Skip to content

Compare methods

Compare source explanations, strengths, limitations and references for up to four methods.

7 results
Criterion Corpus Linguistics Dialectometry
Overview
  • Corpus Linguistics is the study of language based on large, representative collections of texts (corpora) processed by computer. Pioneered by John Sinclair and others, the method uses statistical analysis, concordancing, and computational tools to examine patterns of actual language use. Corpus linguistics has transformed our understanding of English and other languages, revealing frequency patterns, collocation preferences, and register variation that were previously hidden. It serves theoretical linguistics, applied language teaching, and natural language processing.
  • Dialectometry is a quantitative method for measuring linguistic distances between dialects or languages using objective metrics applied to phonological, lexical, or phonetic data. Pioneered by Jean Seguy in 1973, dialectometry compares word lists, pronunciations, or phonetic transcriptions across speech varieties to calculate similarity scores. The resulting distance matrices and dendrograms reveal patterns of dialect relatedness and geographic or social clustering. This method complements traditional dialectology and contributes to historical linguistics and sociolinguistics.
Use and interpretation
  • Use corpus linguistics to investigate actual language use patterns, to test theoretical hypotheses with empirical data, to understand variation across registers and dialects, or to develop language resources for NLP applications. Corpora are essential for studying high-frequency phenomena, dialect variation, language change, and lexical semantics. They complement experimental and introspective methods.
  • Use dialectometry to objectively measure dialect similarity when you have parallel data from multiple varieties, to test hypotheses about language relationships, or to visualize patterns of variation. It complements traditional qualitative dialectology and can process large amounts of data efficiently. The method assumes comparable word lists and is most transparent with phonetically detailed data.
Strengths
  • Provides empirical evidence for language patterns based on actual use, reducing reliance on introspection and invented sentences.
  • Enables quantitative analysis of frequency, collocation, and variation across large, diverse datasets.
  • Facilitates cross-linguistic and cross-register comparison by using standardized, comparable corpora.
  • Supports discovery of patterns (e.g., semantic preferences, register markers) that would be difficult to predict without data.
  • Provides objective, quantifiable measures of dialect similarity, reducing subjectivity in comparisons.
  • Scales well to large datasets (hundreds of varieties, thousands of words), enabling broad pattern detection.
  • Produces visual outputs (dendrograms, maps) that clearly show clustering and relationships.
  • Supports hypothesis testing and comparison across linguistic levels (phonetic, phonological, lexical).
Limitations
  • Corpora are finite samples; low-frequency phenomena may not be represented adequately, and absence from a corpus does not prove a form is impossible.
  • Annotation (POS tagging, parsing) introduces errors that propagate downstream. Manual annotation is expensive; automatic annotation is imperfect.
  • Results reflect corpus composition (e.g., written corpora may not represent spoken language equally); biases in selection affect findings.
  • Frequency patterns do not directly reveal competence; high frequency may reflect performance factors like discourse conventions rather than grammatical structure.
  • Assumes word lists are truly comparable; cognate identification itself requires linguistic judgment and can introduce errors.
  • Distance metrics (like Levenshtein distance) treat all differences equally; they do not weight phonetically significant versus insignificant differences.
  • Results are sensitive to the choice of metric and the composition of the word list; different lists can yield different conclusions.
  • Cannot distinguish between shared innovation and retention of ancestral features, limiting insights into historical relationships without independent evidence.
Common pitfalls
  • Assuming corpus frequency directly reflects grammatical acceptability. Absent patterns may be pragmatically marked rather than ungrammatical; low frequency may reflect discourse context, not linguistic impossibility.
  • Ignoring annotation errors and treating automatically tagged data as gold standard. Validate annotations manually, especially for critical analyses.
  • Overgeneralizing from one corpus to language broadly. Corpus composition matters; results from the British National Corpus may not generalize to spoken English in other regions.
  • Confusing correlation with causation in collocation patterns. Two words may co-occur frequently due to topic (e.g., 'snow' and 'winter') rather than semantic or grammatical dependency.
  • Assuming smaller distances always indicate closer genetic relationship. Extensive borrowing or convergence (areal effects) can reduce distances between non-related varieties.
  • Treating dendrograms as evolutionary trees without additional historical or comparative evidence. Distance-based clustering is a descriptive tool, not a phylogenetic proof.
  • Using inappropriate distance metrics. Levenshtein distance works for phonetic data; percent-difference suits lexical counts. Choose metrics aligned with your data type.
  • Over-interpreting fine differences in distance values. Small changes in metrics or data can alter dendrograms. Report error ranges and run sensitivity analyses.
Introduced by
  • John Sinclair
  • Jean Seguy
Year
  • 1980
  • 1973