Advanced English II (BFE19013)

Corpus Linguistics Foundations

Instructor: Miao Zhang 張淼

What is a corpus?

Are these corpus?

  • a dictionary or word list
  • a text archive
  • a quotation collection
  • one book or one article
  • a random internet dump

A corpus is MORE than a collection

  • large enough to be useful
  • systematic and organized
  • machine-readable
  • designed for analysis
  • representative of a language variety or purpose

Why corpora matter

Even expert speakers:

  • do not know everything
  • may be biased by what they notice
  • cannot remember every pattern
  • cannot easily quantify language use

Features of a corpus

What constitutes a GOOD corpus?

If your corpus contains only formal essays, what kinds of language might be missing?

  • Think about a research question you might want to ask about English in general.
  • What kind of corpus would you need to answer it? What characteristics would it have?

Think about speech, conversation, and informal writing.

1. Representativeness

A corpus should represent the language variety it studies.

If the sample is not balanced, the findings may be misleading.

Discussion

Imagine you want to study social media English by non-native speakers.

What would you include in a corpus to make it representative?

2. Finite size

  • a corpus is usually limited in size
  • it has a target or design
  • it stops when the collection goal is reached

Why is this important for corpus?

3. Machine-readable format

  • searchable
  • analyzable by computer
  • suitable for concordancing and statistics

Concordancing (also known as KWIC: Key Word In Context)

Concordancing of the word “basic”

4. Standard reference

A good corpus can be reused by other researchers.

It gives a stable source for comparison.

  • Replicability: other researchers can repeat the study

Which of these is most corpus-like?

  • a dictionary
  • posts on X worldwide within a certain year.
  • Youtube closed captions for a certain year.

Why do we use corpora?

Language intuition

Everyone has some knowledge of their language.

For example, you naturally know “大红灯笼” sounds ok, but “红大灯笼” is weird.

Likewise, you know “我在C座上课” is fine, but “我上课在C座” is odd.

But, intuition is not always reliable.

What can a corpus reveal that intuition alone cannot?

How is “obesity” talked about in media over time?

Is “record” more often used as a noun or a verb in English?

How do Eastern Asians and Westerners use emojis and stickers differently?

What is the difference between “sincere/candid/genuine”?

Why use corpus?

A corpus helps us:

  • test patterns in real data
  • compare language varieties
  • study typical and rare usage
  • support claims with evidence

Some methodological concerns

Empiricism vs. rationalism

Dimension Empiricism Rationalism
Source of data external evidence internal reasoning
Example task frequencies/probability grammaticality judgement
Goal description explanation
Strength objective and testable can access rare possibilities

Quantitative vs. qualitative

Quick comparison

When would you prefer a qualitative method, and when would you prefer a corpus-based quantitative one?

Compare research goals and data types.

Qualitative study

  • aim: depth and context
  • sample: small and purposeful
  • method: interviews, case studies, observation
  • strength: rich detail
  • weakness: limited generalization

Quantitative study

  • aim: patterns and frequency
  • sample: large and systematic
  • method: counts and statistics
  • strength: objective and generalizable
  • weakness: less nuanced context

Corpus study

is quantitative and empirical.

Question

Which do you think is correct:

  1. I’m happier than yesterday.
  2. I’m more happier than yesterday.

Question

Which do you think is correct:

  1. This car needs washed.
  2. This car needs to be washed.

Where is grammar?

Chomsky’s critiques

  • Corpus consists of real language data, which can be affected by many factors. Linguists should concern only ideal language production that stems from competence, not performance.
  • No body can possibly encounter every single possible sentence in a langauge, so corpora are not sufficient to study language.
  • Corpus linguistics can only describe language, but cannot explain it.

Corpus linguistics responds

  • Corpora are now massive and computational tools handle large data.
  • Good corpus design records speaker and text characteristics.
  • Most corpus work tests hypotheses rather than replacing theory.
  • Experimental methods can complement corpus findings.

Corpus does have limitations

  • There is no negative evidence in a corpus.
    • Absence of evidence is not evidence of absence.
  • Corpora show patterns, but they do not explain everything.
  • Findings depend on the corpus and the research question.
  • Corpus research can be complemented by other methods, such as experiments and surveys.

How do researchers work with corpora?

Basic corpus tools

  1. Concordancers
    • show all examples of a word with context
  2. Aligners
    • compare parallel texts (for translation studies)
  3. POS taggers
    • assign grammatical categories
  4. Parsers
    • analyze structure and function

The evolution of corpora

  • 1960s: Brown Corpus
  • 1980s: scanner era
  • 1990s: digitization
  • 2000s-present: internet-scale data

Design challenge

You want to study how social media English changes over time.

Which type of corpus would be most useful: a sample corpus or a monitor corpus?

  • A reminder: a monitor corpus is a corpus that continues to grow over time, while a sample corpus is fixed and collected at one time.

Types of corpora

Sample vs. monitor

  • sample corpus: fixed, collected at one time
  • monitor corpus: continues to grow over time

General vs. specialized

  • general: broad language picture
  • specialized: one genre, group, or domain

Written vs. spoken

  • written: books, articles, essays
  • spoken: interviews, conversations, meetings

Raw vs. annotated

  • raw: unprocessed text
  • annotated: tagged or labeled text

Final reflection

What kind of research question would you ask if you had access to a large corpus?

Write one idea you could investigate.

Popular online corpora

english-corpora.org

The most popular ones:

  • COCA, BNC, COHA, GloWbE, Oxford English Corpus, NOW (News on the Web) Corpus, etc.

You can build your own corpus!

Sketch Engine is a commercial tool for building and analyzing corpora.

Journals of Corpus Linguistics

Corpora, Corpus Pragmatics, Corpus Linguistics and Linguistic Theory, Digital Scholarship in the Humanities, ICAME Journal, International Journal of Corpus Linguistics, International Journal of Learner Corpus Research, Journal of Corpora and Discourse Studies, Applied Corpus Linguistics

Takeaway

Corpus linguistics is a method for studying real language with evidence.

It helps us move from intuition to investigation.

And that is the foundation of corpus-based research.

Advanced English II