AAC Within-Category Organization — Vendor Scan, Clinical Grounding, Proposed Taxonomy

Oso Voice research — 2026-08-23. Adam flagged that Oso Voice’s computed overflow fallback for oversized categories — alphabetical splits rendered as folders like “Places A–M” (BrightVoiceApp/src/grid/navigation.ts, the alphaChunks/rangeLabel functions) — presumes literacy and isn’t learnable for pre-literate children. This is the research pass before curating functional subfolders for the app’s 5 remaining uncurated categories: descriptors (59 words), people (51), food (51), feelings (30), places (25) (counts confirmed against BrightVoiceApp/src/assets/symbol-packages/ brightvoice-symbols/manifest.json). This is a different question from the prior mixed-Home research (docs/research/2026-08-09-mixed-home-page-research.md), which covered what sits on the Home grid itself, not how a single oversized category is subdivided once opened.

Confidence labels follow the house convention: [measured] (vendor docs/screenshots/QRGs directly inspected), [vendor] (vendor prose without independent verification), [peer-reviewed], [secondary], [not found].

Headline

No surveyed vendor uses alphabetical folder labels for pre-literate users. Every system that documents its rationale (AssistiveWare, WordPower) subdivides large categories semantically first — the taxonomies below recur closely across independent vendors. Adam’s concern is directly supported by a named vendor source, not merely inferred: AssistiveWare states plainly that “alphabetic organization can help literate adults to find words initially as they learn a new AAC system. But it does not help young AAC users still developing language and literacy” [vendor].
One real nuance the anti-pattern needs to hold: WordPower — the most literacy-forward of the surveyed systems — does use alphabetical ordering, but only as a secondary sort inside an already-semantic bucket (“nouns are subcategorized and alphabetized,” a separate “DESCRIBE A-Z” page for overflow adjectives), never as the label a child must navigate by. Oso Voice’s current fallback is the harder case: the category’s own folder name (“Places A–M”) is the alphabetical unit, which requires knowing which letter a word starts with before choosing a folder — a first-letter/spelling judgment, not a sort convenience.
The recurring subfolder depth is one level, not two. AssistiveWare is explicit: fringe folders never nest more than one subfolder deep [vendor]. This matches Oso Voice’s existing category → subcategory shape (subcategories.json) and rules out deeper hierarchies as an option, consistent with clinical-vision principle 3 (flat, shallow navigation).
Direct experimental evidence is about taxonomic-vs-schematic organization at the whole-system level, not about subfolder taxonomies specifically — no controlled study was found that tests one within-category subdivision scheme against another (e.g., “food by meal” vs “food by type”). The clinical grounding in §03 is real but one level removed from the exact design question; treat it as informing the shape of subfolders (concrete, familiar, event-linked labels; icon+text; clustering) rather than validating any specific taxonomy.
Concrete subfolder name lists were only well-documented for Food (both AssistiveWare Proloquo2Go training materials and general vendor pattern) and People (AssistiveWare’s own named list). Places, Feelings, and Descriptors subfolder names are recurring themes across vendor marketing/support copy (home/community/outdoors; happy/sad/mad/scared; size/color/texture/opposites) but no vendor published a copy-pasteable list the way they did for Food/People — flagged as thin evidence in §02.

01 — Vendor scan

System	How an oversized category presents	Subfolder depth	Labeling/icon convention	Stated rationale
Proloquo / Proloquo2Go (AssistiveWare)	Semantic subfolders inside the fringe folder (e.g., Food → Breakfast) `[vendor]`	Max one level of subfolders nested in any fringe folder `[vendor]`	Folder-shaped buttons; unused right-side cells reserved for personalization (family names, etc.) `[vendor]`	Explicit anti-alphabetical statement (see §04); most frequent words placed on the folder’s first page, less-frequent on later pages `[vendor]`
TouchChat WordPower (PRC-Saltillo)	Category page (e.g., PLACES, nouns under GROUPS) then, when oversized, alphabetized listing on that same category page; very-low-frequency items pushed to a dedicated “…A-Z” overflow page (ACTIONS A-Z, DESCRIBE A-Z) `[measured — WordPower Vocabulary QRG 19784v1.02]`	One level (category page → A-Z overflow page for that same word class)	Color-coded by grammatical class (pink=adjectives, green=verbs, yellow=pronouns, blue=little words); nouns live behind one orange GROUPS button, “subcategorized and alphabetized” `[measured]`	Motor planning (“same word, same position, different page”) is the explicit design principle; alphabetization is presented as a findability aid layered on top of semantic subcategorization, not a replacement for it `[measured]`
LAMP Words for Life / Unity (PRC)	No folder browsing at all for the words it does have wired — pressing a core word (e.g., “eat”) opens that word’s motor-planned page, which is where domain nouns like “food” live `[measured — Unity/LAMP comparison doc]`	One level (word → its page); the system doesn’t expose large flat noun categories the way the others do	Consistent color/position across pages (word-association navigation, not folder navigation)	Motor automaticity — the word itself is the navigation trigger, so within-category organization is folded into which core word a fringe item is filed under `[vendor]`
TD Snap / Core First (Tobii Dynavox)	Category pages accessed via toolbar; Core First’s own manual states most of its 233 word lists display alphabetically, with only a small number grouped by function `[secondary — search-derived, not independently confirmed against the primary PDF, which could not be parsed]`	One level (word list itself, not further subdivided in the base page set)	Persistent toolbar; pink link cells for navigation, consistent with the mixed-Home research’s prior findings	Not independently confirmed for this question — flagged `[not found]` beyond the general motor-planning rationale already documented in the 2026-08-09 research
Grid 3 / Grid for iPad (Smartbox)	Category-based grid sets (Super Core topic jumps); could not locate primary documentation on subfolder structure for oversized fringe categories within the time available	Not established	MFK color coding (same scheme Oso Voice uses) `[measured — from prior research]`	`[not found]` for this specific question
CoughDrop	“Core 28 + categories” and similar boards link core to “categorized folders” and “additional categorized support words”; live board could not be rendered by the fetch tooling to confirm subfolder names `[secondary]`	Not established	Not established	`[not found]`
Avaz AAC	Category page (e.g., Food) with fixed peripheral folders — Describe/Quantity/People/Actions in the bottom row, page-specific fringe folders in the right column — rather than the category being split into many named subfolders; within-page grouping is spatial/contextual (e.g., Feelings page clusters “calm” words together and “angry” words together) `[vendor — company blog]`	Effectively flat pages with spatial clustering, not subfolder trees	Empty space and background shading used to visually “chunk” related words	Explicit motor-memory/consistency rationale: fixed peripheral folder positions “so users don’t have to think about where to go” `[vendor]`
Speak for Yourself	Category folders keyed to a core verb (e.g., food lives under EAT); within a folder, items are placed alphabetically, with a few exceptions `[vendor — company FAQ]`	One level (never more than 2 touches to any word)	Not established	Alphabetical-within-folder is presented as a stable, predictable placement convention rather than a literacy aid — no rationale tying it to pre-literate use was found

Cross-vendor invariants worth carrying into the Oso Voice proposal:

Semantic-first, alphabetical-only-as-a-secondary-sort-at-most. Every vendor that subdivides a large category does so by meaning; alphabetical order, where it appears at all (WordPower, Speak for Yourself), is a within-bucket tiebreaker or a labeled overflow escape hatch for high-literacy users — never the folder’s own name.
One level of subfolder depth, never more. Matches clinical-vision principle 3 and Oso Voice’s existing category → subcategory shape.
Folder region and word region stay visually distinct, reaffirming the 2026-08-09 research’s finding — this generalizes below Home too.
The most literacy-independent systems (LAMP/Unity) sidestep the within-category-folder question entirely by making the word the navigation trigger rather than exposing a browsable folder of nouns. This isn’t directly portable to Oso Voice’s current architecture (it would mean fringe nouns living behind their associated core verb, e.g. all food behind “eat,” which is closer to the verb-jump mechanic already shipped in ADR 008 than to a subfolder scheme) but is worth flagging as a structurally different alternative to subfoldering at all.

02 — Recurring subfolder taxonomies per category

Evidence strength varies sharply by category — noted per row.

Food [vendor, well-documented] — the best-documented category across vendors. AssistiveWare/Proloquo2Go training materials (via Boom Learning partner content) name: Food & Breakfast, Lunch & Dinner, Side Dishes & Ingredients & Candy, Snacks & Drinks, Desserts, Flavors & Condiments, Fruit, Vegetables & Meat, Eating/Cooking Utensils. TouchChat WordPower training content (an “I See Food” themed book) names: breakfast, snacks, fruits, vegetables, meals, desserts, condiments. The recurring shape across both: meal-time buckets (breakfast/meals/snacks) crossed with a fruit/vegetable/protein axis, plus drinks as its own bucket.

People [vendor, one concrete list] — AssistiveWare names Proloquo’s People subfolders directly: Family, Friends, Roles, About me, Identity. No second vendor’s exact list was recoverable, but the underlying pattern (family vs. non-family relations vs. community-helper roles vs. self) is consistent with the general AAC vocabulary-organization literature’s description of a People folder holding “family, friends and other important people,” with community-helper roles (teacher, doctor, firefighter) commonly treated as a distinct “roles” bucket [secondary].

Places [thin] — no vendor’s exact subfolder list was recoverable. The recurring theme across secondary sources is a home / community / outdoors (or travel) split, echoing “places we go” style groupings already present in Oso Voice’s own core-word subcategories.json (places we go: beach, farm, home, hospital, library, park, playground, restaurant, school, store, zoo — note this is the core-word Places subcategory, distinct from the fringe “places” category being curated here).

Feelings [thin, but one useful evidence-grounded axis] — no vendor published a copy-pasteable subfolder list. Two recurring, evidence-adjacent themes: (a) basic-vs-expanded emotion vocabulary (happy/sad/mad/scared/tired as a floor, with practitioner guidance explicitly recommending expansion to frustrated/embarrassed/disappointed/lonely/worried regardless of a child’s apparent language level — presume-competence-aligned) [secondary — AAC practitioner blogs]; (b) organizing by valence family with an intensity gradient inside it (e.g., annoyed → angry → furious) rather than a flat alphabetical or single “emotions” list [secondary]. No controlled study was found comparing valence-clustered feelings folders to any alternative.

Descriptors [thin] — no vendor subfolder list was recoverable. AssistiveWare’s general fringe-folder description notes descriptors folders also carry domain-specific adjectives (e.g., “delicious,” “yucky” filed under Food, not a general Descriptors folder) — implying some vendors avoid a single monolithic Descriptors category altogether by distributing adjectives into the category they modify. Oso Voice’s data model keeps a standalone descriptors category, so this vendor pattern isn’t directly portable; noted as a considered-and-set-aside alternative rather than a recommendation.

03 — Clinical grounding

Taxonomic vs. schematic/event-based organization for pre-literate children. [peer-reviewed] Directly carried over from the 2026-08-09 research and reconfirmed here: Fallon, Light & Achenbach (2003, AAC 19(2)) found 4–5-year-olds free-sort symbols by event schema, not category; Drager et al. (2004, JSLHR 47(5)) and Light et al. (2004) found schematic/context-based organization outperforms taxonomic organization for children this age. A more recent controlled comparison — Alzrayer, Ledbetter-Cho & colleagues-type “schematic vs. taxonomic speech-generating device” work (Education and Training in Autism and Developmental Disabilities, 2018, and a related UT Austin dissertation) — found that of the ASD participants taught multistep navigational requests, three met mastery criterion with the schematic display but not the taxonomic one, and a fourth met criterion with both but generalized only from the schematic condition [peer-reviewed, secondary summary — primary text not independently read]. This is evidence about display type at the system level (schematic vs. taxonomic as competing whole-system organizations), not about how to subdivide an already-taxonomic category — the closest available evidence, not a direct test of Oso Voice’s exact question (see Headline #4).
Design of effective grid displays for children with developmental disabilities. [peer-reviewed — primary text read in full] Light, Wilkinson, Thiessen, Beukelman & Fager (2019, AAC 35(1)) — the same state-of-the-science review already cited in the 2026-08-09 research. Confirmed directly from the primary PDF (not a secondhand summary): the paper’s evidence-based guidelines table recommends clustering symbols by internal color when possible, and spatial cues to cluster symbols when internal color can’t be used; background-color cuing showed no measurable benefit for children on displays under 24-32 symbols and may even distract younger children by drawing fixations to irrelevant same-background distractors. Motor-behavior data (reach latency, a proxy for cognitive load) was significantly worse under a “distributed” (non-clustered) layout than a clustered one, for every population tested. Implication for subfolders: whatever taxonomy is chosen, symbols belonging to the same subfolder should ideally also share a consistent color or spatial block within that subfolder’s grid — the same clustering principle that motivated the mixed-Home folder region applies one level down, inside a category page. This paper does not discuss alphabetical organization at all (confirmed by direct reading, contradicting an earlier automated summary that falsely claimed it did — see note in §04).
Pagination and multi-page navigation cost. [peer-reviewed, secondhand] PrAACtical AAC’s synthesis cites Thistle & Wilkinson (2013) on the working memory demands of navigating multi-page systems — every additional navigation step is a place a child can lose track of the message they were building. This is the standing rationale (already in clinical-vision principle 3) for keeping subfolders to one level and for not proliferating subfolder count just because a category is large.
Four navigation philosophies, useful vocabulary for the SLP consult. [secondary] PrAACtical AAC’s “Finding My Words” frames AAC organization as one of: linguistic/taxonomic (by part of speech or noun category), word-association (Minspeak-style multi-meaning icons), environmental/ event-based (organize by activity/context), or pragmatic (organize by communicative function — “I want,” “something’s wrong”). Oso Voice’s proposed subfolders below are linguistic/taxonomic by construction (they subdivide an already-taxonomic category), which the evidence in this section says is the harder scheme for pre-literate children at the whole-system level — worth naming explicitly for the SLP consult rather than treating the taxonomy work as evidence-neutral.
Icon + text labeling. [vendor/practitioner consensus, not a single controlled study found] Global Symbols’ AAC symbol design guidance and general practitioner sources converge on pairing every symbol/folder with both a pictogram and a text caption — dual-coding rather than icon-only or text-only. This matches Oso Voice’s existing clinical-vision principle 6 (icon + label, never decorative/emoji chrome) and needs no new justification, just confirmation it extends to subfolder cells.
Presume competence on emotion vocabulary breadth. [secondary] Practitioner sources argue for expanded (not just basic) feelings vocabulary regardless of a child’s apparent language level, aligning with clinical-vision principle 10 (presume competence) — relevant to not under-curating the Feelings subfolders to only “happy/sad/mad/scared.”

04 — Anti-pattern evidence: alphabetical organization

This is the concern Adam flagged, and it has a real, directly-quotable vendor source rather than being merely inferred:

“Alphabetic organization can help literate adults to find words initially as they learn a new AAC system. But it does not help young AAC users still developing language and literacy.” — AssistiveWare, How is the vocabulary in Proloquo organized? [vendor] (https://www.assistiveware.com/blog/how-is-the-vocabulary-in-proloquo-organized)

The same source adds a second, distinct reason beyond literacy: alphabetical order does not support motor-memory location learning — a word’s position changes depending on which words happen to alphabetize near it as the vocabulary set changes, which conflicts with the position-stability commitment already codified in clinical-vision principle 2 and ADR 006/008. This is a second, independent argument against Oso Voice’s current alpha-split fallback beyond the literacy point Adam raised: even a literate user’s motor plan for “where is the P word” breaks if adding one word later shifts every chunk’s alphabetical boundary — which is exactly how alphaChunks/rangeLabel in navigation.ts currently computes ranges (deterministic but re-derived from the live word list each time, not frozen positions).

Important correction to the record: an earlier automated fetch summary in this research session claimed the Light, Wilkinson, Thiessen, Beukelman & Fager (2019) state-of-the-science review “explicitly cautions against alphabetical organization.” Direct reading of the primary PDF (all 9 pages covering the children’s-display sections) found no mention of alphabetical organization anywhere in that paper — it is scoped to VSD vs. grid display variables (color, clustering, spatial cues, navigation-bar placement), not folder-labeling schemes. That claim has been discarded; it is flagged here so it is not accidentally reintroduced by a future pass over this document. The AssistiveWare quote above is the real, verified source for this section’s claim — treat it as [vendor]-strength (a named commercial AAC company’s own design-rationale statement), not [peer-reviewed]. No peer-reviewed study directly testing alphabetical-vs-semantic folder labeling with pre-literate AAC users was found in this pass — flagged [not found], consistent with the parallel finding in the 2026-08-09 research that hybrid/label-scheme comparisons are an understudied area of the AAC display-design literature generally.

Nuance that keeps this honest (from §01): WordPower’s real, measured use of alphabetization (nouns “subcategorized and alphabetized,” a dedicated “DESCRIBE A-Z” overflow page) shows alphabetical order isn’t universally rejected in the field — it survives as a secondary, within-bucket sort for users with some literacy, layered under semantic grouping. What no vendor does, and what the AssistiveWare quote specifically targets, is make the alphabetical range the folder’s own name — which is Oso Voice’s current fallback and the one form of alphabetical organization the literature/vendor consensus actually argues against.

05 — PROPOSED starting taxonomy for Oso Voice’s 5 categories

This is a first-pass proposal for the SLP consult and Adam to review and ratify — not a decision. It is built from the real word lists in BrightVoiceApp/src/assets/symbol-packages/brightvoice-symbols/manifest.json (descriptors: 59, people: 51, food: 51, feelings: 30, places: 25 — counts confirmed) and shaped like the existing voicepack-studio/scripts/subcategories.json ("word": "subcategory name" flat map), so it can plug into the same curation pipeline that already produced the core-words/objects/actions subcategories. Every subfolder stays to the one level the vendor scan and clinical-vision principle 3 both support. None of these words currently carry toolbar/jump behavior, so clinical-vision principle 13 (protest/regulation vocabulary never carries navigation) isn’t implicated by this proposal — it’s purely a within-category label change, not a behavior change.

Food (51 words) — mirrors the AssistiveWare/WordPower meal-time × food-type pattern from §02 most closely of the five:

{
  "apple": "fruits", "banana": "fruits", "blueberry": "fruits", "grape": "fruits",
  "orange": "fruits", "peach": "fruits", "pear": "fruits", "strawberry": "fruits",
  "watermelon": "fruits",

  "carrot": "vegetables", "corn": "vegetables", "potato": "vegetables",
  "pumpkin": "vegetables", "salad": "vegetables",

  "cereal": "breakfast foods", "egg": "breakfast foods", "pancake": "breakfast foods",
  "waffle": "breakfast foods", "muffin": "breakfast foods", "bread": "breakfast foods",
  "butter": "breakfast foods", "jam": "breakfast foods", "peanut butter": "breakfast foods",

  "hamburger": "meals", "hot dog": "meals", "pizza": "meals", "taco": "meals",
  "sandwich": "meals", "soup": "meals", "spaghetti": "meals", "pasta": "meals",
  "rice": "meals", "meat": "meals", "french fries": "meals",

  "cookie": "snacks & treats", "cracker": "snacks & treats", "pretzel": "snacks & treats",
  "popcorn": "snacks & treats", "chocolate": "snacks & treats", "candy": "snacks & treats",
  "cake": "snacks & treats", "donut": "snacks & treats", "ice cream": "snacks & treats",
  "yogurt": "snacks & treats", "cheese": "snacks & treats",

  "milk": "drinks", "juice": "drinks", "soda": "drinks", "lemonade": "drinks",
  "tea": "drinks", "coffee": "drinks"
}

People (51 words) — separates real people-nouns from the pronoun/wh-word subset that happens to live in this fringe category today (worth flagging to the SLP: pronoun/question words filed under a fringe “people” category is an odd data-model choice worth a second look independent of this taxonomy work):

{
  "mom": "my family", "dad": "my family", "brother": "my family", "sister": "my family",
  "grandma": "my family", "grandpa": "my family", "aunt": "my family", "uncle": "my family",
  "cousin": "my family", "nephew": "my family", "niece": "my family", "son": "my family",
  "daughter": "my family", "family": "my family", "baby": "my family",

  "friend": "people I know", "classmate": "people I know", "neighbor": "people I know",
  "babysitter": "people I know", "coach": "people I know", "teacher": "people I know",
  "principal": "people I know", "stranger": "people I know",

  "doctor": "people who help", "dentist": "people who help", "nurse": "people who help",
  "vet": "people who help", "firefighter": "people who help", "police officer": "people who help",
  "chef": "people who help", "waiter": "people who help", "driver": "people who help",
  "mail carrier": "people who help",

  "boy": "people words", "girl": "people words", "man": "people words", "woman": "people words",
  "kid": "people words", "adult": "people words", "people": "people words",

  "I": "me & you", "me": "me & you", "mine": "me & you", "you": "me & you",
  "yours": "me & you", "it": "me & you",

  "what": "asking about people", "who": "asking about people",
  "when": "asking about people", "where": "asking about people", "why": "asking about people"
}

Feelings (30 words) — valence-family clustering per §03’s intensity-family finding, with an explicit calm/regulation bucket distinct from “sad”:

{
  "excited": "happy feelings", "hopeful": "happy feelings", "loved": "happy feelings",
  "proud": "happy feelings", "relieved": "happy feelings", "thrilled": "happy feelings",
  "silly": "happy feelings",

  "bored": "sad feelings", "disappointed": "sad feelings", "lonely": "sad feelings",
  "upset": "sad feelings",

  "annoyed": "mad feelings", "cranky": "mad feelings", "frustrated": "mad feelings",
  "grumpy": "mad feelings", "jealous": "mad feelings", "mad": "mad feelings",

  "nervous": "scared feelings", "worried": "scared feelings", "shy": "scared feelings",
  "embarrassed": "scared feelings", "shocked": "scared feelings",

  "calm": "calm feelings", "content": "calm feelings", "safe": "calm feelings",
  "brave": "calm feelings", "sleepy": "calm feelings",

  "surprised": "surprised feelings", "curious": "surprised feelings", "confused": "surprised feelings"
}

Places (25 words) — home/community/outdoors, the recurring §02 theme, kept to 3 subfolders given the category’s small size (matches the granularity of the app’s own existing core-word subcategories, e.g. “greetings” at 6 words):

{
  "backyard": "places at home", "bathroom": "places at home", "bedroom": "places at home",
  "garage": "places at home", "garden": "places at home", "kitchen": "places at home",

  "bakery": "places in town", "church": "places in town", "city": "places in town",
  "gas station": "places in town", "gym": "places in town", "mall": "places in town",
  "movie theater": "places in town", "museum": "places in town", "street": "places in town",
  "airport": "places in town", "train station": "places in town",

  "barn": "places outside", "camp": "places outside", "forest": "places outside",
  "lake": "places outside", "mountain": "places outside", "ocean": "places outside",
  "pool": "places outside", "river": "places outside"
}

Descriptors (59 words) — the weakest-grounded of the five (§02 flagged this as thin); grouped by sense/function rather than any vendor-verified scheme, and includes one imperfect placement (“round” has no natural home and is filed with condition/state words) that the SLP consult should specifically revisit:

{
  "big": "size & amount", "little": "size & amount", "small": "size & amount",
  "tall": "size & amount", "short": "size & amount", "long": "size & amount",
  "wide": "size & amount", "many": "size & amount", "few": "size & amount",
  "empty": "size & amount", "full": "size & amount",

  "blue": "colors", "green": "colors", "red": "colors", "yellow": "colors",

  "soft": "how it feels", "hard": "how it feels", "rough": "how it feels",
  "smooth": "how it feels", "sticky": "how it feels", "fuzzy": "how it feels",
  "wet": "how it feels", "dry": "how it feels", "heavy": "how it feels",

  "sweet": "how it tastes", "sour": "how it tastes", "salty": "how it tastes",
  "spicy": "how it tastes", "yummy": "how it tastes", "gross": "how it tastes",

  "fast": "how it moves & sounds", "slow": "how it moves & sounds",
  "loud": "how it moves & sounds", "quiet": "how it moves & sounds",
  "strong": "how it moves & sounds", "weak": "how it moves & sounds",
  "hot": "how it moves & sounds", "cold": "how it moves & sounds",

  "angry": "how I feel", "happy": "how I feel", "sad": "how I feel",
  "scared": "how I feel", "scary": "how I feel", "tired": "how I feel",
  "sick": "how I feel", "hurt": "how I feel", "funny": "how I feel",

  "good": "condition & change", "bad": "condition & change", "clean": "condition & change",
  "dirty": "condition & change", "broken": "condition & change", "new": "condition & change",
  "old": "condition & change", "same": "condition & change", "different": "condition & change",
  "easy": "condition & change", "finished": "condition & change", "round": "condition & change"
}

Open questions for the SLP consult

Descriptors’ “how I feel” subfolder overlaps semantically with the entire Feelings category (both contain “sad,” “angry,” “scared” in spirit). Worth asking whether Descriptors should keep an emotion-adjacent bucket at all, or route those words toward the Feelings category instead — a data-model question this research doesn’t resolve.
People’s pronoun/wh-word subset (“I,” “you,” “what,” “who”…) filed under a fringe People category is a pre-existing app data-model choice, not something this taxonomy pass should quietly re-litigate by relabeling — flagged for a separate decision, not folded into this proposal.
No vendor evidence was found strong enough to prefer any one of these five taxonomies’ exact bucket boundaries over a reasonable alternative split — treat the specific word-to-subfolder assignments above as a defensible first draft, not a researched optimum.

Primary sources

AssistiveWare, How is the vocabulary in Proloquo organized? assistiveware.com
AssistiveWare, How are fringe words organized in Proloquo? assistiveware.com
PRC-Saltillo, WordPower Vocabulary Quick Reference (19784v1.02, 2022). PDF
Liberator/PRC, Comparison of Unity and LAMP Words for Life. PDF
Speak For Yourself, How is Vocabulary Organized in Speak for Yourself? speakforyourself.org
Avaz Inc., How Vocabulary Layout Shapes Communication: A Peek Inside Avaz AAC. avazapp.com
PrAACtical AAC, Finding My Words: Considerations in Language Displays and Organization. praacticalaac.org
Robin Hurd, Organizing Vocabulary for an AAC System. PDF
Light, Wilkinson, Thiessen, Beukelman & Fager (2019). Designing effective AAC displays for individuals with developmental or acquired disabilities: State of the science and future research directions. AAC 35(1), 1-14. DOI 10.1080/07434618.2018.1558283. (Primary PDF read in full for this research; confirmed no mention of alphabetical organization — see §04 correction.)
Fallon, Light & Achenbach (2003). AAC 19(2), 74-85.
Drager, Light et al. (2004). JSLHR 47(5), 1133-1148.
Light, Drager, McCarthy et al. (2004). AAC 20(2), 63-88.
docs/research/2026-08-09-mixed-home-page-research.md (this doc’s sibling — Home-page composition, not within-category organization).

Confidence ladder

[measured] WordPower QRG (read directly from PDF); Light et al. 2019 (read directly from PDF, full text pp.1-9).
[vendor] AssistiveWare’s alphabetical-organization statement and fringe folder depth rule; Speak for Yourself’s alphabetical-within-folder convention; Avaz’s peripheral-folder consistency rationale.
[peer-reviewed] Taxonomic-vs-schematic developmental findings (Fallon, Drager, Light 2004 — carried over from the 2026-08-09 research, not re-verified against primary text in this pass); grid-display clustering evidence (Light et al. 2019, verified in full this pass).
[secondary] PrAACtical AAC’s navigation-philosophy framework; the schematic-vs-taxonomic SGD comparison study (summary only, primary text not read); feelings-intensity-gradient practitioner guidance; Core First’s claimed mostly-alphabetical word-list default (search-derived, primary PDF unparseable in this pass).
[not found] Any peer-reviewed study directly comparing alphabetical vs. semantic folder labels with pre-literate AAC users; any vendor’s published Places/Feelings/Descriptors subfolder name list as concrete as the Food/People ones; Grid 3/Grid for iPad’s and CoughDrop’s within-category subfolder conventions specifically.