Chapter 4

The web

The Phi lexicon has an entry for welisha, color, and it does not stay politely on its own page. It points first to haoni welisha, voice color, the ordinary expression for timbre. The compound registry leads elsewhere. kerou welisha, stone color, is gray. horathe welisha, dawn color, is pink. A learner who went looking for one plain word has acquired three more ideas before managing to close the tab. This is generous behavior from a dictionary, though it can make a short study session rather longer than advertised.

Each expression leaves its method visible. kerou means stone and horathe means dawn. Both stand before welisha under the modifier-first rule from the last chapter, so the source of each hue arrives before the category it colors. Voice works the same way. A suffix for gray or an unrelated root for timbre would hide those relations. Phi uses words the learner may meet elsewhere and lets each of them teach again.

The feeling of a sound

In 1929 Edward Sapir gave five hundred people, most of them school students, pairs of invented words. He told them that both words meant the same kind of thing, a table, for example, but that one named a larger instance and one a smaller. Across four pairs contrasting a with i, large majorities chose the a form for the larger thing. In his best-known example, mal sounded like the large table and mil like the small one (Sapir 1929). The effect did not tell anyone what either word meant. It gave them a preference once the field had been narrowed for them.

Newer and broader tests find other modest correspondences. In an online study, 917 people in twenty-five language groups heard bouba and kiki and matched them with round and spiky shapes. The pooled estimate for making both matches in the expected direction was 72 percent. Seventeen groups were reliably above chance, and the sample covered ten writing systems: a broad tendency, not an exceptionless rule (Ćwiek et al. 2022). Blasi and colleagues took a different route, analyzing 6,447 word lists that represented 4,298 languages. In the broadly attested core vocabulary, n recurred in words for nose and l in words for tongue. The extended lists were far fewer. Among them, high front i was associated with small across 78 lineages and three of the five world regions with enough coverage. The tests balanced languages by genealogy and screened associations across regions (Blasi et al. 2016). Human vocabularies are not wholly arbitrary; certain sounds lean.

They do not lean toward everything. The studies cited here establish limited correspondences for size and shape, along with recurring biases in basic vocabulary. The review by Dingemanse and colleagues also describes imitative forms (Dingemanse et al. 2015). None establishes that a nasal universally means groundedness, or that a fricative belongs by nature to reflection. Those are Phi's associations. The language teaches them openly as a local practice, not as facts discovered inside the human ear.

The entry for muila, earth, declines to invent a symbolic story. Its articulatory note follows the closed lips and nasal voice of m, the separate vowels across the hiatus, and the tongue's final contact for l. That is an account of how to say the word. When Phi has a genuine mnemonic, it says so elsewhere: nuwera, bed, begins with the same nu heard in nulae, sleep. The resemblance may help a learner, but nu is not a hidden word for sleep and cannot be used to coin another member of the family.

Phi's manual places the boundary plainly: "Omitting it is more accurate than inventing a story after the word already exists." Sound symbolism is optional, a proposal about how a word may be heard rather than a required biography for every syllable. It can support memory and make coinage answerable to an aesthetic, but it cannot make the word self-defining. Someone still has to teach the gloss. Once nuwera means bed in the student's mind, its echo of sleep may help it stay there.

Family resemblance

Sound does a second job when words resemble their semantic neighbors. sileta, sun, and silero, star, share their first two syllables: si.le. Only the ending parts them. The likeness teaches a true relation, since the sun is a star, while the different final syllables preserve the ordinary distinction between the star of daylight and the wider class seen in the night sky.

Nothing here is derivation. There is no productive si.le element meaning celestial light, ready to attach to a new ending whenever an astronomer feels energetic. Both are complete words, and their kinship is designed into their sound. A learner can use the resemblance as a hook, but cannot use it to manufacture a third member of the family.

Form can relate to meaning in more than one way. The kinship of sileta and silero is closest to what linguists call relative iconicity: the relation between two forms echoes a relation between their meanings. Systematicity is different: it is a statistical pattern across a group of words whose forms help a learner predict a category or use. One pair cannot establish it. Natural languages mix both with arbitrariness. Experiments suggest that iconic forms can help word learning and communication, systematic patterns can help category learning, and arbitrary forms help keep meanings individually distinct (Dingemanse et al. 2015). Those are broad tendencies, not yet test results for Phi. A vocabulary in which every relative sounded alike would make a family reunion difficult. Everyone would answer when one name was called.

Phi accepts that tension rather than choosing one side. The learning chapter recommends studying in clusters, such as nulae, sleep, with nuwera, bed, and whemura, dream. The grouping is guidance rather than code. These words share enough depth and softness to sit together in memory, but each has its own shape and must be learned on its own terms. A familiar sound can help with the next door. Too much familiarity makes the doors hard to tell apart.

Sound families reduce the loneliness of an arbitrary list, but they create neighbors in the ear. This is the web's first cost. A student may remember that sileta and silero belong together and still reach for the wrong ending. For now, Phi can say that the relationship was made audible on purpose and that the resulting risk is watched.

What composition keeps visible

The color compounds make a different bargain. In kerou welisha, both parts keep their ordinary meanings, and their order shows the analysis: this is the color associated with stone. horathe welisha does the same with dawn and pink. The expressions are longer than dedicated roots would be, but a speaker does not have to store two more unrelated sound shapes.

The saving is real, though smaller than the word "transparent" sometimes suggests. Stones come in more colors than any sensible paint chart would attempt, and dawn has been known to arrive gray. The compound registry fixes the intended reading because the world alone cannot. A newcomer may make a good guess; someone who knows the registry can rely on it. Convention remains present, only now it is attached to an image that explains why the convention was chosen.

That image is part of Phi's philosophy of vocabulary. A dedicated word for pink would be compact and exact once memorized. Dawn color keeps the sky in the utterance. Gray retains stone. Timbre retains voice and color together. It asks the listener to hear sound as something with shade and texture. The expressions do more than save roots: they expose Phi's account of the ideas.

Exposure can also become insistence. A speaker may need pink while discussing a paint sample under electric light, with no dawn in mind at all. The inherited image still comes along. That repeated explanation is the web's second cost: a transparent expression shows its reasoning even when the speaker only wanted the result. Phi accepts that small weight where the relation is judged worth keeping. Where a compound becomes cumbersome, ambiguous, or aesthetically wrong, the word-creation protocol permits a root instead.

The choice is practical. A good compound lets its parts continue working; a good root makes a recurring idea easy to hold whole. Neither is a moral victory. The lexicon needs both, because a language built entirely from explanations would eventually require a packed lunch.

Three syllables and no attic

Every Phi lexicon word ends by its third syllable. The ceiling applies to ordinary vocabulary and to every optional module; a difficult philosophical concept receives no extra syllable merely because it arrived wearing spectacles. Productive personal names follow a separate charter and may reach four, but the dictionary may not. If a proposed word will not fit, its maker must find another form or reconsider the expression.

The available space is not small. Under Phi's phonotactic rules, 376,740 legal three-syllable forms can be built. Even after exact duplicates and the practical close-neighbor gate are removed, well over 300,000 remain. That scale is large enough to settle the fear that Phi will run out of words. It says much less about whether two chosen words will remain distinct across a room, through an accent, or beside a running tap. Combinatorics has excellent hearing under ideal conditions.

For that reason, new words face two different collision checks. The validator ratchets a baseline of existing content pairs at one-character edit distance; any newly created pair fails. A second instrument, audit_phonetic_neighbors.py, treats some substitutions as closer than others. Its weights account for differences in vowel height and rounding, and for shared consonant places and manners of articulation. It ranks the resulting neighbors, gives extra attention to function words and frequent forms, and can generate listening prompts for repeated comparison.

The score cannot rename a word. The repository says this plainly because an algorithmic resemblance is not a listening failure. A suspicious pair goes to recorded listening across separate sessions and different conditions, and even repeated trouble from one speaker remains evidence from one speaker. The instrument identifies where attention is due. Actual ears retain the final vote, which is fortunate, since they are the ones expected to use the language.

The three-syllable ceiling tightens this work. It keeps words compact and makes the visible shape of Phi consistent, but it removes the easy escape of adding another syllable whenever a candidate feels crowded. Some coinages will take several attempts. A lovely four-syllable solution may be lovely somewhere else. The inconvenience is collected once, during coinage, in exchange for shorter words everywhere else.

The price of a new root

Before Phi coins a word, its protocol asks whether existing words can already carry the concept and whether a compound would reveal more than a root. It asks where the concept belongs: in the common vocabulary, in one or more optional modules, or beside the language as source material. It then asks what neighboring ideas need to remain distinct and what legitimate use the proposed definition might burden. Only after the meaning has edges does the maker begin choosing sounds.

The candidate must fit the phonology, survive both neighbor checks, and feel distinct in careful and comfortable speech. Its entry then has to explain its semantic range and grammar. Every content root receives a physical account of its articulation and a place in the relevant semantic domains. Sound symbolism and pillar connections appear only when they earn the space. The result is more work than typing a gloss beside an attractive noise. Attractive noises are plentiful and have never shown much reluctance to enter dictionaries.

The protocol is a quality checklist, not a court hearing. Phi is a personal language, and a concept may be coined because it is useful or simply because its maker loves it. The questions do not demand that a word justify its existence to an imaginary public. They make its price visible. Every root occupies sound space and a place in the learner's memory; it also creates new neighbors. In return it may make a precise idea easy to say, or give a discussion a word to gather around.

Optional modules keep part of that exchange voluntary. A specialist can learn the words needed for medical care, ecological systems, or collective governance without making every beginner carry them. The grammar remains one grammar, and a core paraphrase remains available when module vocabulary would exclude someone. The lexicon can expand without turning every specialist term into a beginner's obligation, and a misplaced term can move when use exposes the mistake.

That is why the lexicon's restrictions come with instruments rather than reverence. The checks leave room to correct a bad sound account, separate a confusing family, or replace a compound whose picture has become a nuisance. Actual use will decide which corrections are needed.

The sun's beginning

Return to the person who opened welisha and found more than color. Suppose the next search is for east. Phi has ample unused forms and could spend one. The protocol first looks for a relation already present, and the lexicon has two pieces waiting: sileta, sun, and thorui, beginning.

Set them in the language's ordinary order and the source comes before the event. The compound registry fixes the directional reading, while the image keeps the act of orientation visible. Rather than reach east through another arbitrary syllable, the speaker turns toward the daily edge where light starts.

The learner had opened the dictionary looking for east. What they found was sileta thorui: the sun's beginning.

References

Blasi, Damián E., Søren Wichmann, Harald Hammarström, Peter F. Stadler, and Morten H. Christiansen. 2016. Sound-meaning association biases evidenced across thousands of languages. Proceedings of the National Academy of Sciences 113(39): 10818-10823. https://doi.org/10.1073/pnas.1605782113.

Ćwiek, Aleksandra, Susanne Fuchs, Christoph Draxler, Eva Liina Asu, Dan Dediu, Katri Hiovain, Shigeto Kawahara, Sofia Koutalidis, Manfred Krifka, Pärtel Lippus, Gary Lupyan, Grace E. Oh, Jing Paul, Caterina Petrone, Rachid Ridouane, Sabine Reiter, Nathalie Schümchen, Ádám Szalontai, Özlem Ünal-Logacev, Jochen Zeller, Marcus Perlman, and Bodo Winter. 2022. The bouba/kiki effect is robust across cultures and writing systems. Philosophical Transactions of the Royal Society B 377(1841): 20200390. https://doi.org/10.1098/rstb.2020.0390.

Dingemanse, Mark, Damián E. Blasi, Gary Lupyan, Morten H. Christiansen, and Padraic Monaghan. 2015. Arbitrariness, iconicity, and systematicity in language. Trends in Cognitive Sciences 19(10): 603-615. https://doi.org/10.1016/j.tics.2015.07.013.

Sapir, Edward. 1929. A study in phonetic symbolism. Journal of Experimental Psychology 12(3): 225-239. https://doi.org/10.1037/h0070931.

Phi sources

The Phi material in this chapter comes from documents/reference/compounds.md, project/development_protocol.md, and documents/design/word_shape_and_external_boundaries.md. Module scope follows documents/modules/README.md. The listening procedure is documented in documents/evaluation/listening_audit.md and implemented by scripts/audit_phonetic_neighbors.py. The sound-symbolism account draws on manual/part2_soul/ch04_philosophy_of_sound_meaning_grammar/02_sound_as_symbol.md, and the word examples use their entries under vocabulary/.