One of the most important tasks performed by Trust Miners is awarding Exfluenscores to Enhancers based on the number and severity of errors corrected and labelled in the Enhancer's work. A new Quality Assurance Table will be used across the Exfluency™ platform for categorising errors and assigning severity to each of them. The higher the severity, the more serious the error, and the lower the final Exfluenscore™.
Trust Miners’ role is a challenging one. On the one hand, they are required to maintain top-notch quality of target texts by correcting and marking every error. On the other hand, they do not want to ruin the reputation of their colleagues who work as Enhancers. Bearing in mind the fact that a lower Exfluenscore will eventually damage an Enhancer’s Trust Chain™ – which is immutable and cannot be repaired by deleting scores received so far –, Trust Miners’ decisions should be prudent and well-informed. The purpose of this guide is to help in making such decisions. However, it is addressed to both Communities involved, as Enhancers will also benefit from understanding the differences among individual error categories as well as their characteristics.
A word of advice to Enhancers. For the benefit of a harmonious collaboration between the two Communities, when you disagree with the error category used by a Trust Miner™, but the severity of your counter-proposed category is the same as the original one, there is no point in splitting hairs and discussing the issue. After all, at the end of the day the Exfluenscore will stay unchanged.
|
Error classification |
Severity |
|
Preferential |
0 |
|
Formatting |
1 |
|
Addition |
3 |
|
Style |
3 |
|
Ambiguity |
5 |
|
Grammar |
7 |
|
Orthography |
7 |
|
Value |
7 |
|
Mistranslation |
9 |
|
Omission |
9 |
|
Terminology |
9 |
How is Exfluenscore calculated?
Each document is divided into linguistic assets. Each asset within the document has the exact same value (the number of words does not matter). The highest score for an asset is 9. When the Trust Miner decides that the asset is translated with no errors, then a score of 9 is assigned automatically. If there is one or more errors inside one asset, all should be corrected but only one with the highest severity must be marked.
Example 1. If the document contains 10 assets and all of them are correct, the score will be calculated as follows:
(10 assets x score of 9) : 10 = 90 : 10 = 9.0
Example 2. If the document contains 10 assets and one asset has a Style error whose severity is 3, the score for that particular asset is 9 (the maximum number of points per asset) - 3 (style error) = 6. If the remaining 9 assets are free from any errors and only one asset has the Style error, the Exfluenscore for the project given to the Enhancer will be calculated as follows:
(9 assets x score of 9) + (1 asset x (score of 9 - 3 for error)) : 10 = (81+6) : 10 = 87 : 10 = 8.7
Preferential
Preferential changes must be made seldom, as Trust miners should only make changes to assets that are incorrect. When Trust Mining, you might find an error that is repeated throughout the project. A severe error marked again and again can have detrimental effects on an Enhancer’s Trust Chain™. When you encounter an error for the first time, make sure to mark the error referring to the Error Classification Guide and leave a justification. If you encounter the same error again within the same project, please mark this as Preferential. The Preferential error carries a severity of 0, therefore it doesn't have a detrimental effect on the Enhancer's score. This way, Enhancers are marked fairly and provided with an explanation.
Formatting
Formatting errors include wrong or missing formatting attributes and/or placeholders, therefore this category is scored 1. Enhancers should ensure all formatting attributes and placeholders are placed correctly in the assets.
Addition
An addition error occurs when superfluous elements of meaning are introduced, including aspects of tone (irony, intensification, etc.). However, explicitation is permissible. Explicitation is defined as “[…]a translation procedure where the translator introduces precise semantic details into the target text for clarification or due to constraints imposed by the target language that were not expressed in the source text, but which are available from contextual knowledge or the situation described in the source text.” (Delisle, J. (1991). Translation Terminology, Foreign Language Teaching and Research Press, p. 139)
An example of necessary addition, i.e. explicitation, could be observed in a translation of an article reporting a president's speech on peace. Since English uses natural gender, a pronoun replacing the noun 'president' will be 'he', whereas 'peace' would become 'it'. In languages with grammatical gender, such as German or Polish, the two nouns may both be 'he'. Consequently, if a pronoun is used in the English source text, a full noun may be necessary to avoid ambiguity in the translation.
Style
General
Style errors include any target-language errors that do not change the meaning of the source text or clearly violate rules of spelling, grammar, or punctuation, but detract from the quality of the translation with nonidiomatic, inappropriate, or unclear wording/phrasing.
NOTE: Errors resulting from machine-translation artificiality and any unnatural language produced by MT should be corrected and classified as Style.
Usage and word collocations
A usage error occurs when conventions of wording or phrasing in the target language are not followed (“We don’t say it that way”). This category includes idiomatic use of prepositions (e.g., “married to,” not “with”) and collocations (“committed a crime,” rather than “performed a crime”). NOTE: If incorrect usage ends up changing the meaning (e.g., “She took a train in Berlin,” when “to Berlin” is meant; or “He illustrated the conclusion” rather than “He drew the conclusion”), this should be marked as a mistranslation error.
Register
A register error occurs when incorrect verb forms are used (tú vs. usted in Spanish or du vs. Sie in German) or vernacular forms/vocabulary are employed when formal ones should be used (for example because the text is addressed to educated readers or has a formal character).
Cultural norms
Phrases that are incorrect in a given context or that are not used at all in a native-level language. This applies for example to formal/informal greetings, letter opening and closing formulas, use of imperative, etc.
Ambiguity
The simplest example of ambiguity can be found on a lexical level. For example, the word "gezellig" in Dutch can be translated as "cosy", "friendly", "comfortable", or "homey", yet none of these precisely captures the full meaning of the word*. A similar example is the Polish noun “żal” that can have English translations as different as “sorrow”, “remorse”, “grudge”, or “pity”. In a translation from Polish into English, the noun has to be disambiguated because there is no single equivalent in the target language.
Another level, where ambiguity can impede correct interpretation of a text, is syntax. See the example below.
Syntactical ambiguity can end up being particularly problematic in instructions and patient information leaflets. Examples of ambiguous instructions can be found here.
However, in situations where the source language can be interpreted in more than one way due to its grammatical characteristics, selecting only one option in the target language should not be considered an ambiguity error. As an example, in English the pronoun ‘you’ plus a verb can be translated either as the 2nd person singular or the 2nd person plural. Another example are the two Spanish verbs ‘ser’ and ‘estar’. Both can only be translated as ‘to be’ in English.
Grammar
General
Grammar errors include lack of agreement between subject and verb, incorrect verb inflections, and incorrect declension of nouns, pronouns, or adjectives.
Incorrect: “The boy don't like dancing.”
Correct: “The boy doesn’t like dancing” or “The boys don’t like dancing.”
Syntax
A syntax error occurs when the arrangement of words or other elements of a sentence does not conform to the syntactic rules of the target language. Errors in this category include improper modification, lack of parallelism, unnatural word order, and run-on structures.
Incorrect: “German charming blue old plastic big Trabant.”
Correct: “Charming big old blue German plastic Trabant.”
Word Form
A word form error occurs when the root of the word is correct, but the form of the word (e.g. number or case of noun or pronoun) is incorrect or non-existent in the target language.
Incorrect: “It was a conspiration.”
Correct: “It was a conspiracy.”
Orthography
Punctuation
Incorrect: “There is however another type.”
Correct: “There is, however, another type.”
Capitalization
Incorrect: “…the jewish festival of the passover.”
Correct: “…the Jewish festival of the Passover.”
Spelling
Incorrect: “Their bot equaly intresting.”
Correct: “They're both equally interesting.”
Value
A value error occurs when a number, date, or time is incorrect or has not been properly localised.
Mistranslation
The most important aspect of any mistranslation error is a change in meaning in the target text compared to the intended meaning in the source text.
Faux amis
A faux amis error occurs when words of similar form but dissimilar meaning across the language pair are confused. A classic example of a false friend are English 'actually' and Spanish 'actualmente' - they look similar but one cannot be used as a translation of the other.
Cohesion
A cohesion error occurs when a text is hard to follow because of inconsistent use of structural elements such as terminology, pronouns, inappropriate or missing conjunctions, etc. Cohesion is the network of lexical, grammatical, logical, and other relations that provide links between various parts of a text, assisting the reader in navigating the text. Although cohesion is a feature of the text as a whole, Trust Miners need to mark errors for individual elements that disrupt cohesion.
An example of a cohesion error would be changing a translation of a word for no reason. If a 'cup' is translated consistently as a 'cup' and then in one place the same 'cup' becomes a 'glass', we have a cohesion error.
Faithfulness
A faithfulness error occurs when the target text does not stick to the meaning of the source text as closely as possible. The translation should convey the meaning and intent of the source text without rewriting or improving it. If a “creative” rendition changes the meaning, an error should be marked. If recasting a sentence or paragraph - i.e. altering the order of its major elements - breaks the flow, changes the emphasis, or obscures the author’s intent, an error should be marked.
Literalness
A literalness error occurs when a translation that follows the source text word for word results in an awkward and/or unidiomatic rendition.
Misunderstanding
A misunderstanding error occurs when the translation clearly results from a misinterpreted word or idiom, or the incorrectly parsed structure of a phrase or sentence.
Indecision
An indecision error occurs when more than one option for a given translation unit is offered. Even if both options are correct, an error should be marked.
Inconsistent translations of a term or phrase at different points in the passage may constitute a cohesion error.
Other mistranslations
Any error that changes the meaning of a word, phrase, or passage but does not fit into the previous subcategories is marked as a mistranslation error.
Omission
An omission error occurs when one or more elements of meaning in the source are left out of the target. This covers not only textual information but also the author's tone (irony, intensification, etc.). Implicitation is permissible, defined as “[…]a translation procedure intended to increase the economy of the target text and achieved by not explicitly rendering elements of information from the source text in the target one when they are evident from the context or the described situation and can be readily inferred by speakers of the target language” (Translation Terminology, p. 145).
Implicitation, e.g. omitting words or phrases because they cannot or should not be translated, is the other side of the explicitation coin. Consider, for example, an expletive or pleonastic use of pronouns 'it' and 'there'. In many languages, translations of 'there is a dog in the room' or 'it is raining' will have to omit the pronoun – this is implicitation.
Terminology
A terminology error occurs when a content word or phrase (noun, verb, modifier) is chosen with an incorrect or inappropriate meaning in relation to the source text. This error also applies when a term appropriate to a specific subject field is not used when the corresponding term is used in the source text. In Exfluency, the terminology error subcategory includes the cases where a terminology database is not followed.
* Tokowicz, N. (2022). Translation Ambiguity Affects Language Processing, Learning, and Representation.
Comments
0 comments
Article is closed for comments.