📄 English Dialects in San Antonio

Summary

Baird asks whether San Antonio has one Anglo English dialect or several. Earlier dialect-geography work by Sawyer and Atwood described the city as Southern in pronunciation and denied any Spanish influence on English. Baird re-examines this with 1975 sociolinguistic fieldwork. Hamilton recorded spontaneous speech in grocery stores, and students recorded mall-goers reading a list of street names, presented as a traffic survey. Baird then analyzed one name, "Blanco," in detail. He sorted its eight phonetic variants into four groups (Spanish, Standard English, Southern, and a blended Southwest) and cross-tabulated them with informant demographics in SPSS. Pronunciation tracked ethnicity and self-rated Spanish knowledge but not mall location. Monolingual English speakers with no Spanish still used the blended forms. Baird concludes that Anglo monolingual speech in the city holds three dialects: Southern, Southwest and Standardized. Spanish-origin place names carry visible Spanish influence. He argues that understanding a city's language requires fusing linguistic geography (many items, few informants) with sociolinguistics (few items, many informants).

Key points

• Hamilton grocery survey: visits to all 29 Handy Andy markets produced 253 valid tokens from 61 speakers, about 60% Mexican-American. Target words came from Sawyer, e.g. greasy, pin/pen, new. The data already suggested Southern, Southwest and Standardized varieties.
• Street names survey: the list had 24 real downtown streets plus the invented "Quintanilla," chosen after a pilot of 15 fake names. Students interviewed people at seven malls and downtown.
• "Blanco" analysis: 66 audible responses came from three malls with 17, 22 and 27 informants. The eight variants were collapsed into four categories.
• Overall shares: Standardized 46%, Southern 25%, Southwest 25%, Spanish 4%.
• By ethnicity:
  • Anglos and Blacks: Standardized 52.5%, Southern 35%, Southwest 12.5%, Spanish 0%.
  • Mexican-Americans: Southwest 45%, Standardized 29%, Southern 16%, Spanish 10%.
  • Speakers with no Spanish used Southwest forms about as often as Southern (27%).

Interesting ideas

• I1. Planted nonce item: a made-up street name hidden in the middle of a real list showed whether speakers apply a pattern productively rather than recalling a memorized form, and its position mattered for getting responses.
• I2. Variants as source A, source B, or a principled mix: each pronunciation was labelled by which parts come from each source system, so blending became an explicit, countable category instead of noise.
• I3. Cross-item consistency as a signal: speakers who said "La Villita" with a Spanish -ll- did the same with "Quintanilla," so one item predicted behaviour on another.
• I4. Proxy metadata can mislead: mall location stood in for residency and showed no effect, while asking about residency directly later did produce correlations.
• I5. Intuitive grouping treated as a testable hypothesis: Baird grouped variants by judgment first, then checked the grouping statistically against independent variables.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (learn): seed evaluation sets with invented words and names placed among real ones, as a negative control that separates true generalization from lookup.
• I2 (tree): in UD dialect conversion, tag each annotation decision as convention A, convention B, or hybrid, so mixed treebanks become measurable rather than silently merged.
• I3 (tokens): check loanword and proper-name handling for consistency across related items; disagreement between sibling forms can flag tokenizer or lemma errors.
• I4 (learn): when slicing evaluation by text source or domain, verify that the metadata really measures what is assumed, before concluding a factor has no effect.
• I5 (learn): let LLM-teacher or hand-made categories enter as hypotheses that must earn support from agreement with independent witnesses, rather than being accepted as gold.