The success of a machine translation depends on two things: (1) the website used and (2) the target language.
For the target language, it can generally be said that the closer the target language is linguistically, the more successful the translation will be. For English, translation into other Indo-European languages such as French, German and Spanish renders relatively accurate, readable content. There may be occasional grammar errors and an odd translation here and there, but overall the result is understandable and unproblematic. On the other hand, translation into Japanese and Chinese often renders laughable if not dangerously inaccurate results. This is true as well for human translation, where knowledge of the differences between the source and target languages will make the difference between a bad and a good translation. Although AI has improved over the years, it still cannot overcome basic problems like a human can. Critically important for translation from Japanese to English into Western languages and Chinese, it cannot provide information that is not in the original text.
Consider the following experiment (Yoneoka, Saito, and Tomei 2015?):
This sign in front of a computer room was translated by XX students majoring in English, Chinese and Korean first on their own (human translation) and then using the Higosign method.
Sign X. マナーを守ってご利用ください。 Please observe and protect manners.
The translations were judged by 2 native speakers for each language using a 5 point scale. Here are samples for each result.
Human translation samples: 1. Please observe the rules when you use a computer. Perfectly understandable. 2. Please keep the manners in the computer room. Understandable, but "keep the manners" is awkward. 3. Please use this room to fallow the rule. Spelling mistake, grammar mistakes prevent understanding. 4. Please protect manners and use. Word choice, lack of object prevent understanding. 5. Please following the manor and using there. Grammar, spelling, lack of object, mistaken pronoun
The average ratings for each language were: English 3.6, Korean 1.38, Chinese 3.13. That is, human translation produced relatively successful results for Korean, but were much less successful for English and Chinese.
After the Higosign method was used, here are the output samples: 1 Please mind your manners when you use the computers. 2 When you use this computer room, please use it politely. 3 Please observe the rule of the computer room. And, please use those computer. 4 Please refrain from the nuisance of the computer room. It becomes troublesome other users. 5 People to use these computers need to conform to the courtesy. 5 this place should observe and use rules.
The average ratings were: English 3.18, Korean 2.36, Chinese 2.05. This means that English and Chinese improved, but Korean went down!