Corpus Linguistics

It depends on your research question. For high-frequency phenomena (common words, basic sentence types), tens of thousands of words suffice. For rare phenomena or detailed variation, millions of words may be needed. A good rule of thumb: you should have at least 30-50 occurrences of your target phenomenon for basic statistical analysis. However, even small specialized corpora (thousands of words) are valuable for endangered language documentation.

A concordance is a list of all occurrences of a target word or phrase, typically shown in context (usually one or two lines on either side). The target is centered for easy scanning. Concordances reveal collocations (words that regularly appear near the target), register variation, and syntactic patterns. Modern corpus software generates concordances with frequency counts and sorting options.

Several metrics quantify collocation strength (association). Mutual Information (MI) measures deviation from random co-occurrence; Log-Likelihood Ratio (LLR) is a statistical test; T-score and Delta-P measure association differently. No single metric is universally best; different metrics suit different research questions. Inspect raw frequencies alongside metrics to avoid artifacts where low-frequency collocations show high association by chance.

No. Absence from a corpus shows it is infrequent or absent in that particular collection, not that it is impossible in the language. Sentences may be rare due to pragmatic or discourse constraints, not grammaticality. Combine corpus evidence with acceptability judgments, experimental data, and theoretical reasoning to draw conclusions about grammaticality.