Translation Errors
Carnegie Mellon

Machine Translation error examples


wrong word order | wrong sense of the word |
wrong number agreement | wrong person agreement | wrong gender agreement | wrong tense agreement |
incorrect word | no translation |
translation requires an extra word | translation is missing a word |


wrong word order

English has a rather fixed word order. If the words are not in the right order and, unless they are moved around, the sentence is not grammatical. For example, if you were to correct the translation:

We were looking forward climbing the mountain really tall.
you would need to drag "mountain" and drop it at the end of the sentence, after "tall":
We were looking forward climbing the really tall mountain.

You need to make sure that the alignments are correct, namely, if the source language was Spanish, and we had "...la montaña muy alta", "really" would need to be aligned with "muy", "tall" with "alta" and "mountain" with "montaña". [explain in more detail, once the interface is done]



Need to edit a word
wrong sense of the word

The same word in one language can have more than one meaning or sense and thus more than one translation in another language.
If we are translating from Spanish to English, the word "banco" can mean both "bank" and "bench". Suppose we only have [banco->bank] in the lexicon and are given the sentence "me senté en un banco" to translate, then the Machine Translation system would produce:

I sat on a bank
even though, in principle, it is possible that the Spanish sentence means that somebody is seating in a financial institution, world knowledge tells us that it is much more likely that somebody is seating on a bench. And thus, you would need to modify the English translation above to say:
I sat in a bench


wrong number agreement

In English, some words agree in number, such as determiners and the nouns they specify, as well as subjects and verbs. And example of wrong number agreement between a noun and its determiner can be seen in the following example:

There was no way we could go up without a ropes
should be:
There was no way we could go up without a rope (singular)
or:
There was no way we could go up without ropes (plural)

depending on whether the original word for "rope(s)" is singular or plural. In the first case, the word "ropes" would need to be edited (by clicking inside its box) and you would need to indicate that the source of error is "wrong number agreement" and that it should agree with "a". In the second case, you just need to drag "a" to the trash box.
An example of wrong number agreement between subject and verb can be seen in this example:
They likes to climb in the winter
which should be:
They like to climb in the winter (plural)
For more examples of the last type of error, see
person agreement



wrong person agreement

In English, the subject and the verb agree in person. Even though most verbs have just 2 forms according to the person and the number (3rd person singular vs. non-3rd person singular), some verbs like "to be", actually have a richer conjugation. If the subject and verb do not agree in person and number, the sentence is not grammatical. Here are a couple of examples with wrong person and number agreement:

Laura am a good sport climber
should be:
Laura is a good sport climber
(3rd person singular)

and:
We is going down to Seneca Rocks this weekend
should be:
We are going down to Seneca Rocks this weekend
(2nd person plural)

To correct this type of errors, you need to click on the verb box to edit it, change it to the right form and click on the "wrong person agreement" choice. In the first case, the verb needs to agree with "Laura", in the second case, it needs to agree with "we".



wrong gender agreement

In English, there are very few words that show gender agreement (an example being personal pronouns which need to agree in gender and number with their antecedent). However, in other languages, gender agreement is much more important, since determiners, adjectives, nouns and (some) verbs are marked with gender.
Here is an example of wrong gender agreement in Spanish:

El silla alto era roja
should be:
La silla alta era roja

since "silla" is the head noun of the noun phrase and it's a feminine noun, both the determiner ("el") and adjective ("alto") need to agree in gender (and number). Note that the predicative adjective "roja" also needs to agree with the noun. When correcting this type of error, you need to click on both "el" and "alto" and change them to their feminine form ("la" and "alta") and indicate that they need to agree with "silla".



wrong tense agreement

Verbs have different tenses and aspects (present, past, future, progressive or continuous, etc.), and there usually isn't a one to one correspondence between the tenses and aspects of one language and those of another language. However, with the tense in the verb in the source language and the context, we can almost always determine what is the right tense of the verb in the target language. For example, if we were translating from Spanish to English, and we had the sentence "estábamos esperando que Juan pusiera las áncoras", where the verb is in the past tense, if the translation is:

We wait for John to put the anchors
we know that we should correct it to be:
We waited for John to put the anchors
or maybe:
We were waiting for John to put the anchors

In order to do that, you should click on the verb "wait", change it to the correct form and click on "wrong tense agreement".



incorrect word

Some times languages use lexicalized expression, namely, two or more words which stopped having their literal meaning and now constitute a single entry with a fixed meaning. An example of lexicalized expression can be seen in "John kicked the bucket", which has nothing to do with kicking or buckets. If we have to translate this sentence in Spanish, we wouldn't translate it as:

Juan pataleó el cubo
but rather:
Juan se murió

So in this case, you should probably edit "pataleó" and change it to "se murió" (make sure that's how I want them to do this, 2 words?), indicating that the source of error is "incorrect word", and then drag "el" and "cubo" to the trash bin.

On the other hand, some words have the same or a very similar meaning, but are used in different context, and if the wrong word is used for a particular context, there is a lexical error. It can be that a word changes according to the phonetic context, for example:
We had an sixty meter rope and two ninety meter ones
should be:
We had a sixty meter rope and two ninety meter ones

according to its syntactic context:
all climber was patiently waiting for her turn
should be:
every climber was patiently waiting for her turn
or according to its semantic context:
In English some names are countable and some are mass nouns, and there are some quantifiers that can only co-occur with countable nouns (many) and some other which can only quantify mass nouns (much). An example of errors due to this property of names is:

It was a nice spot, there were some flowers and much trees
which should be:
It was a nice spot, there were some flowers and many trees

Moreover, some words have specific selectional restrictions, and cannot appear with some other words. For example:
It was very freezing
should be:
It was freezing
or:
It was very cold

Another common case of selectional restrictions is the cases of governed prepositions, namely some verbs (and a few nouns) are always followed by the same (set of) preposition(s), and the use of other prepositions in conjunction with those verbs results into an ungrammatical sequence.
For example, if we were translating the Spanish sentence "Yo estaba preocupada sobre la casa" to English, it is possible that we would get the following ungrammatical translation:
I was worried on the house

but we know that in English, the transitive form of "to worry" is always followed by the preposition "about", and thus you would need to correct the translation so that is says:
I was worried about the house
to do that, you just edit "on" (by clicking on it) and indicate that the source of the error is "incorrect word".

It could also be that there is a morphology error. English is a fairly regular language, most plurals are obtained by adding an "s" to the singular noun, most simple past tenses are obtained by adding "ed" after the verb stem, etc. However, there are some exceptions, what are called irregular nouns and verbs, for example:

Too bad John and Sean's wifes could not make it
should be:
Too bad John and Sean's wives could not make it



no translation provided

Some times there is a word that is not in the lexicon and therefore our system cannot provide any translation for it, and thus the system outputs the word in the original language. In such cases, you will just need to edit the word in the original language and provide with the right translation in this sentence, indicating that the word was not translated.



Need to add a "New Word"
translation requires an extra word

When translating from one language to another, more words might be needed in the target language than there are in the source language. For example, if we're translating English to Spanish, and there is a direct object, often times it won't be preceded by a determiner, but in Spanish, direct objects typically need a determiner.
So the English sentence, "I took the kids to school", would not be translated:

Llevé los niños a escuela
but rather:
Llevé los niños a la escuela

In order to do that, you need to click on "New Word" and write the word you want to add, in this case the determiner "la", then drag it to the right position indicating that there was a missing word (you might need to move the current words a bit to make some space) and make sure it is aligned to the right word. In this case, both "la" and "escuela" should be aligned to "school".



translation is missing a word

Since our system still needs to be refined, there might be words in the source language that are just missing in the translation. An example of this could be the translation from "quieres vino?" into English:

want wine?
which should be:
do you want wine
In this particular case, whereas "do" is just missing from the translation, "you" could be considered to be an extra word required by the translation, since English sentences always requires a subject, unlike Spanish.
In either case, you need to click on "New Word" and write the word you want to add, dragging it in the right position and indicating what is the source of error.









Ariadna Font Llitjos
Last modified: Tue Sep 24 13:36:40 EDT 2002