2018-2019
Corpus-based translation studies: aims & issues –
Gastcollege Rudy Loock
Introduction/definitions
What is a corpus?
- Linguistic database
- Generally large quantities of data
- Representative sample (of a genre)
- Machine-readable (online interface/specific software)
o You do not read the full corpus yourself
o If you do not find a correct corpus, you can make one yourself, but it must
be machine-readable
- Sometimes annotated (tagging/parsing)
o You get extra information, e.g. linguistic information your searches will
be more sophisticated
Aim = not to read a corpus, but search a corpus
Technical definition (< corpus linguistics)
A collection of machine-readable authentic texts (including transcripts of
spoken data) which is sampled to be representative of a particular language
or language variety and which may be annotated with various forms of
linguistic information.
- Not just any random selection of electronic texts collected from the web
o It is more sophisticated than that
o Non-specialists also write on the web: be careful with the samples you
choose, some sources are not reliable (quality issue)
- Machine law: garbage in garbage out (GIGO)
o It is easy to put in data, interpret the data etc. but it has to be correct
- Issue: What do I put in my corpus? this is where a lot of people get it wrong
Raw text annotation
- Tag words in your corpus
o Each word receives a kind of identity card (pronoun, adjective, noun,
verb...)
o A series of abbreviations
o Some are extremely specific
PoS-tagging (part of speech) Parsing
- You only see the raw text, but annotation and parsing are there, but hidden
- Annotation also often includes lemmatisation
1
, 2018-2019
o All inflected forms of a word (inflections, but not derivations) are grouped
together (lemma)
o You type one word and you get all its inflected forms
Selection of texts according to a set of criteria
What kind of data do I want? What language do I want? What variety of language
do I want?
- You need to know what you are doing and be very precise, otherwise: garbage
in garbage out this is the step where you need to think the most and this is
often where the problem lies for most people
1. Criteria, e.g. genre, dates, geographical variety, general/specialised
language, native/non-native, original/translated...
2. Standardisation and conversion to files with specific formats you need to
transform the texts with a formula
3. Optional: tagging and/or parsing
a. You have software doing this
b. 90-95% success rate
4. Use of the corpus thanks to a specific software (online interface or
concordancer)
a. Concordancer: you put all the text in here, code it and you can do
searches
Manufactured DIY (Do-It-Yourself) corpora
- English corpora, Sketchengine, Dutch Parallel Corpus
- Specific needs not met by online corpora
- Manual/semi-automatic compilation
- Use of offline concordancers (+ online possibilities)
Corpora & translation studies: two different uses
Translation tools
Prescriptive approach
- Use of information as inspiration for help in the decision-making process
- Not necessarily multi-million-word, manufactured, tagged, “clean” corpora
- Help improve translation quality
- What a translation should be
Research tools
Descriptive approach
- < approach of corpus linguistics
- Very often requires big, clean tagged corpora for results to be significant
- Translated language as variation, no quality issue
- What a translation is
Definition: translation studies
What is translation (process)? (theoretical) What are translations like? (descriptive)
Can we improve translations and translator education? (applied research)
2
Corpus-based translation studies: aims & issues –
Gastcollege Rudy Loock
Introduction/definitions
What is a corpus?
- Linguistic database
- Generally large quantities of data
- Representative sample (of a genre)
- Machine-readable (online interface/specific software)
o You do not read the full corpus yourself
o If you do not find a correct corpus, you can make one yourself, but it must
be machine-readable
- Sometimes annotated (tagging/parsing)
o You get extra information, e.g. linguistic information your searches will
be more sophisticated
Aim = not to read a corpus, but search a corpus
Technical definition (< corpus linguistics)
A collection of machine-readable authentic texts (including transcripts of
spoken data) which is sampled to be representative of a particular language
or language variety and which may be annotated with various forms of
linguistic information.
- Not just any random selection of electronic texts collected from the web
o It is more sophisticated than that
o Non-specialists also write on the web: be careful with the samples you
choose, some sources are not reliable (quality issue)
- Machine law: garbage in garbage out (GIGO)
o It is easy to put in data, interpret the data etc. but it has to be correct
- Issue: What do I put in my corpus? this is where a lot of people get it wrong
Raw text annotation
- Tag words in your corpus
o Each word receives a kind of identity card (pronoun, adjective, noun,
verb...)
o A series of abbreviations
o Some are extremely specific
PoS-tagging (part of speech) Parsing
- You only see the raw text, but annotation and parsing are there, but hidden
- Annotation also often includes lemmatisation
1
, 2018-2019
o All inflected forms of a word (inflections, but not derivations) are grouped
together (lemma)
o You type one word and you get all its inflected forms
Selection of texts according to a set of criteria
What kind of data do I want? What language do I want? What variety of language
do I want?
- You need to know what you are doing and be very precise, otherwise: garbage
in garbage out this is the step where you need to think the most and this is
often where the problem lies for most people
1. Criteria, e.g. genre, dates, geographical variety, general/specialised
language, native/non-native, original/translated...
2. Standardisation and conversion to files with specific formats you need to
transform the texts with a formula
3. Optional: tagging and/or parsing
a. You have software doing this
b. 90-95% success rate
4. Use of the corpus thanks to a specific software (online interface or
concordancer)
a. Concordancer: you put all the text in here, code it and you can do
searches
Manufactured DIY (Do-It-Yourself) corpora
- English corpora, Sketchengine, Dutch Parallel Corpus
- Specific needs not met by online corpora
- Manual/semi-automatic compilation
- Use of offline concordancers (+ online possibilities)
Corpora & translation studies: two different uses
Translation tools
Prescriptive approach
- Use of information as inspiration for help in the decision-making process
- Not necessarily multi-million-word, manufactured, tagged, “clean” corpora
- Help improve translation quality
- What a translation should be
Research tools
Descriptive approach
- < approach of corpus linguistics
- Very often requires big, clean tagged corpora for results to be significant
- Translated language as variation, no quality issue
- What a translation is
Definition: translation studies
What is translation (process)? (theoretical) What are translations like? (descriptive)
Can we improve translations and translator education? (applied research)
2
