1
CHAPTER I
INTRODUCTION
A. Background of the Study
In our current world, there is no more limitation since technology is
developing rapidly. People are facilitated sufficiently to get everything easily,
including information. There is a lot of information accessed by people from
media. The media can be written media such as newspapers, books, or magazine,
audio-visual media such as television and radio, and online media such as internet.
In accessing them, however, people still get obstacle in case of language.
Therefore the existing technology provides very vast information which is not
only from one area/country that means one language but also from many countries
with their own languages. Ramis (2006, 1) says that "still, the language barrier is
the only obstacle for this vast information to be fully shared by all users". Ramis
states that accessing information optimally requires people to be literate in more
than one language. It becomes problems for those who only master one or two
languages. In this context, translation plays an important role to help people to
access and understand the information from other languages. People finally do not
have to master many languages because translation has done the job.
Bell (1997:6) defines translation as the replacement of a representation of
a text in one language by a representation of an equivalent text in a second
paragraphs, or a whole text from a Source Language (SL) to a Target Language
(TL). Along with the developing technology, translation is also influenced by that
development. Currently, translation can be done both manually and automatically.
Manual translation is fully done by human meanwhile automatic translation is
done by computer system which is in practice, with or without human assistance
(translated from Nababan, 1999: 134). The latter which we can call as Machine
Translation (MT) have been the focus of research in translation since 1950s. From
the research, US, Canada, and European countries have developed several systems
of MT. The Systems are among others Météo, Systran, Eurotra, Ariane, and Susy
(Ramis, 2006: 2).
Indeed, the systems are not perfect tools to result satisfying translation
from one language to another language because of several limitation owned by
any kind of machine. In his Ph. D. paper, Gispert reports the Bar-Hillel analysis
that Fully Automatic High-Quality Translation (FAHQT) was an unreachable goal
and that the enthusiasm of the MT research is up and down. It shows scepticism in
viewing the existence of MT in contributing for the development of translation
itself. But looking at the facts of great demand of tools translating texts to help
people facing interlingua condition, the MT is still continuously improved more
and more.
In this study, the researcher focuses on the error analysis of Machine
Translation output because the researcher realizes that it is impossible to analyze
all aspects of Machine Translation whose scope spread from linguistic aspect to
output to repair. Particularly, the researcher chooses Google Translate, the
most-often-used Machine Translation. The researcher is going to investigate what errors
are found in a text translated by Google Translate. It is because the researcher
finds several errors when one text in one language is translated into English by
Google Translate and finds that it will translate a text differently at two different
period of time. Google Translate is one of Google-search-engine features. Google
Translate is a Machine Translation based on Systran system (Research at Google,
2012).
It is explained that Google Translate is statistically based machine
translation developed from Franz Joseph Och research. It can translate one text
into more or less fifty languages; one of them is from English to Indonesian and
vice versa. One of obvious examples is the translation of iPad user guide from
English to Bahasa Indonesia. "Keep all your app subscriptions in one convenient
place" is translated to "Jauhkan semua langganan aplikasi Anda di satu tempat
yang nyaman". The original meaning in Source Text shifts when it is translated
using Google Translate. Instead of "Jauhkan", "Keep" should be translated to
"Simpan". As said previously, Machine Translations including Google Translate
still have limitation. In this case, it is obviously shown that the translation is not
equivalent. The Target Text translated by Google Translate gives totally wrong
instruction for Target Readers.
According to the real example above, the researcher is interested to
analyze the errors done by Google Translate in translating an English text into
guide, National Geographic article, and the Ownership Agreement document. By
the result of analysis, the researcher hopes that users can minimize their misuse in
using Google Translate to translate one text and can use that Machine Translation
appropriately.
B. Problem Formulation
There are two problems formulated in this research, namely
1. What errors are found in the three texts of the Ownership Agreement
document, a National Geographic article, and iPad user guide which are
translated into Indonesian using Google Translate?
2. What suggestions are proposed to reduce errors in using Google-Translate?
C. Objectives of the Study
By the problem formulated above, the objectives of this research are to
find and analyze errors in the Bahasa Indonesia texts of the Ownership Agreement
document, National Geography’s article, and iPad user guide which are translated
by Google Translate. And the second is finding suggestions to reduce errors in
using Google-Translate.
D. Benefits of the Study
Theoretically the researcher expects that this research contributes as one of
Machine Translation and that this finding can be secondary data of errors found in
the translation done by Google-Translate.
Practically, the benefit of this study is not to create absolute solution
which can finish all problems since this study only focuses on linguistic analysis
in Machine Translation, but at least it is to suggest some ways which hopefully
can reduce errors done by Google Translate as a tool to produce raw translation.
E. Definition of Terms
In order to have the same perceptions and terminologies used frequently in
this research, the researcher defines the following terms.
1. Machine Translation
The term Machine Translation (MT) is the now traditional and standard
name for computerized systems responsible for the production of translations
from one natural language into another, with or without human assistance
(Hutchin, 1992: 3).
2. Error Analysis in Machine Translation
Error Analysis in machine translation is counting errors of texts translated
by Google Translate with a classification of errors. It is also an index of the
amount of work required to correct 'raw' Machine Translation output to a standard
3. Google Translate
Google Translate is an online machine translation system published by
Google which provides automatic translation for more or less 57 languages in the
world including Indonesian. (Research at Google, 2012).
Google Translate is produced with some approach: the computer is fed
with billions of words of text, both monolingual text in the target language, and
aligned text consisting of examples of human translations between the languages.
Then it applies statistical learning techniques to build a translation model.
7
CHAPTER II
THEORETICAL REVIEW
A.Review of Related Studies
1. Review of Error Analysis of Google Translate Translating Indonesian Text
into German Text
The researcher reviews the study on MT has been done by Iman Santoso.
On his paper entitled “Analisis Kebahasaan Hasil Terjemahan Google-Translate
Teks Bahasa Indonesia Ke Dalam Bahasa Jerman”, he investigated some errors of
Google-Translate in translating Indonesian texts into German texts. In his
research, he applied linguistic approach by using four aspects of linguistics, i.e.
Orthography, Morphology, Syntax, and Semantics. After he classifies the data, it
was shown that Google-Translate had errors in those four aspects. The errors are
simple ones which actually can be translated well by human translators. It
indicates that MT, in this case Google-Translate didn't have sufficient ability yet
to translate the texts equivalently.
According to the data, from the four aspects mentioned the most errors
belong to morphological errors. There are 25 morphological errors of 59 total
errors. An example of orthographical errors is mis-spelling of proper name.
"Gayus" and "Aburizal Bakrie atau Ical" in Indonesian text are translated into
"Gaius" and "iCal Bakrie" in German text. An example of morphological errors is
an Indonesian word "Pendiri" which is translated as "Gründer". The word is
example of syntactical errors is misstucturing of the sentence "Buku komiknya
sendiri akan beredar di AS mulai akhir Desember dengan harga 6,99 dollar AS".
By Google Translate it is translated "Comic Buch selbst wird in den USA
verbreitet werden ab Ende Dezember 2010 mit dem Preis von $ 6,99". The
translation contains errors in passiving, the appropriate translation must be "Das
Comicheft wird dann in den USA ab Dezember-Ende 2010 mit dem Preis von $
6,99 verkauft". In semantic, errors are found in translating the phrase "naik daun"
to be "stieg". "naik daun" is a metaphor which is synonymous with "being
famous". The appropriate translation must be "berühmt werden.
Based on his study Santoso concludes that Google-Translate translated
texts from Indonesian into German word by word by ignoring its context. Users
have to edit the text after translated by Google-Translate in order to get better
result. He also highlighted that the shorter the texts, the better the result.
2. Review of The Translation Analysis of English Imperatives Translated
into Indonesian by Google Translate
The analysis of Google-Translate errors also done by Kemala Meilinda
Putri. She analyzed the errors of Google-Translate in translating English
imperatives into Indonesian.
She finds that Google-Translate did two categories of errors, those are
function word and miss-selection of words with similar meaning
Putri states that Google-Translate cannot identify different markers of
English and Indonesian imperatives. Similar to Santoso, she said that the machine
merely translated the sentence word for word.
In her suggestions, she asked users to concentrate and to read the result
again.
According to the two previous study about error analysis on machine
translation, currently the researcher is going to investigate the errrors on machine
translation and tries to give suggestions to users to reduce the errors. This research
does not only concern to what errors occured, like the previous studies, but also
gives some alternatives to overcome the problem.
B.Review of Related Theories
1. Translation
There are many definitions of translation. Every expert certainly has his/her
own definition based on his/her own perspectives. Bell defined the translation
based on some resources as.
[...] the expression in another language (or target language) of what has been expressed in another, source language, preserving semantic and stylistic equivalences [...] Translation is the replacement of a representation of a text in one language by a representation of an equivalent text in a second language. (1997: 5-6)
An expression in one language should have its
translated-to-target-language expression. The expression then should be equivalent in term of semantic
(meaning) and stylistic (the style of language). It is apparent that the keyword
as similar as possible in case of delivering message in the source language (SL) to
the target language (TL) reader. Bell also adds a quite understandable definition
for “equivalence”.
Texts in different languages can be equivalent in different degrees (fully or partially equivalent, in respect of different levels of presentation (equivalent in respect of context, of semantics, of grammar, of lexis, etc.) and at different ranks (word-for-word, phrase-for-phrase, sentence-for-sentence) (1997: 6).
The equivalence which is varied shows that it is apparent, and has been for
a very long time indeed, that the ideal of total equivalence is impossible to achieve
(Bell, 1997: 6). Therefore there will be no absolute equivalence which can convey
meaning to the TL as similar as in the SL. But at least, it is sufficient to look for
the closest equivalence for each word, phrase, and sentence in the TL.
The following picture is the transformation of a source language text into a
target language text. The process takes place in the memory where it analyzes the
source language text into a universal semantic representation and synthesizes the
semantic representation into the target language text. The following page shows
Picture 2.1. Process of Translation
2. Machine Translation
Hutchin (1992: 1) explains that Machine Translation is a traditional call for
programs which can produce ‘raw’ translations of texts in relatively well-defined
subject domains, which can be revised to give good quality translated texts at. It is
a tool which can be used to translate automatically one text to another language,
with or without human assistance.
There are some system designs for Machine Translation. The first is the
direct translation approach: the Machine Translation system is designed in all
details specifically for one particular pair of languages in one direction. The
second is the interlingua approach, which assumes the possibility of converting
texts to and from “meaning” representations common to more than one language.
Translation is thus in two stages: from the source language to the interlingua, and
from the interlingua into the target language. The third type is the less ambitious
transfer approach. Rather than operating in two stages through a single interlingua
meaning representation, there are three stages involving, usually, syntactic
representations for both source and target texts. The first stage converts texts into Analysis
Synthesis Source
Language Text
Target
Language Text Semantic
intermediate representations in which ambiguities have been resolved irrespective
of any other language. In the second stage these are converted into equivalent
representations of the target language; and in the third stage, the final target texts
are generated.
3. Evaluation of Machine Translation
Hutchin (1992: 161) states that the translations produced by MT systems
are inadequate. The MT systems cannot result good translations which are
equivalent in term of semantic and stylistic. He added that Fully Automatic High
Quality Translation (FAHQT) which can produce good translation is not present
possible thus the translations produced by Machine Translation merely stops in
phase of ‘raw translations’ which still need revising or post-editing.
The way to increase the quality of the translation is by evaluating the MT
systems. Here Hutchin (1992: 162) shows the principal areas in which evaluation
can take place, what aspects should be taken into consideration, and some of the
methods may be employed. The types and stages of evaluation are divided into
five categories.
a. Prototype evaluations : Evaluation will be restricted to the testing of
processes alone, without consideration of potential operational environments.
b. Development evaluations : Assuring that the systems does what it is
claimed to do. Economic viability, such as a commercial product, potential
c. Operational evaluations : assessment of how much and what kind of
human input is required to produce acceptable translations, what technical
facilities are required, how improvements can introduced, etc.
d. Translator evaluations : Finding out all of problems undergone by
Machine Translation from translator’s perspective, e.g. how much work will
be involved in pre-editing, post-editing, or interactive operation, whether
productivity will be increased, how much time and effort is saved, and the
impact on working practices.
e. Recipient evaluations : the final stage of evaluation is concerned
primarily with quality, cost, and speed.
4. Linguistic Evaluation of Raw Output
Common to all the stages is the testing of the linguistic quality of the
output or the quality of raw translations (Hutchin, 1992: 163). There are two types
of the basic computer processes.
a. Glass-box evaluation
- Assessment by those who have access to all the workings of the system.
- Available only to the researchers and developers of prototype systems.
b. Black-box Evaluation
- Assessment by those who can work only with inputs and outputs.
- Potential purchasers and users are restricted to black-box evaluations.
In addition, in examinations of systems, it needs to distinguish further
between overall assessments of 'quality' and more detailed identifications of
a. Quality Assessment
To test the quality of a translation, there are three obvious objects which
can be tested, they are: (1) its fidelity or accuracy, the extent to which the
translated text contains the 'same' information as the original; (2) its intelligibility
or clarity, the case with which a reader can understand the translation; and (3) its
style, the extent to which the translation uses the language appropriate to its
content and intention. (Hutchin, 1992: 163)
However, applying the test on those three points is excessively subjective.
The readers who are asked to assess the text, either assessing how informative the
text was, evaluating isolated sentences, or measuring style, will have their own
judgments. Their judgments possibly will be different in comprehension and
perspective in which the result will be difficult to measure.
b. Error Analysis
In most instances the most useful practical information is obtained from
error counting. It is an index of the amount of work required to correct 'raw'
Machine Translation output to a standard considered acceptable as a translation
(Hutchin, 1992: 164).
Commonly, the error analysis is done by counting each addition or
deletion of a word, each substitution of one word by another, and each instance of
the transposition of words in phrases, and then calculating the percentage of
corrected words (errors) in the whole text. Similar to Quality Assessment,
however, the method cannot be completely objective because revisers differ in
which are dependent on the particular circumstances in which the revision is
taking place. Although it seems no better way to evaluate the output objectively,
the creation of operational evaluations can be a solution to end those problems.
Hutchin (1992: 164) says that what is needed is a classification of errors
by type of linguistic phenomenon and by relative difficulty of correction. He
stated that some lexical errors are easily resolved by simple changes to
dictionaries, while other may have implications for grammatical rules and for a
whole range of vocabulary items. Some grammatical mistakes may be corrected
by simple adjustments to a few lexical entries; others might involve alterations to
the basic design of whole translation modules.
In addition, there are still obvious difficulties although operational
evaluations are created. They are the selection of texts tested, what should be
tested, and how performance should be measured.
5. Koponen’s Classification in Assesing Machine Translation
Koponen (2010: 4-5) on her journal “Assessing Machine Translation
Quality with Error Analysis” proposed errors categories and the way the tested
texts are selected. He divides errors categories into two big classes, i.e. relation
between source and target concepts and relation between concepts.
a. Relation between source and target concepts (Koponen, 2010: 4-5)
i. Omitted concept: ST concept that is not conveyed by the TT.
ii. Added concept: TT concept that is not present in the ST.
iii. Untranslated concept: SL words that appear in TT.
v. Substituted concept: TT concept is not a direct lexical equivalent for ST
concept but can be considered a valid replacement for the context.
vi. Explicitated concept: TT concept explicitly states information left implicit in
ST without adding information.
b. Relations between concepts (Koponen, 2010: 6)
i. Omitted participant: ST relation not conveyed by the TT due to an omitted
head or dependent.
ii. Omitted relation: ST relation not conveyed by the TT due to
morpho-syntactic errors that prevent parsing the relation although both concepts are
present in the TT.
iii. Added participant: TT relation not present in ST introducing an added
concept.
iv. Added relation: TT relation not present in ST arises due to morpho-syntactic
errors.
v. Mistaken participant: Head or dependent of the relation different in ST and
TT, not same entity.
vi. Mistaken relation: relation between two concepts different in ST and TT,
changed role.
vii. Substituted participant: Head or dependent of the relation different in ST and
TT, same entity.
viii. Substituted relation: relation between two concepts different in ST and TT,
Then for selecting the text, Koponen (2010: 2) suggested to choose three
source texts which are different in types. These different genres are selected to
observe whether the machine translation accuracy varies with text type.
6. Semantics
O’Grady (2010: 203) states that there is more to languages than just form.
Sound pattern, morphological structure, or synactic organization are just the form
of utterances which is obviously heard and seen. However, in order for language
to fulfill its communicative function, utterances must also convey a message; they
must have content (O’Grady, 2010: 203). He adds that what is meant by content in
this context is meaning.
O’Grady explains that the meaning is on the relation between words and
sentences (2010: 204-208). In the first relation, words and phrases can enter into a
variety of semantic relations with each other such as synonymy (automobile =
car), antonymy (dark >< light), polysemy (bright shining, intelligent), and
homophony (bank=financial institution and bank=an edge of a river). Then in the
second relation, sentences also have meanings that can be analyzed in terms of
their relation to other meanings such as paraphrase, entailment, and contradiction
Meanwhile for the meaning itself, O’Grady gives four proposals of
approach to meaning. One of them is componential analysis or semantic
decomposition. This approach tries to analyze the meaning of certain types of
speech classes in terms of semantic feature. The example is man and boy could be
grouped together as [+human, +male] while man and woman could be put in a
7. Farrus’ Solutions in Optimizing The Use of Machine Translation
Farrus (2011) stated that to resolve the errors described in the previous
section, several techniques need to be applied according to the idiosyncrasy of the
problem. He classified it into two types:
a. Text Edition
Some of the errors need to be analyzed by processing the text before or
after the translation.
b. Grammatical Category-Based Approach
Grammatical categories have been successfully used in statistical
translation in order to deal with several problems like reordering (Crego and
Marino [2007] as quoted by Farrus [2011]) and automatic error analysis (Propovic
et al [2006] as quoted by Farrús [2011]). The aim is to add the grammatical
category, via a tag, corresponding to the word to translate, so that the statistical
model is capable of distinguishing the word depending on its category and to learn
from context.
C. Theoretical Framework
All these theories previously explained will help the analysis which later is
used to answer the two problem formulations.
The first theory to the fifth theory about the concept of translation and
machine translation, machine translation evaluation, and error categories are
applied to answer the first problem formulation. The theories is used to classify
Next, the theory of Farrus’ Solutions is applied to look for the most
effective method for reducing errors as the answer of the second problem
formulation. The suggestion leads the researcher to look for the most effective
20
CHAPTER III
METHODOLOGY
A. Object of the Study
The object of the study is sentences taken from three informative texts.
The texts are iPad user guide (the first chapter "At Glance"), National Geographic
article (If They Could Only Talk), and the Ownership Agreement document (The
Yacht Ownership Agreement). The researcher believes that the three texts are the
most truthful English text because they are published by offical company or
website, i.e. Apple Inc., National Geographic, and Limcharoen Hughes &
Glanville. Moreover most of them represent texts which are often accessed by
common readers. iPad user guide is instructions of how to use iPad, the National
Geographic article is information about the legend of walking statues in Easter
Island, and the Ownership Agreement document is about a contract letter of Yacht
ownership.
The forms of data are sentences which are vary in their length and
complexity. The Ownership Agreement document and the National Geographic
contain fairly long and complex sentences, meanwhile the iPad’s user guide
contain many short imperative sentences and sentence fragments. Then all of the
sentences are translated into Indonesian using Google Translate and in the end the
researcher has English sentences and Indonesian sentences from the three texts. In
detail the researcher has 309 English sentences and 309 Indonesian sentences
the National Geographic article, and 97 English sentences and 97 Indonesian
sentences from The Ownership Agreement document.
B. Method of the Study
This study uses library research. The method is conducted to gather the
information and theories of machine translation and linguistic aspects in order to
be able to analyze the data. It is also done to find the indicator of errors which will
be used to assess the Bahasa Indonesia translation of the three English texts.
The data in this study are primary data which means that the data are not
taken from the other studies or researchers. The data are collected by the
researcher from the English software user guide, the English magazine, and the
English legal document.
C. Research Procedure
1. Kinds of Data
The data in this research are objective data. They consist of whole
document of the first chapter of iPad user guide, of one article from National
Geographic magazine, and of the Ownership Agreement document.
2. Data Collection
In collecting the data for this research, the researcher took several steps in
order to get complete and sufficient data.
Firstly, the researcher chose three texts. They were selected to represent
ability varied with text type (Koponen, 2010: 2). The three texts were the first
chapter “At Glance” (iPad user guide), If They Could Only Talk (National
Geographic Magazine), and Fractional Yacht Ownership Agreement (legal
document). Then number of sentences from each text were counted and put in the
table. By applying simple purposive sampling, the researcher finally had 77
sentences from software user guide, 34 sentences from the magazine, and 24
sentences from the legal document. After that all sentences were translated using
Google Translate and the translation was also put in the table. The table was
arranged in such a way so that each sentence would be side by side with its
translation. In the table “No.” is number of each sentence in one article, Source
Text (ST) is the original sentence, and Target Text (TT) is the sentence translated
by Google Translate. The table was displayed as the following.
Table 3.1. Data No. 1
No. Source Text (ST) No. Target Text (TT)
184 Keep your calendar current on iPad, or sync it with your Mac OS X or Windows calendar
184 Jauhkan kalender Anda saat ini pada iPad, atau sync dengan Mac OS X atau Windows kalender
3. Population and Sample
The population of data in this research are 309 sentences from the user
guide, 148 sentences from the magazine, and 97 sentences from the legal
document. The researcher did not use all of the data because they were too many.
The researcher took samples by applying the method of simple random sampling
whole data in each text. Thus, the data used for the analysis are 77 sentences from
the user guide document, 37 sentences from the magazine, and 24 sentences from
the legal document.
4. Data Analysis
After the data had been collected the researcher one group of errors
categories, i.e. errors related to individual concepts. The errors category were
based on theory of omission, addition, explicitation, and substitution created by
Koponen (2010: 4-6). Here was the table used in the analysis.
Table 3.2. Errors in each datum
Code Errors related to individual concepts (a)
Omitted concept
(b) Added concept
(c) Untranslated
concept
(d) Mistranslated
concept
(e) Substituted
concept
(f) Explicitated
concept D/S X/√ X/√ X/√ X/√ X/√ X/√
D1/S184 X X √ √ X X
N1. .... N2. ....
The researcher arranged the data in form of table in order to put them
easily. "Code" in each table refered to the text in the table 1. The code consisted
of D (number of text) and S (number of sentences). The sign " X/√" was used to
confirm that there are/there are not error occurences in one sentence: "X" means
no error occurence and "√" means there are errors. The table 2 totaly has six
confirmation columns. In addition, the row below the confirmation row in each
table is for explaining in detail about what errors are found based on error
occurences in the confirmation column. In this row, N1 and N2 were code for
In order to answer the first problem the researcher analyzed what error
occurences were found in each sentence. This process took several steps. The first
was classifying errors of a sentence based on errors classification in the table 2.
Then, the researcher had to confirm the error occurences. The detail information
of what errors found were always put after confirming the error classifications
below the confirmation row. The last is analysing the errors using semantic
approach in which it still involves some linguistic aspects such as morphology,
syntax, and semantic itself. Those three steps are very important to identify errors
in the translation in order to help the finding suggestions.
Finally, in order to answer the second problem the researcher needed to
look at the table 2. The researcher took samples from each classification for the
finding suggestion. The finding suggestion took several steps. Firstly, the
researcher listed all of error types and put them in a table. Then every error was
tested with suggestions based on Farrus' proposed solutions whose results were
later shown in the confirmation columns. What were tested were samples for
every error type which were taken from the texts. After the testing, the researcher
kept trying the finding suggestion for errors which were not matched to Farrus'
proposed solutions. This process was done since this study of linguistic analysis in
technology was very open for every possibility. The table used for the finding
Table 3.3. Mistranslated Concept
No. Code SL TL Description of Errors
Translating in an isolated form
Text Edition
Information
26 37/D1/S1 84
Keep Jauhkan choosing an inappropriate equivalence
√ Type the word and search the appropriate equivalence of “keep”
(keep menyimpan)
010912
Each error category then would have its table as above. By showing like
D. Research Framework
The following was the map of how this research conducted.
ST 1
National Geographic's article
ST 2 Environmental Cooperation document
ST 3
iPad's manual user guide
Translated by GOOGLE TRANSLATE
TT 1
National Geographic's article
TT 2 Environmental Cooperation document
TT 3
iPad's manual user guide
Error Analysis with Koponen’s Errors
Category
Errors Suggestion to reduce errors Linguistic Analysis
27
CHAPTER IV
ANALYSIS
A.Errors in Google Translate
The first problem discussed is the errors of Google Translate when it is
used to translate the three texts: “At Glance” (iPad user guide), “If They Could
Only Talk” (The National Geographic), and “Fractional Yacht Ownership
Agreement” (a legal document). The data analyzed are the whole texts which are
directly copied and pasted in the Google Translate. The data input are not edited
yet in order to observe how Google Translate deal with all of the texts. Thus there
are differences between inputting the whole text and a word/phrase/sentence. The
latter belongs to the suggestion proposed.
The researcher analyzes each Indonesian translation using the categories
explained in the third chapter. The data are shown in form of table attached in the
appendices.
In this part, the researcher discusses the errors in each category in detail.
From the finding, the researcher sees that not all error categories were found in the
three texts. For the discussion the researcher does not use all of the data. Only the
data which represent similar errors are discussed in the following pages.
1. Omitted Concept
a. Omission of one element in the phrases
Text No. ST TT
The given translation “Layar” TL is considered not representing the meaning of
ST’s compound word “Home screen” because one element of the noun
compound is omitted in the TT. It is therefore classified as omitted concept. The
translation of “Home screen” should be “Layar Utama”.
b. Omission of additional word to complete the information
Text No. ST TT
At Glance
26/D1/S132 Viewing in portrait or landscape
melihat dalam potret atau lanskap
The sentence in the ST “Viewing in portrait or landscape” is translated literally
by Google Translate to be “melihat dalam potret atau lanskap”. Semantically
the translation is not equivalent because the given translation “potret atau
lanskap” does not contain semantic properties “direction” like in the ST. If it is
not added additional information, the meaning in the TT will be “seeing
something inside portrait or landscape”. Thus it should be “melihat dalam arah
potret atau lanskap”.
Text No. ST TT
If They Could Only Talk
104/D2/S52 the statues had been created by pre-Inca from Peru
patung-patung telah
diciptakan oleh Pra-Inca dari Peru
In this datum, "Pre-Inca" is the name of tribe. Google Translate gives the
translation "Pra-Inca" which is considered not enough as the equivalence. It
should be added by the words “orang-orang” to give appropriate description
about its semantic properties “ancient American people”. Thus the translation
c. Omission of predicate
Text No. ST TT
At Glance 28/D1/S140 Bluetooth is on Bluetooth
Google Translate fails to translate “Bluetooth is on”. In the TT, the translation is
only “Bluetooth”. There is no equivalence for “is on” in the TT because
syntactically the verb phrase is omitted which automatically reduce the whole
meaning of the clause. Therefore it should be “Bluetooth menyala”.
d. Omission of possesive pronoun
Text No. ST TT
At Glance 47/D1/S208 Your YouTube account YouTube account
Syntactically, “Your” as a determiner which signifies possession fails to be
translated by Google Translate where it totally changes the phrase meaning
because the sense of possesion in the TT’s phrase does not appear. The
translation should be “YouTube account Anda”.
e. Omission of relative pronoun (yang)
Text No. ST TT
At Glance
52/D1/S232 Search the App Store for apps you can purchase or download
Cari App Store untuk aplikasi Anda dapat
membeli atau
men-download
Syntactically, “yang” as the connector of an adjective clause is omitted in the
TT which reduces the meaning of the whole phrase. Looking at the TT’s context,
one of the interpretations is “Search the App Store for apps and then you can
purchase or download”. The modifier of the word “apps” in the TT changes to
should be translated into “Cari App Store untuk aplikasi Anda yang dapat Anda
beli atau download”.
Text No. ST TT
If They Could Only Talk
114/D2/S68 the islanders could no longer build seagoing canoes for fishing
penduduk pulau tidak bisa lagi membangun perahu berlayar di laut untuk menangkap ikan
The expression “perahu berlayar di laut” is a clause which is not similar to a
phrase “Seagoing canoes”. The translation of the word “seagoing” should be
added explicitly to the phrase because it becomes the modifier of the head
“canoes”. The word “yang” should be added as the signifier of adjective clause
as in the ST’s context. Therefore it should be translated as “perahu yang dipakai
berlayar di laut”.
Text No. ST TT
The Yacht Ownership Agreement
170/D3/S44 damages not covered by insurance
kerusakan tidak diasuransikan
In this datum, the word “yang” as the signifier of adjective clause in the TT is
omitted. That omission causes the change of whole meaning in the TT. “not
covered by” which is the post-modifier of “damages” becomes the predicate in
the TT which then breaks the meaning of the whole sentence because the ST’s
expression is a phrase, not a sentence. It should be translated into “kerusakan
f. Omission of determiner
Text No. ST TT
If They
Could Only Talk
77/D2/S12 they watch over this remote island from a remote age
mereka mengawasi pulau terpencil dari usia jauh
In this datum, “this” in the sentence “they watch over this remote island ....” is
omitted. No equivalence is found in the sentence translated. “this” is the
pre-modifier of “remote island” which functions to determine which island referred
in the sentence. Its position is important because it becomes the specifier of the
phrase “remote island”. So it should be translated into “mereka mengawasi
pulau terpencil ini ....”.
g. Omission of punctuation
Text No. ST TT
If They
Could Only Talk
81/D2/S24 Tuki’s question – how did they do it? – has vexed legions of visitors in the past half century.
Tuki yang tanya
bagaimana mereka
melakukannya? – telah jengkel legiun pengunjung dalam setengah abad terakhir.
In this case, the element which is omitted is not a word but its punctuation, i.e.
the hyphen (–). The hyphen in a sentence has important function to indicate that
they are grammatically linked or to indicate word division at the end of a line.
For this sentence, the hyphen functions to indicate that there is a combined
meaning. The omission of the first hyphen in the sentence makes the relation of
each element in the sentence changes. In the given translation by Google
Translate, the verb phrase and its object after the second hyphen have no subject
because the omitted hyphen makes the subject “Tuki’s question” and the noun
next element. That changes shift the meaning mentioned in the ST. The correct
translation should be “Pertanyaan Tuki – bagaimana mereka melakukannya? –
telah membuat bingung ribuan pengunjung dalam setengah abad terakhir.”
h. Omission of Reference
Text No. ST TT
The Yacht Ownership Agreement
198/D3/S84 Both Co-Owners shall equally distribute any cost and expense accruing from the appointment of such arbitrator [...] The Co-Owner acknowledge of the Agreement
The Co-pemilik mengakui Perjanjian
In this datum, "the" as the definite article of Co-Owner is not translated
appropriately since it does not make "pemilik" refering to the previous person
mentioned. That word without any additional word makes it has a new meaning
(a new referent). It should be added by “tersebut” to make it spesific and still
connected with the previous referent. It should be translated into “Pemilik
Bersama tersebut”.
2. Added Concept
a. Addition of relative pronoun
Text No. ST TT
If They Could Only Talk
89/D2/S32 Just three decades ago cars, electricity, and phone services were scarce
Hanya tiga dekade lalu mobil, listrik, dan layanan telepon yang langka
In the datum 89/D2/S32 “yang” appears in the given translation by Google
Translate whereas there is no match for that word in the ST. That addition
“mobil”, “listrik”, and “layanan telepon”. Actually, “were scarce” should
function as the predicate for the whole sentence. Therefore the correct
translation should be “Hanya tiga dekade lalu mobil, listrik, dan layanan
telepon langka”.
b. Addition of relative pronoun
Text No. ST TT
If They
Could Only Talk
107/D2/S56 “The experts can say whatever they want,” says Suri Tuki, 25, José Tuki’s half brother
"Para ahli dapat mengatakan apapun yang mereka
inginkan," kata Suri Tuki, 25, saudara tiri José Tuki itu
In the datum 107/D2/S56 “itu” appears and makes the TT sentence sounds not
effective since its existence is not necessary. Moreover the ST sentence does not
contain that pronoun. The correct translation is enough without “itu”, it becomes
“saudara tiri José Tuki”.
Text No. ST TT
If They Could Only Talk
135/D2/S120 The Moai cannot be dated directly
Para Moai tidak bisa ditanggalkan secara langsung
The addition of “Para” in the given translation changes the semantic property of
“Moai” in the TT. In the ST, “Moai” is the huge sculpture which is inanimate,
but in the TT, the addition of “Para” then gives the word “Moai” semantic
feature of animate because “Para” itself is only used for human and animals or
in other words only for animate things. So, the correct translation should be
3. Untranslated Concept
Text No. ST TT
At Glance 25/D1/S120 Syncing Sync
In this case, “syncing” is translated as “sync”. Although the translation is not
similar, the researcher considers it as untranslated concept because the word is
still in the ST vocabulary. “-ing” which is missing is considered as inflectional
affixes which does not change the type of meaning it denotes.
Untranslated concept is also found for the same word in the sentence
number 48/D1/S208 and 39/D1/S184. Google Translate gives different treatment
because the word “sync” remained translated as “sync”. The correct translation
for “sync” is “mensinkronisasi”.
Text No. ST TT
At Glance 47/D1/S208 YouTube acount YouTube account
“YouTube” is untranslated because it is a name. However “account” should be
translated because there is an equivalent for that word in Indonesian, i.e.
“akun”. Besides the word “akun” has been the translation in the previous and
the next context in the data. Thus it should translated into “akun YouTube”.
Text No. ST TT
The Yacht Ownership Agreement 148/D3/S1 Yacht Yacht
There is no equivalence of “yacht” in the given translation whereas the TL
vocabulary has the appropriate term for that word, i.e. “kapal pesiar”. The other
term “in pari passu” in the sentence “such insurance shall be payable in pari
passu by the Parties hereunder” is still kept as the original. Lexically it can be
Text No. ST TT
The Yacht Ownership Agreement 172/D3/S44 Co-owner Co-owner
In the datum 172/D3/S44 “Co-owner” is still kept as the original. It is possibly
because there is a hyphen which connects relation between affix “co” and the
head “owner”. The connector here makes Google Translate considers that word
as one unity which then it does not translate the word since there is no
equivalence.
4. Mistranslated Concept
a. Mistranslating a noun as a verb
Text No. ST TT
At Glance 1/D1/S4 Multi-Touch display Multi-Touch menampilkan
Here, it can be seen that Google Translate gives the wrong translation for the
word “display”. On the user guide, “Multi-Touch display” is a noun phrase
which is written in isolated form to explain one part on iPad picture. The word
“display” should be translated as “tampilan” since it is a noun or the head of the
noun phrase. It is not the verb “menampilkan” since it will change the phrase to
be a clause. So the correct translation should be “tampilan Multi-Touch. It
shows that Google Translate cannot detect the word categories; when one word
which morphologically has similar form functions as a noun or a verb in a
sentence.
Similar cases are also found in some data. In the datum number 12/D1/S76
“this switch doesn’t mute audio playback”, “mute” is translated into “bisu”
(adjective). According to the sentence context, it should be translated into
the position as verb. Thus the translation becomes “tombol ini tidak membisukan
pemutaran audio”. In datum number 20/D1/S100, the word “on” in the sentence
“Shows that airplane mode is on” is mistranslated as a preposition “di”,
whereas “on” which is after auxiliary verb “is” functions as predicate in that
clause. It is appropriately translated as “menyala” thus the sentence becomes
“Menunjukkan bahwa mode pesawat menyala”.
b. Choosing an inappropriate equivalence
Text No. ST TT
At Glance 2/D1/S12 Headphone jack Headphone dongkrak
Looking at the phrase, it is seen that Google Translate is already able to give the
translation for each word in the TL. Unfortunately, the given translation is not
equivalence to the context. “Headphone” in the ST is accepted as “Headphone”
in the TT, but “dongkrak” as the translation given for “jack” is not correct.
Although “jack” also contains meaning “a device to for lifting heavy object”,
the most appropriate meaning to the context is “a jack socket or something like a
port”. Therefore the correct translation is “port headphone”. It is seen that
Google Translate can give the translation of one word but the word chosen is not
yet appropriate with the context since possibly the choice of the translation is
random without considering the context.
A similar case also occurs in the datum number 7//S48. The translation
“Depan tombol” for “Home button” is incorrect on the choice of meaning and
the order because it does not represent the original meaning in the ST. The ST’s
expression means “a button which is in the front” but the TT’s expression means
Text No. ST TT
At Glance 10/D1/S52 On the Home screen, tap an app to open it
Pada layar Asal, tekan sebuah aplikasi untuk membukanya
In this error, Google Translate gives the opposite translation for one word. It
mistranslates the word with the translation which is totally different in meaning
and then also changes the meaning for the whole sentence. “Tap” is translated
into “Tekan” whereas two actions resulted from those two words are different.
“Tap” is “to hit lightly on something and “Tekan” (push) is “to press
something”. Therefore “Tap” should be translated as “Ketuk” where it will
have an appropriate result when applied in the iPad. If it The same case also
occurs in the datum number 29/D1/S152. The phrase “double tap” in the
sentence “Double-tap to zoom in or out” is still translated into “dua kali
menekan”. Certainly it will give different result compared to the appropriate
instruction “dua kali ketuk”.
Text No. ST TT
At Glance 37/D1/S184 Keep your calendar current on iPad, or sync it with your Mac OS X or Windows calendar
Jauhkan kalender Anda saat ini pada iPad, atau sync dengan Mac OS X
atau Windows
kalender
In this datum, it is seen how Google Translate fatally gives the translation which
is totally not equivalent. In the sentence “Keep your calendar current on iPad
[...]”, the word “Keep” is translated into “Jauhkan” which means “Keep away”
in the SL. The word which contain semantic property “store in regular place” is
better translated as “Menyimpan”. Therefore, the correct translation should be
the datum number 66/D1/S284. “then” in the sentence “then tap On or Off” is
translated into “maka”. It shows that Google Translate is not able to translate
the grammatical meaning of “then” as an adverb (lalu) in the context. The
correct translation should be “lalu” because it is an adverb of time.
Text No. ST TT
If They Could Only Talk
72/D2/S4 He left his one-room
home on the
southwest coast and hiked north across the island to Anakena beach
Dia meninggalkan satu kamar rumahnya di pantai barat daya dan menaikkan utara melintasi pulau untuk Anakena pantai.
Google Translate gives the incorrect translation for the word “hiked” as
“menaikkan”. It should be translated into “mendaki” because the context is
about travelling on the island. The verb “menaikkan” is not appropriate as the
translation in the TT because its meaning is “to lift something or someone for
other people”, not “to climb up a mountain”.
The mistranslation of verbs also found in the datum number 79/D2/20 and
136/D2/S120. “it was settled” in the sentence “After it was settled, it remained
isolated for centuries” and “cannot be dated” in the sentence “The moai cannot
be dated directly” are translated into “diselesaikan” and “tidak dapat
ditanggalkan” whereas they should be “dihuni” since the context is about place
for inhabitant and “tidak dapat ditentukan waktu tepatnya” since it is more
appropriate than “ditanggalkan” which has ambigous meaning in the TT. For
the datum number D2/S120, the meaning of “ditanggalkan” can be interpreted
as “being put off” which is clearly not suitable with the context (deciding the
Text No. ST TT
If They Could Only Talk
100/D2/S48 the tourism chamber ruang pariwisata
Google Translate cannot give the official term of tourism in the TT for the
phrase “the tourism chamber”. Instead of giving the correct one, it translates the
phrase literally into “ruang pariwisata”. The official term in the TT should be
“dinas pariwisata” which is known widely by TT readers.
Text No. ST TT
The Fractional Yacht
Ownership Agreement
153/D3/S12 The parties pihak
In the datum 153/D3/S12 “pihak” as the translation given by Google Translate
is not enough to describes “the parties” in the ST. The ST phrase has a definite
determiner “the” which is to specify the head noun and an inflectional affix “-s”
which makes the head plural. Therefore it should be translated into “para pihak
tersebut”. The word “para” functions as the plural expression and “tersebut”
function as the definite determiner.
Text No. ST TT
The Fractional Yacht
Ownership Agreement
155/D3/S20 Save construed otherwise by the context, the following expressions shall bear the meanings assigned to them hereunder
Simpan ditafsirkan lain oleh konteks, ungkapan berikut akan menanggung
arti yang
ditugaskan kepada mereka dibawah ini
This datum shows that Google Translate fails to give appropriate meaning for a
word which is polysemous. “Save” in the sentence “Save construed otherwise
by the context [...]” is translated into “Simpan” which then functions as the
subject whereas its function should be as the conjunction of the whole sentence.
translated into “Kecuali ditafsirkan lain oleh konteks [...]”. It is also similar to
the number 164/D3/S36 where “insure” in the sentence “The Parties have
mutually agreed to insure the Yacht for a period to be determined by both
Parties” is translated into “memastikan”. The word “insure” is indeed another
term for the word “ensure (memastikan)”, but in this context it is contextually
translated into “mengasuransikan”. Therefore the translation in the sentence
becomes “[...]sepakat untuk mengasuransikan kapal pesiar [...]”.
c. Misordering a phrase
Text No. ST TT
At Glance 3/D1/S16 Volume buttons Volume tombol
Here Google Translate can give the correct translation which is equivalent to
each of a word. However, the order of the words changes the phrase meaning.
The TT’s phrase “Volume tombol” means “the amount of space” whereas the
ST phrase means “a knob control the loudness”. It should be translated into
“tombol volume”. It is seen that Google Translate can translate a phrase
equivalently, but still in an incorrect order.
More examples, “White icon” (27/D1/S140), “Your iOS devices”
(51/D1/S216), and “a new list” (69/D1/S308) are translated into “Putih icon”,
“perangkat Anda iOS”, and “baru daftar”. The given translations are actually
equivalent. They are just incorrect in case of word order. They should be
translated as “icon putih”, “perangkat iOS Anda”, and “daftar baru”. The
Text No. ST TT
At Glance 75/D2/S4 Anakena beach Anakena pantai
In this case, Google Translate actually can give the equivalence for each word
but there is still problem related to a word order. The word translation of
“beach” which functions as the head of the phrase should be placed before the
modifier. So the correct translation should be “pantai Anakena”.
It is similar to the datum number 90/D2/S32, 92/D2/S32, and 109/D2/S60.
The sentence “now Hanga Roa, the only town, buzzes with Internet cafés”,
“Saturday night”, and “moai building” are translated into “Hanga Roa
sekarang [...]”, “malam Sabtu”, and “moai bangunan”. In the first sentence,
“now” is not the modifier of “Hanga Roa”. Grammatically it functions as the
adverb of time, so it should be translated into “sekarang, Hanga Roa [...]”.
Then in the second datum, Google Translate fatally invert the words in the
phrase. What is meant by “Saturday night” is “the night on Saturday”. But the
given translation “malam Sabtu” means “the night on Friday” because if
“malam” is put before the name of the day, the time always refers to the night
on the day before. It should be translated into “Sabtu malam”. And the last
datum is “moai building” which will be correct in term of order if it is translated
into “bangunan moai”.
From those examples, Google Translate can give the correct translation for
Text No. ST TT
If They Could Only Talk
76/D2/S8 Sleepless roosters crowed
Ayam jantan berkokok tanpa tidur
In this datum, “Sleepless” which is the modifier of “rooster” in the ST becomes
the adverb in the TT sentence. The misordering of words changes the sense of
the original ST’s clause in the TT because the function of “sleepless” as noun
modifier shifts to be verb modifier (adverb). It should be translated into “Ayam
jantan yang tidak tidur berkokok”. It is seen that the movement of modifier can
occur in the process of translation by Google Translate.
Text No. ST TT
The Fractional Yacht Ownership Agreement
168/D3/S40 Uninsured damages
Diasuransikan kerusakan
The noun phrase “Uninsured damages” is mistranslated into “Diasuransikan
kerusakan”.“uninsured” is the modifier of “damages” which does not function
as a verb because it will change the phrase to be a clause. It should be translated
into “Kerusakan yang diasuransikan” which is more suitable to the adjective
clause expression owned by the ST’s expression.
Text No. ST TT
The Fractional Yacht Ownership Agreement
148/D3/S1 FRACTIONAL YACHT
OWNERSHIP AGREEMENT
YACHT Fractional KEPEMILIKAN PERJANJIAN
In this case, besides untranslated, the given translation is incorrect in term of
word order. They should be translated into “PERJANJIAN KEPEMILIKAN
KAPAL PESIAR BERSAMA” because in the TL the order of a phrase always
start with the head (PERJANJIAN). Apparently, Google Translate also has
d. Mispositioning modifier as the subject of a clause
Text No. ST TT
At Glance 22/D1/S104 Shows that your carrier’s 4G LTE network (iPad Wi-Fi + 4G) is available
4G Menunjukkan bahwa
operator Anda LTE
jaringan (iPad Wi-Fi + 4G) adalah tersedia
The noun phrase is translated in a very random order. The sentence “Shows that
your carrier’s 4G LTE network (...) is available” are translated into “4G
Menunjukkan bahwa operator Anda LTE jaringan (...) adalah tersedia”.
Google Translate fatally puts “4G” which is a pre-modifiers of “network”
becomes the subject of the whole sentence. It totally changes each function of
each word and also changes the whole meaning. The correct translation should
be “Menunjukkan bahwa jaringan 4G LTE operator Anda tersedia”. That
occurence is also found in the translation of the sentence “manage your cellular
data account” (57/D1/S260). It is translated into “Anda mengelola selular data
account”. “Anda” as the premodifier of cellular data account” changes to be
the subject for the whole sentence which is syntactically not equivalent and
causes the change of the meaning. The correct translation should be “Mengelola
akun data selular Anda“.
e. Mistranslating phrasal verb
Text No. ST TT
At Glance
17/D1/S88 Push the tool straight in until the tray pops out
Mendorong alat lurus di sampai baki muncul keluar
Several phrasal verbs are not successfully translated by Google Translate. One
example is “push in” in the sentence “push the tool straight in ....”. The
something in (preposition of place)”. The correct translation should be
“Mendorong alat lurus”. Apparently Google Translate cannot detect the
existence of a phrasal verb when it is separated by an object. When finds a
phrasal verb in a separated form, it will translate them word by word.
No. ST TT
142/D2/S136 Wearing away the porous tuff
Mengenakan menghilangkan batuan berporosi
The mistranslation of phrasal verb is also found in this datum when Google
Translate “wearing away” into “mengenakan menghilangkan”. That phrasal
verb is translated literally possibly because it is already added by inflectional
affix which makes it more complex to translate. It is enough to translate that
phrase into “menghilangkan batuan berporosi”.
f. Mistranslating preposition
Text No. ST TT
At Glance 31/D1/S156 Save images from websites to your Photo Library
Simpan gambar dari website untuk Perpustakaan Foto Anda
A preposition which has several meaning such as for and to is given the
translation which does not suit to the context. In this case, the sentence “Save
images from websites to your Photo Library” is translated into “Simpan gambar
dari website untuk Perpustakaan Foto Anda”. Google Translate chooses
“untuk” for “to” rather than “ke” because it cannot detect its function. Actually,
the preposition “from” which comes before can be the signifier that “to” should