AccStack catalog

What the published descriptions say about the position of word stress: the stress type (fixed, weight-sensitive, lexical, morphologically conditioned, tone or pitch accent, no word stress, and others), the rule where there is one, the supporting passage quoted with its page, every work consulted, any disagreement between descriptions, and a confidence rating. Click a language to open its entry.

They give the stress pattern for each word length, read left to right, one digit per syllable: 1 is a stressed syllable and 0 an unstressed one. 10 100 1000 means initial stress in words of two, three and four syllables; 01 010 0010 means penultimate stress. Where the source states secondary stresses they are marked in the same strings (see the legend on the catalog page). Each entry also states the rule in words.

Every confidence level is shown by default; the pills above the table narrow the list. To include the other levels (moderately confident, disputed, not confident, not yet rated), select them in the Confidence filter. The count at the top always gives the number of distinct languages matching the current filters.

Earlier releases of the catalog distinguished two sets: the strict set (a single quantity-insensitive rule for the word, stated without exceptions) and the dominant set (Gordon's (2002) broader criterion: the stated pattern is the predominant one). Membership in both is still recorded on each entry and can be selected with the filter "Previous edition". The catalog now codes, for every fixed-stress entry, the kinds of exception its source states, and the filter Fixed stress: exceptions allowed keeps a language in the selection only if every exception its source states is of a kind you allow. Three presets are provided: exceptionless (absolute), no exception or qualification of any kind; exceptionless (core vocabulary), exceptions confined to loanwords, compounds, phrase-level conditioning or secondary stress, or a bare qualification such as "usually"; and dominant pattern (Gordon 2002), any quantity-insensitive positional rule stated as the predominant pattern. The Protocol gives the counts for each.

For every fixed-stress language the entry lists the kinds of exception its source states: a qualification only, loanwords or names, compounds or reduplications, phrase or utterance conditioning, secondary stress only, native lexical items, morphological classes or affixes, clitics or particles, other conditions. Allow the kinds you are prepared to accept; a language stays in the selection if every exception its source states is of an allowed kind, and the count beside each kind is the number of fixed-stress languages, within your other filters, whose source states it. The entry quotes the source's own statement of each exception.

For languages whose descriptions speak of tone or pitch accent, the entry has a block headed "Tone". It reports what the description that treats tone states: the kind of system, how many contrastive tones it distinguishes and what it calls them, the unit that carries tone, whether it mentions downstep or floating tones, how tone relates to stress, and the kind of evidence the statement rests on, followed by the passage itself with its page and, where needed, a translation. Items the description does not state are marked "not stated" rather than guessed. The block reports the descriptions and has been checked once, independently, at the cited page; it is not a tonal analysis of the language, and it does not change the classification of the stress system. The same fields can be filtered in the left rail under "Tone", and the Protocol says how they were read.

Open the entry and look for "Hear the language". Some entries include a short audio clip from an openly licensed collection, with source, license and attribution; many more link to recordings in external archives. The Recordings filter lists them.

Open the entry. "Source passages" quotes the relevant passage verbatim with its citation and page, and notes the form in which the source was consulted (page scan, open full text, publisher's text, search excerpt, or as quoted in another work). "Full bibliography for this language" lists every work known for the language, with a note on how each can be obtained, so that the description itself can be consulted. The Protocol describes how every entry was verified against its source.

open an issue at github.com/accstack/issues or write to comments@accstack.org. Name the language and, if possible, the description and page that settle the matter. Corrections that change a classification quote the source, and the superseded classification remains visible on the entry.

If one description is better supported (more detailed, based on original data, or more recent and aware of the other), the entry follows it and quotes the disagreement in full under "A description dissents". If none can be preferred, the language is classified "sources conflict (unresolved)". Conflicting statements that were verified in their sources are weighed when the confidence level is set. A database coding (StressTyp2, Gordon 2002, WALS) that departs from the description it cites is not a dissenting description: it is shown separately under "Note on earlier databases" and does not lower the confidence level. The "Why" line of the entry says which applies.

The variety its quoted description describes. Where the source names a variety other than the one the Glottolog entry refers to, or where the match between source and language is uncertain, the entry says so under "Which variety". A description that turns out to concern a language already in the catalog is not entered twice.

Usually because no description accessible to us states its stress pattern or prosodic type. The bibliography page covers every language of the frame, including those without a catalog entry, and notes for each whether a stress statement has been found in an accessible source, is known only second-hand, exists in a description we could not obtain, is absent from the accessible descriptions, or whether no descriptive source is known at all. Languages are added as verified statements are found. Two groups are handled differently: sign languages are left out on purpose, because what their descriptions call stress is a prominence of the hands and not of the voice (their sources are still listed in the bibliography), and constructed languages such as Esperanto have entries but stay hidden until the "include artificial languages" box next to the search field is ticked.

The bibliography is drawn from Glottolog 5.1, from the sources cited in the catalog, and from works found during compilation; it certainly still has gaps. open an issue at github.com/accstack/issues or write to comments@accstack.org with the citation and, if one exists, a link to an open-access copy; works that state the stress pattern will be examined and added to the entry.

Large language models (Claude), working under the protocol described on the Protocol page, not linguists. Every entry was coded from a copy of the source and then independently verified in the same copy at the cited page, with disagreements adjudicated; the whole catalog was afterwards re-verified twice more against the sources. No entry has been reviewed by a linguist, and verification means one model run checking another. The Protocol lists the kinds of error this method is known to make.

That an independent check, given the quotation and the classification with the task of refuting them, found the quotation word for word at the cited page of the same copy of the source, agreed that the classification follows from it, and confirmed that the source describes this language and the word rather than the phrase. Every entry in the catalog has been verified in this sense; where the check did not simply confirm the coding, the disputed fields were adjudicated by a further examination of the source.

Quotations are verbatim and were verified word for word against the source at the cited page. They are given in the language of the source; where the source is not in English an English translation is added, marked as ours. Errors in scanned text, wrong page numbers and translation errors remain possible, and reports of them are welcome.

Because no copy of the original description could be obtained. Some entries rest on the description as quoted in StressTyp2, WALS or a secondary work, and some on a search-engine excerpt only; both are marked under "Evidence", and an entry with no verified first-hand statement is rated not confident. Such entries are updated when the original is obtained.

They state the form in which the cited source was consulted: a page scan of the printed work; an open full text; the publisher's text; a search excerpt only, seen without the full page; or the passage as quoted in another work. For a number of entries, mostly those taken over from StressTyp2 and Gordon (2002), this was not recorded.

They describe how a copy of each work can be obtained. Open access: full text freely available online. Restricted access: full text online behind a login or paywall, or in limited (search-only) view. Print, library copy: no digital copy is known, but a research library holds a print copy, obtainable by scan request or interlibrary loan. Print, no copy located: known from a catalog record or citation only; no holding found in the libraries checked. Not located: no copy of any kind has yet been found.

No; the catalog does not redistribute the works it cites. Where an open-access copy exists the bibliography links to it; otherwise the access label indicates how to obtain one through a library. Quotations on the site are short extracts from the cited pages, given with full attribution so that each classification can be checked; the text of StressTyp2 itself is not reproduced.

It began with the fixed-stress languages of StressTyp2 (Goedemans, Heinz & van der Hulst 2015) and the appendix of Gordon (2002), each re-checked against the description the database cites; entries whose database coding does not match that description carry a correction stated in full, listed by the "Earlier databases" filter. WALS chapters 14A–17A were used to identify weight-sensitive languages and as an independent coding for comparison: each entry notes whether the StressTyp2 and WALS codings agree with the source. All other entries were compiled directly from the descriptive literature.

Very confident: several independent first-hand descriptions agree. Fairly confident: one first-hand description, or minor disagreement with a clear resolution. Moderately confident: descriptions disagree; one analysis is better supported. Disputed: specialists disagree in print. Not confident: no first-hand statement has been confirmed; the classification is provisional. Each entry's "Why" line gives the reason for its rating. Entries added since the last rating, and the entries without a Glottolog code, have no level yet and are shown as not yet rated.

"Fairly confident" means that one analysis clearly prevails even though the evidence is thin: a single verified first-hand source with nothing contradicting it; several supporting sources where the only conflicting statements could not themselves be verified; or supporting works for an entry whose own quotation could not be verified first-hand. "Very confident" requires the entry's own verified first-hand source and at least one other independent first-hand work in agreement, with no work examined conflicting.

No. It means that the available descriptions agree and that the quotation was verified in the source. The rating measures the state of the documentary evidence, not the truth about the language: if the descriptions share an error, the catalog inherits it.

Yes. On both the catalog and the bibliography page, "Download this list as CSV" saves the languages currently selected by your filters, with quotations, citations, conflicting statements, database corrections and confidence levels.

The catalog data (codings, rule strings, confidence lines and the compilers' prose) and the bibliography records are released under CC BY 4.0: reuse them with attribution to AccStack. The quoted passages remain their authors' copyright and are quoted with attribution; the material carried from StressTyp2 keeps that database's own terms (CC BY-NC-ND). StressTyp2 (Goedemans, Heinz & van der Hulst 2015) is CC BY-NC-ND and is cited for its codings and references only. Audio clips keep the licenses of their source collections, listed clip by clip. Quotations from descriptions are short extracts reproduced with attribution so that each classification can be checked. For any other reuse, open an issue at github.com/accstack/issues or write to comments@accstack.org or write to comments@accstack.org.

Cite the description the entry quotes, which is the primary evidence, and the catalog with its release date (shown in the page footer), since entries change between releases: Schwartz, Matthew D., Kevin M. Ryan, Matthew K. Gordon, Joe Pater and Larry M. Hyman. 2026. AccStack: Agentically Compiled Catalog of Stress and Tone, release of 13 October 2026. [date accessed] https://accstack.org/catalog/ Each entry links to the language's Glottolog page for its code and classification.

Yes. Secondary stress is coded separately from the primary rule, with its own quotation, work and page: the entry says whether the source states a secondary-stress rule, states that there is none, mentions secondary stress without placing it, mentions it only for compounds or phrases, or does not discuss it, and whether the statement comes from the same work as the primary rule. Where a later work revises the primary rule, the entry names it. Secondary stresses stated by the source also appear in the rule strings, and the filter "Secondary stress in the source" groups the entries by status.

Yes. Besides the works on word stress, each language's list includes the works found in a survey of the literature on its tone system: grammars, phonologies, articles and theses that describe the tones, or that state the language has none. A work whose statement about tone was read is marked "states tone" and shows the passage with its page and, where the source is not in English, a translation; whether the quotation was verified at the cited page is shown beside it. The other works on tone are listed with how a copy can be obtained. The filter "Tone statement found in the bibliography" selects the languages for which such a passage exists.

They are included, classified as "tone or pitch accent": the prosodic type is recorded with the passage that states it, and no stress rule is given unless the source also states one. Languages whose descriptions state that they have no word stress are classified as such. 299 entries rest on a description of the tone system alone (a work read for the bibliography's survey of tone that states the language's tones, or states that it has none, checked at the cited page); their stress type is "not determinable" unless the same work states a stress rule, and like every entry added since the first edition they carry no confidence rating yet, so they appear when "not yet rated" is ticked.

Whenever entries change, a new data release is generated and the catalog and bibliography pages are rebuilt from it; the release date is shown in the page footer. Languages are added as verified stress statements are found in their descriptions.