📄 Pulmonic Ingressive Speech in Icelandic

Summary

This survey study asks how Icelandic speakers perceive and report pulmonic ingressive speech (talking while breathing in, called innsog in Icelandic), both in Iceland and in North American heritage communities. Earlier work had documented the phenomenon mostly through scattered observation. The authors tested three expectations: that it is mainly a female habit, that it is fading, and that it is practically absent in North American Icelandic.

They ran an online Qualtrics questionnaire of 24 items, mixing closed and open-ended questions. After filtering out incomplete or unusable submissions, 54 respondents remained. Responses showed that people still use ingressive speech. Men and women were said to use it about equally, although respondents reported hearing it a little more from women. Only the North American group described a decline. Ingressive speech was also reported among North American speakers, contrary to the third expectation.

The study adds rare first-hand data on Icelandic to research on a supposed North Atlantic/Baltic ingressive zone. The authors caution that the sample probably leans toward academics and active heritage-group members.

Key points

• Data collection ran from 15 May to 17 July 2022 and gathered 130 responses. Of these, 70 were dropped for being almost empty, and 6 more were dropped because they contained nothing on ingressive speech or showed no tie to Icelandic. The final set was 54.
• The Iceland group had 25 respondents: 23 in Iceland, 2 in Denmark, and 18 of the 25 in Reykjavík. The North America group had 29: 24 in Canada and 5 in the US. Most of the North American group grew up in rural places.
• Only 5 of the 29 North American respondents (17%) still use Icelandic in daily life. The rest only remember hearing it.
• Questions asked about the social functions of innsog, such as acknowledging, keeping the talk going, showing sympathy or surprise. They also asked which items carry it: Já, Jú, Nei, or longer stretches up to whole sentences.
• Open-ended answers were coded with a constant comparison method and stored in an Access database. Age was split into under-50 and over-50 groups, and a survey oversight placed age 50 in both.

Interesting ideas

• I1. Respondents who clearly misunderstood the key term were removed, since their yes/no answers could not be trusted; a comprehension check protects data quality cheaply.
• I2. The survey kept "I use it" separate from "I hear it from others", and the two gave different gender pictures; reported behaviour and perceived behaviour are not the same evidence.
• I3. A single short token like já was treated as multi-functional, with a checklist of discourse roles (acknowledge, keep the flow going, show surprise, dismiss).
• I4. Overlapping age bins at the value 50 forced an awkward workaround, showing how boundary definitions silently distort comparisons.
• I5. Free-text answers were grouped by repeatedly comparing new cases against emerging categories (constant comparison), so the categories grow out of the data.

For Mova

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

• I1 (learn): Before a witness parser or LLM teacher may vote on silver data, it could be required to pass a small probe set, excluding witnesses that misread the target construction.
• I2 (world): Facts in the text world could carry their evidential source (self-report vs. observation of others), so answers can say who claims what instead of merging the two.
• I3 (frames): Backchannel words like "yeah" or "right" could get a multi-label speech-act analysis from context rather than one fixed meaning, which also helps irony and implicature detection.
• I4 (world): Measure and value ranges in compiled knowledge should use explicit half-open intervals, with a check that flags overlapping or gapped bins.
• I5 (learn): Parser errors could be clustered the same way, comparing each new failure to existing groups, to seed self-induced repair rules instead of hand-picking error types.