Error
Code
Error Type Error Description
2A Article usage An article usage error occurs when (1) the incorrect article is used, (2) when an article is needed, but none is used, or (3) when an article is not needed, but one is used.
2B Preposition usage A preposition usage error occurs when (1) the incorrect preposition is used, (2) when a preposition is needed, but none is used, or (3) when a preposition is not needed, but one is used.
2C Orthography
(Spelling,
Capitalization, &
Punctuation)
A spelling error occurs when a word in the translation is spelled incorrectly according to target-language conventions. A capitalization error occurs when the conventions of the target language concerning upper and lower case usage are not followed. A punctuation error occurs when the conventions of the target language regarding punctuation are not
followed, including those governing the use of quotation marks, commas, semicolons, and colons. Incorrect or unclear paragraphing and spacing is also counted as a punctuation error.
2D 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 unnatural word order and lack of agreement between subject and verb.
2E Tense A verb tense error occurs when the verb in the target language does not correctly indicate the time of the occurrence of an action in the source text.
2F Word form A word form error occurs when the root of the word is correct, but the form of the word is incorrect or nonexistent in the target language.
2G Word usage A word usage error occurs when a verb, noun, adjective, adverb, pronoun, preposition, conjunction or interjection is used incorrectly according to target-language conventions.
2H Miscellaneous A language error occurs that cannot be described by any of the other language error codes.
Error Marking
All errors for Translations 1 and 2 were coded using the error typology developed by the researcher. If an error occurred at the level of a single word, the error code appears in bold immediately after that word. Take for example the following sentence from Student 1, Translation 1:
Original text:「台灣人能活到『七老八老』，主要歸功於社會經濟快速繁榮」
Student translation: “Taiwanese’2F long lifespan1F 2G mainly attributes2F to the burgeoning2G 1D economy.”
Since an apostrophe cannot be added to a word that ends in e without also adding s, the word Taiwanese’ is marked with the error code 2F to indicate that the word form (here, the suffix) is incorrect. The second error in the above example is more
complicated. First of all, the word lifespan should be plural, since it refers the lifespans of all Taiwanese. However, the word life expectancy is the correct term, since lifespan refers to the actual life from birth to death of an individual, while life expectancy is an average of the lifespans of the population. For the purposes of the study, words are only marked with one error code. In the case that a word has multiple errors, only the most serious error is marked. Since it was determined that the main error is that of incorrect terminology, and that the incorrect suffix is a secondary error, the word lifespan is marked with the code 1F. The third and final errors in the example are those of missing words. Clearly the word is should have been used in the first case, and social (「 社會」) should appear before economy because it is used in the original Chinese text. When a word or words should have been used, but were omitted, the error code appears bolded, underlined and standing alone to show the missing word or words, as in 2G and 1D above. In the case that
there are multiple words, if they are a lexical chunk, they are only counted once and not as separate errors. Similarly, if words that appear in the translation are a lexical chunk and also an error, the chunk is considered to be one error. Take for example the use of a lot of in Student 1’s Translation 1, paragraph 2 (see Appendix A). The
meaning of the phrase is correct, but the phrase itself is informal and not suited to the register of the translation. Therefore, the entire phrase is counted as one error and marked thusly: a lot of1E. Hence, all instances of underlined words signify that the phrase has been marked as one error. The only cases of double error coding occur with syntax. In such cases, individual words in the sentence were coded for errors, and the syntactic error was counted once and shown by underlining the sentence or phrase where it arose. After errors were coded, they were then recorded on the error totals chart and calculated (see Table 4).
Table 4
Example of Student 1’s error-coded Translation 1 Error totals
1. Style/Rendition Errors
# of Errors 2. Language Errors # of Errors
1A: Addition 4 2A: Article usage 5
1B: Literalness 5 2B: Preposition usage 9
1C: Mistranslation 5 2C: Orthography 2
1D: Omission 5 2D: Syntax 1
1E: Register 8 2E: Tense 1
1F: Terminology/Cohesion 8 2F: Word form 8
1G: Miscellaneous 2 2G: Word usage 10
2H: Miscellaneous 0
Style total 37 Language total 36
Translation Scoring Scale
In order to make comparisons between Translations 1 and 2, and in order to make comparisons among all study participants’ translations, the researcher
developed a 200 point scoring scale in which style/rendition and language usage are weighted equally at 100 points each (see Table 5). The reasoning behind this was twofold. Firstly, the researcher wanted to be able to make meaningful comparisons between the two main categories and, by weighting them equally, this was more easily accomplished. Secondly, the researcher believes that style/rendition and language usage both play an equally important part in high-quality translations and this is represented in the scoring method.
Table 5
Translation scoring scale
Style/rendition (50%) (100 points)
Language Usage (50%) (100 points)
Points subtotal /100 /100
Style + language score /200
Percentage score %
In order to apply the scoring scale to the study participants’ translations, it was necessary for Translation 1 to be divided into ten sentences (from its original nine).
Translation 2 presented no problems as it was already divided into ten sentences.
Thus, each sentence in Translations 1 and 2 was awarded a total of ten points for Category 1: Style/rendition and ten points for Category 2: Language usage, for a total of 20 points per sentence. A one-point penalty was assessed for each error, with the maximum possible deduction of ten points per sentence per category. The maximum possible deduction was 20 points per sentence, regardless of the actual number of errors committed. Thus, if ten errors were made from both Categories 1 and 2, the participant received a score of zero for that sentence. However, if 11 errors were made from both Categories 1 and 2, the participant still received a score of zero for that sentence and was not further penalized. This scoring scale allowed the researcher
to compare and rank the study participants’ translations and to assign each one a score based on the percentage of correct translations per sentence.
Data Collection Procedure
In November 2013, research participants were recruited from among the researcher’s students. Data collection was carried out over a three-month period from December 1, 2013, to March 1, 2014. Due to time and space constraints and for the sake of convenience, the research participants were allowed to perform the
translations at home. At the beginning of each month of the study, the students
received, in chronological order, a translation from previous translation exams held by Taiwan’s Ministry of Education (MOE). The students then had a month to complete the translation (although the suggested time for completion was one to two hours) and were told dictionaries and grammar books could be used, but were asked not to request outside help or use online translators or the Internet. The reasoning behind this, and the reason communicated to the students, was so that their real results could be detected and analyzed in order to help them avoid similar mistakes in the future.
Students were intentionally not given any specific instructions on translation
methodology before the first translation. This aspect of the project was left ambiguous in order to determine what type of techniques students would use and to ascertain which were most successful.
After a student submitted a translation, the researcher corrected their errors and provided other feedback, such as general comments and suggestions on how to improve, via email. This served a dual purpose. Firstly, it was hoped that the feedback would encourage students to continue the project and help them learn from their
translations. Additionally, this gave the researcher the chance to compare Translation 1 with Translation 2 and analyze how much, if at all, students had improved.
Once the translations were complete, the participants with the five highest and five lowest scores (evaluated on a preliminary global basis by the researcher) were then selected and invited to partake in personal interviews. The interviews were scheduled at the participants’ convenience and took place in the relaxed atmosphere of a coffee shop. Before the interviews, participants received a list of ten questions in English and were asked to prepare answers beforehand. The interview itself consisted of two parts. During the first part, the researcher asked the participant the ten
questions (see Appendix D). In the second half of the interview, the researcher went over the students’ translations with them and asked about any special features or asked for clarification about students’ motivations for translation choices. The flowchart for the data collection procedure is shown in Figure 1.
Figure 1. Data collection procedure for translation error analysis.
Errors and scores for Translations 1 & 2 were statistically analyzed The scoring scale was designed, tested, modified and finalized
Translations 1 & 2 were error coded
The error typology was designed, tested, modified and finalized Research participants completed a voluntary online biographical
questionnaire
Mar. 1 - Apr. 31, 2014: Retrospective interviews were conducted Feb. 1 - 28, 2014: Participants completed Task 3 (Translation 2) & received
feedback
Jan. 1 - 31, 2014: Participants completed Task 2 & received feedback Dec. 1 - 31, 2013: Participants completed Task 1 (Translation 1) & received
feedback
Nov. 2013: Research participants were recruited
Dec. 1, 2013: Task 1 (Translation 1) was sent to participants via email
Chapter 4: Results & Discussion
After the translation data had been collected, the researcher developed an error typology to evaluate, assign scores to, and rank the linguistic and stylistic errors of the research participants’ translations. The errors were then hand-coded and the results were tallied and quantified. The main types of errors most commonly made by students were identified, and the frequency of these errors was calculated.
Participants’ translation scores were then compared to determined if and how they had changed from the first translation to the second. Finally, personal interviews with the highest- and lowest-performing students were conducted to discover what general commonalities they shared and to ascertain which translation techniques used were the most effective. The most relevant portions of the interviews have been transcribed and are discussed at the end of this chapter.
Development of the Error Typology The Error Typology Prototype
The first research question of this study concerns how to develop an error typology for Chinese-to-English translation that will accurately detect and sufficiently describe student translation errors. The researcher’s error typology drew inspiration from two different sources. The typology’s two main categories, which are
style/rendition errors and language errors, were inspired by Anthony Pym’s (1992) binary and non-binary classification of errors. In Pym’s system, binary errors are either clearly correct or incorrect, whereas non-binary errors are errors in which there is no definitive right or wrong answer. Likewise, in the researcher’s error typology, style and rendition errors are a nod to non-binary translations, while the language error categories are more definitively binary errors. By separating the errors in this
manner, it was anticipated that insight could be gleaned into the translation difficulties the study participants faced. It was hoped that the results might suggest participants’
overall strengths and weaknesses, which could then be generalized and serve as guide for teachers to target problematic areas. Additionally, since the category of
style/rendition is often viewed as being subjective, it was thought that a more detailed and objective error typology could be developed to better ensure fairness in the error coding and scoring of translations.
The second source for the researcher’s error typology was the American Translators Association (ATA) Framework for Standardized Error Marking Explanation of Error Categories. The ATA lists 23 different errors: addition,
ambiguity, capitalization, cohesion, diacritical marks/accents, faithfulness, faux ami, grammar, illegibility, indecision, literalness, mistranslation, misunderstanding,
omission, punctuation, register, spelling, style, syntax, terminology, unfinished, usage, and word form/part of speech. Some of the ATA’s categories were deemed unsuited to the current study, however, for various reasons. The categories of “diacritical marks/accents” and “faux ami”, for example, are more relevant to translations of Romance languages into English, rather than Chinese-to-English translations. Other categories, such as “grammar”, “syntax”, and “word form/part of speech”, seemed to overlap. It appears that “grammar” would encompass “syntax” and “word form/part of speech”; thus, in the researcher’s error typology, the grammar category has been parsed into several subcategories in order to better describe the most common types of grammatical errors that occurred. Due to the nature of the current study, the ATA’s categories of “unfinished” and “illegibility” did not apply, as participants had ample time to complete their translations and all translations were typed. Finally, the
researcher believes that the ATA’s categories of “faithfulness” and “misunderstanding”
are too subjective for the current study. According to the ATA, “A faithfulness error occurs when the target text does not respect the meaning of the source text as much as possible. Candidates are asked to translate the meaning and intent of the source text…”
Since it is possible that source texts may be interpreted slightly differently by different translators, and since the researcher felt that objectively determining the
“intent” of the source text would be rather difficult, this category was not utilized.
Likewise, the researcher felt it nearly impossible to ascertain that a participant had misunderstood the source text without conducting personal interviews. Thus, this category was also removed.
Hence, the researcher’s error typology for this study, in keeping with Pym’s non-binary and binary errors, has divided translation errors into two main categories:
style/rendition errors (non-binary) and language errors (binary). The subcategories borrow heavily from the ATA’s error categories, but have been combined or split when necessary in order to describe in a detailed manner the types of errors committed by study participants without being too broad or too specific. Table 6 shows the prototype of the current error typology.
Table 6
Error Typology Prototype
1. Style/Rendition Errors 2. Language Errors
1A: Addition 2A: Punctuation
1B: Cohesion 2B: Spelling & Capitalization 1C: Literalness & Mistranslation 2C: Syntax
1D: Omission 2D: Verb tense
1E: Register & Terminology 2E: Word usage
1F: Miscellaneous 2F: Word form
2G: Miscellaneous
The prototype shown in Table 6 was used in a pilot test to mark the errors of five student translations for Translation 1. It was found that the first and second most common Style/Rendition errors were Literalness & Mistranslation (1C) and Register
& Terminology (1E), respectively. It was felt that these categories were too broad, and so for the current study, they became four separate categories: literalness, mistranslation, register, and terminology. Using the prototype typology, Cohesion (1B) and Miscellaneous (1F) errors were found to be the lowest. Therefore, since cohesion errors often coincided with terminology errors, cohesion and terminology were combined into one category for the current error typology. The third and fourth categories with the highest number of errors for the prototype were Omission (1D) and Addition (1A). These two categories were nearly equal in number of errors, with omission being slightly higher. Thus, since they each had relatively high error frequencies, it was decided that combining them would not be useful because they serve as better descriptors of students’ abilities when they stand alone. There were only three Miscellaneous (1F) errors, which along with cohesion, accounted for the fewest number of errors. This was a good indication that the error typology prototype was broad enough to describe most of the errors in the participants’ translations.
When employing the prototype typology to mark five student translations from Translation 1, the most common language errors were Word usage (2E) errors, which accounted for almost half of all language errors. Thus, the category was deemed too expansive in scope and somewhat problematic, as it did not appear to be descriptive enough of the type of errors that were occurring. In this preliminary analysis, it was found that the use of articles and prepositions was particularly problematic, with students adding or deleting them at will. For this reason, two new categories were created for articles and prepositions, which helped take some of the
burden off the Word usage (1E) category. The most common language error after word usage was Word form (2F). The results were unsurprising as Mandarin Chinese does not conjugate verbs or employ plural word endings, and the category remained unchanged. Following word form, categories Syntax (2C) and Punctuation (2A) were almost equal. Somewhat surprisingly, Spelling & Capitalization (2B) had few errors, but this is probably due to a lack of time constraints and students’ free access to dictionaries and grammar books. Because of the small number of errors in spelling and capitalization, Punctuation (2A) and Spelling & Capitalization (2B) were merged into one category. The Syntax (2C) category remained intact, but the definition was broadened from simply unnatural word order to also include lack of subject/verb agreement. Utilizing the prototype typology, Verb tense (2D) accounted for the lowest number of errors, which was quite surprising based on the results of previous studies.
Verb tense (2D) was not, however, combined with another category as it was thought that a larger sample size would perhaps yield different results. The fact that there were no miscellaneous errors in the language errors category was a good indication that the typology was comprehensive enough to describe most errors that occurred.
The Finalized Error Typology
Thus, the error typology for the current study is as follows: Category 1:
style/rendition errors, with the subcategories of addition (1A), literalness (1B), mistranslation (1C), omission (1D), register (1E), terminology and cohesion (1F), miscellaneous (1G), and Category 2: language errors, with the subcategories of article usage (2A), preposition usage (2B), orthography (spelling, capitalization, and
punctuation) (2C), syntax (2D), verb tense (2E), word form (2F), word usage (2G), and miscellaneous (2H). Please refer to Tables 7 and 8 for concise error descriptions.
Table 7
Error Typology for Category 1: Style/Rendition Errors Error
Code
Error Type Error Description
1A Addition
An addition error occurs when the translator introduces superfluous information, including lexemes and stylistic effects.
1B Literalness A literalness error occurs when a translation that follows the source text word for word results in awkward, unidiomatic, or incorrect renditions.
1C Mistranslation A mistranslation error occurs when the meaning of a segment of the original text is not conveyed properly in the target language.
1D Omission An omission error occurs when an element of information in the source text is left out of the target text or when the translator fails to include lexemes necessary in the target language according to target-language conventions.
1E Register A register error occurs when the language level or degree of formality produced in the target text is does not correspond to that of the source text.
1E Register A register error occurs when the language level or degree of formality produced in the target text is does not correspond to that of the source text.