also important. If you define the unknown words before the translation, the translation quality will be improved considerably.

6. What kinds of translation programs are available?

Translations requiring laborious leafing through thick heavy books and dictionaries are a thing of the past. The heavyweights have generally been replaced with a variety of intelligent electronic reference books and translation software. Here are the following different types:

• Digital/online dictionaries: By typing the search word, the term quickly appears in the required language. Some digital or online dictionaries also contain more entries than the thick heavy books kept on your desk. You can buy a range of dictionaries from language publishers on CD-ROM, either with general vocabulary or specialist lexicon. There are many free reference materials available on the Internet. However, the quality of these is often inadequate.
• Word-for-word translation programs: These programs replace every word of the source language sentence with a word in the target language. Depending on the complexity of the target language and the degree of kinship of both languages, a word-for-word translation can deliver either completely usable results or total rubbish like “equal goes it loose” for “gleich geht es los”. “Casa Blanca” in Spanish would be translated as “house white” in English. Word-for-word translation were originally a translation aid for making content of a foreign text available on an interactive basis, for looking up terminology and/or creating terminology databases and for semi-automated (interactive) translation. In the meantime, they are making space for full-text translators.
• Sentence-for-sentence translators: These programs are known as full text translators. Once again here, the users are not completely relieved of work: to obtain good translations, a lot of texts have to be processed. Long, complex sentences need to be broken down before the translation into smaller sentences in order for the output to be comprehensible. Slang words or heavily abbreviated terms should be avoided. Also, people usually don’t get around the post-editing of a translation. The effectiveness of a translation system depends on the translation quality of the relevant program.

7. Buying tips / what kind of translation program would be suitable for me?

A good translation program should meet the following criteria:

• The translation quality is extremely important. Enquire about the possibility of getting a trial version (e.g. online, or demo CD) or demonstrations (e.g. at an exhibition) before buying the product.
• Simply add new words to the dictionary: in the translation of texts with certain terminology requirements, you should be able to add terminology (specialist terms, proper names) to the dictionary. Pay attention to how easy and intuitive it is to enter words without excluding the different linguistic possibilities. A terminology import is also a desirable feature to transfer existing bilingual word lists into the user dictionary quickly and easily.
• Is it possible to set various specialist areas, e.g. banking, production engineering, medicine, etc.?
• Individual adjustment possibilities, e.g. old/new German spelling rules, British / American English.
• Translation memory for storing translated and revised sentences for re-use.
• Microsoft® Word integration for integrating the translation software into the text processing program Microsoft® Word so that the translation can be carried out directly in word.
• Translation of Internet sites retaining the site structure.
• Dictionary function: ability to look up unknown words quickly, ideally by means of a simple click.
• Can the dictionary be expanded with specialist vocabulary?
• Integrated text recognition software: if letters and faxes that need to be translated are not available in file format, a translation software interface for text recognition (OCR) is an advantage.
• Text-to-speech function: Text-to-speech function for all texts into the languages of the translation program.
• Easy to use software.

8. What is the difference between free online translators and PC versions?

When it comes to free translation services on the Internet, an Internet browser as an entry window is used for the text. The text to be translated is entered or typed using the clipboard function. Here, text formats are not taken into consideration, i.e. bold type, underlining, various font sizes etc. disappear and the user only gets plain text format back.
– Advantages: Operation is very easy. The user does not need to be trained into the program and does not have to be bothered with linguistic details. As you do not have to buy software, you just have to pay for the Internet connection.
– Disadvantages: No extensions can be made, e.g. addition of specialist terms or proper names to the dictionary. User-specific settings for translation options are also not possible, e.g. choosing between American English and British English, between old and new German spelling rules or searching for unknown words in the text. The translation quality can suffer as a result. The translation produced by free services is often limited to a certain number of characters.
Full text translation programs for the PC have their own user interface or can – depending on the manufacturer – also be directly integrated into the text processing program, e.g. Microsoft Word.
– Advantages: Users can work in their usual text processing environments. The programs can – depending on the versions – be extended with own vocabulary and specialist dictionaries. This improves the quality of the translation. They also offer individual adjustment possibilities. Other functions are generally included, e.g. look-up of words in the dictionary and website translation.
– Disadvantages: Linguistic knowledge may be required – depending on the intensity of use. This means that, depending on the prior knowledge of the user, they may need to be trained into the program. A serious disadvantage of the online translation tool is that you are releasing your texts onto the Internet with all its data insecurity. Confidential or delicate content should not be released into unknown portals.
Fact: Free translation services are there to help you translate short texts and to give you an idea of the content of foreign-language texts as well as for the casual translation task.Translation programs for the PC allow you to adapt specialist vocabulary to meet sector or company requirements. They are a suitable tool for specialist texts and for frequent translation needs.