When I tell people what I do for a living, they sometimes look at me in surprise and respond by saying: "But can't you just get a computer to do all that for you nowadays?"
The answer, of course, depends on what they mean by "all that" and is largely determined by the level of quality required by the end user.
MT (machine translation) is certainly capable of delivering usable results in some contexts. For example, an extremely successful MT system called METEO was developed in Canada to translate weather reports. However, weather forecasts clearly rely on a finite set of descriptions and grammatical structures that can be mapped relatively easily and it should be noted that the weather reports submitted to METEO for translation had to follow a standard format and adhere to a fixed vocabulary. Even then, the accuracy rate achieved was only estimated to be around 90%.
Similarly, the United States Air Force and NASA are both known to have used MT for translating from Russian into English, but they only required a gist translation and so the output of the MT engine was, once again, sufficient for their needs.
Where MT does not fare so well is when you want it to produce an accurate and polished translation of a complex document, such as a legal contract, or if you require a publishable piece of marketing copy or a comprehensible set of user instructions. This is because human language is highly complex and meaning is often buried beneath the surface of the words and structures used.
For example, I once translated a document that contained the term "Gesundheitspräventionsprogramm", which was listed as a benefit of working for a particular employer. Despite knowing exactly how this word ought to be translated (i.e. as "preventative health programme"), I thought it would be fun to see what a well-known translation engine would make of it. As I had anticipated, it simply deconstructed the component parts of the word in the exact order they appeared and returned the translation "health prevention programme". I am not sure about you, but I certainly would not be happy working for somebody with that kind of a programme! In this context, the MT engine failed because it was unable to relate the words it was translating to the real world. Unlike a human translator, it is simply attempting to match patterns, rather than actually trying to grasp the underlying meaning.
Given the many technological advances we have seen in recent decades, it is perfectly reasonable to ask whether computers might now be able to translate natural language. However, unless you are looking for a very rough translation, the answer is "no" and this is likely to remain the case for the foreseeable future.