Chapter 9The Dialect Sample and the Corpus-Quantification Method

9.1 Dialect coverage

The evidence for Hokkaido Ainu is unevenly distributed. Saru and Chitose have substantial grammatical descriptions and lexical resources, while documentation of other varieties differs in size and kind. Nakagawa identifies Saru as the principal basis of his grammar, with frequent Chitose examples and comparisons across Hokkaido Nakagawa (2024: 16). Refsing's grammar describes Shizunai Refsing (1986); Satō's default variety is Chitose Satō (2008: preface).

The label HK identifies Hokkaido material for which a more specific dialect has not been assigned. It does not establish that a form is shared throughout Hokkaido. Place names and speaker information refine the attribution: a regional label may cover material from several communities, periods, or speakers. The source's own description remains necessary when those differences bear on the grammatical claim.

9.2 Sampling and dialect classification

Hattori and Chiri's 1960 study compared a 200-item basic-vocabulary list from nineteen localities Hattori & Chiri (1960); Nakagawa (2024: 100). That survey is distinct from Hattori's 1964 dictionary, whose ten regional sets include nine surveyed varieties and Torii's Kuril material Hattori (1964); Nakagawa (2024: 99). A classification needs to be read with the dataset on which it is based.

Refsing points out that dialect investigations often depended on only one or two speakers per locality, limiting the precision with which geographical boundaries could be established Refsing (1986: 55). A difference between two recorded speakers may reflect regional variation, but the sample may also leave individual or generational variation unresolved. The competing classifications are discussed in Chapter 159 (Dialect Classification and Dialectometry).

9.3 Sakhalin and Kuril comparison

The labels SA and KU identify Sakhalin and Kuril material. Sakhalin sources include texts and grammatical descriptions; the Kuril record is more fragmentary Janhunen (2022: §2(3)); Tangiku (2022). Comparisons need to preserve that difference in documentation. An apparent gap in a historical vocabulary cannot establish that a construction was absent from the spoken language. The principal structural comparisons appear in Chapter 162 (Sakhalin and Northern Kuril Comparisons).

9.4 Text collections as samples

A corpus assembled from published texts reflects the histories of those collections. Endō describes a strong regional bias in the oral-literature record: Hidaka and Iburi are well represented, while northern and eastern Hokkaido have far fewer surviving materials Endō (2022: §5). A large number of sentences therefore need not imply broad coverage of dialects or speakers.

Narrative performance, conversation, elicited examples, and written compositions also provide different opportunities for a construction to occur. Repeated passages or repeated publication of the same recording can further affect counts. The collections and their provenance are described in Chapter 8 (The Oral-Literature Corpus and Spoken-Language Data). A frequency statement needs to identify the particular material searched, including any exclusions.

9.5 Interpreting corpus figures

A count of a written form is a count of matches under particular search conditions. It may include several meanings or grammatical functions. Spelling variants, word division, and the treatment of person markers affect which occurrences are found. Counts of individual tokens, multiword sequences, and sentences containing a form have different denominators and cannot be compared as if they measured the same thing.

A grammatical frequency claim requires the retrieved examples to be classified for the use under discussion. For comparison between dialects or genres, the relevant subcorpora and their sizes must also be specified. An occurrence rate per thousand sentences expresses the number of occurrences divided by the number of sentences, multiplied by one thousand. It measures that sample; it does not by itself show the proportion of speakers who use the construction.

Figures quoted from a publication belong to that author's dataset and counting method. Counts obtained from an online corpus depend on the version and date searched. A reproducible result therefore needs a source or snapshot, the query and normalization used, the counting unit, and the criteria used to classify matches. A missing search result leaves open whether the form is absent, spelled differently, or outside the indexed material.

9.6 Example provenance

An example's citation identifies the published source in which it can be found. Narrator, locality, and recording information distinguish the speech event from the edition that presents it. Where an example is encountered through a later study, the note identifies that intermediary. These details allow a reader to trace both the expression and its interpretation.

The label constructed identifies an illustration composed for the grammar. Such an example is separate from attested speech and does not supply evidence for a frequency claim. Transcription, segmentation, glossing, and translation can each involve editorial decisions; the conventions are explained in Chapter 10 (Interlinear Glossing, Abbreviations, and Citation Conventions).

9.7 Evidence and generalization

The evidence labels distinguish source disagreements, corpus observations, and hypotheses. A corpus-confirmed claim requires the relevant observation to have been checked; corpus-suggested marks a tentative interpretation. Contested concerns competing analyses, while speculative marks an unverified proposal. New analyses originating in this grammar carry original-needs-review. The full conventions appear in Chapter 10 (Interlinear Glossing, Abbreviations, and Citation Conventions).

9.8 Observations outside published descriptions

A speaker's explanation, a learner's observation, or a discussion of a particular passage can identify a grammatical question that published descriptions leave open. Its value depends on the expression and context it identifies. A claim based on such an observation needs an attribution and supporting examples; a tentative suggestion remains tentative until its evidence has been examined.