Words with chemical elements: what the dictionaries actually reveal

Every word in our dictionaries can be spelled with chemical element symbols. That is not a discovery — it is a design choice. Carbonat's word lists are pre-filtered curated sets: if a word survived into the dictionary, it already passed the element parser. So the question is not “which dictionary words work?” (all of them do), but rather: given that these 85,911 words already work, what is interesting about how they work?

To answer that, I measured element usage patterns, variation counts, and word length distributions across all seven supported languages. The results are more honest than exciting: most words have exactly one valid layout, a handful of single-letter elements do most of the heavy lifting, and 26 of the 118 elements never show up in the parser's preferred split. But every single element does appear somewhere across alternative variants, and the long tail is genuinely interesting once you know where to look.

This page walks through the data in full. I want to be transparent about what the numbers actually say — and equally transparent about where the data pipeline shapes the results before you ever see them. If you are here for the quick version, the summary box below has the headline numbers. If you want the deep read, keep scrolling.

The global top five are all single-letter symbols: O (Oxygen, 41%), I (Iodine, 41%), N (Nitrogen, 40.1%), S (Sulfur, 35.6%), and C (Carbon, 30.5%). This is not a coincidence: single-letter elements can fill any single-character gap in a word, which makes them the universal connectors of the system. A two-letter symbol like Er needs the next letter to cooperate; O just needs to exist.

The first two-letter symbol to break into the global rankings is Er (Erbium) at 17.4%, which makes sense: “er” is one of the most common letter pairs across European languages — verb endings, agent suffixes, comparatives. After that come Re (Rhenium, 11.9%) and Te (Tellurium, 10.6%) — again reflecting common bigrams rather than any chemical significance. These elements dominate the word data not because they are chemically important, but because their symbols happen to look like normal spelling.

There is an asymmetry worth noting. The 14 single-letter element symbols (B, C, F, H, I, K, N, O, P, S, U, V, W, Y) collectively account for a massive share of all element appearances in the data. But they represent only 12% of the periodic table by count. Meanwhile, the 104 two- and three-letter elements share the remaining slice. The element distribution in words is not a reflection of the periodic table — it is a reflection of orthography.

Every single one of the 118 elements appears in at least some valid word split across the seven dictionaries. But 26 of them never show up in the parser's preferred path — the first variant it returns. They only appear in alternative splits. That makes them invisible in the default view unless a user explicitly selects a different variant.

Why does this happen? The parser tries 1-character symbol prefixes before 2- and 3-character ones. When a word contains “co,” the parser tests C (Carbon) before Co (Cobalt). If C leads to a valid continuation, the parser takes that path and Cobalt never gets chosen as the first result. Cobalt still participates in an alternative layout — it is a valid split — but it is not the default.

The 26 most interesting cases are elements like In (Indium), which appears in 6,961 words across alternative variants, Co (Cobalt) in 5,095, and Si (Silicon) in 4,014. These elements are not rare or exotic — they are just consistently outcompeted by the single-letter element that shares their first character.

Each language dictionary was built independently, so the sizes and compositions differ. Italian has the largest set at 14,802 words, while French has the smallest at 7,510. These numbers are not representative of total vocabulary size in each language — they reflect curation depth and availability of elementizable entries in the source word lists I used.

The more telling metric is average variants per word. Italian leads at 2.28, meaning the typical Italian word in our dictionary has close to two valid element layouts on average. Welsh is at the other end with 1.58. Only two languages — Italian and German — have a median variant count above one (both at 2), meaning more than half their words have at least two valid splits. In every other language, the median is exactly one.

This difference tracks with orthographic structure. Italian and German both favour longer words with dense vowel-consonant alternation, which creates more fork points for the parser. Welsh, despite having the second-largest dictionary, produces fewer variants per word because its consonant clusters (ll, rh, ch) do not map cleanly to two-letter element symbols, leaving fewer ambiguous positions.

The global top five (O, I, N, S, C) dominate everywhere, but their relative ordering shifts in ways that reflect each language's orthographic habits. Welsh pushes Nitrogen to first place and introduces Tungsten and Gold into its top ten. Italian and Spanish elevate Oxygen above all others. German and Dutch lean more heavily on Erbium (Er) and Rhenium (Re) thanks to their “-er” and “-re-” word patterns.

The cards below show the top five elements for each language with their appearance rate. Compare them side-by-side to see how spelling conventions shape which elements the parser reaches for.

A small but satisfying corner of the data: some dictionary words are themselves the names of chemical elements. When “carbon” appears in the English dictionary, the parser does not treat it specially — it just splits it into C + Ar + B + O + N like any other word. The element Carbon does not “recognise” its own name; it is merely one of five symbols that happen to tile the word.

Across all seven languages, I found 15 unique element names present in at least one dictionary. English has the most with 12 matches, while Spanish has none. The pattern is predictable: languages that borrow English scientific terminology directly (like Dutch and English) contain more element names, while languages with their own chemical naming traditions include fewer.

There is a small irony here. “Iron” appears as a word in the English dictionary — split as Ir + O + N — using Iridium, not Iron's own symbol (Fe). “Silicon” is in the word list too, split as S + I + Li + C + O + N — six elements, none of which is Silicon (Si). The parser splits the word successfully, but Si itself only appears in alternative variants because S + I is tested before Si. The words survive; their namesake symbols take the back seat.

The global median variant count is 1. That means half of all dictionary words have exactly one valid element-symbol layout — no alternatives, no choices to make. At the 75th percentile it rises to only 2, and the 95th percentile reaches 4. The combinatorial explosion that people imagine — dozens of alternative tile layouts for every word — is not how the system actually behaves for the vast majority of inputs.

The 99th percentile is 8, and the absolute maximum in the dataset is 66 (the Italian word “inasiniscono”). So yes, extreme branching exists, but it is genuinely extreme. The distribution is not a bell curve — it is a cliff with a very long, very thin tail. The standard deviation of 1.54 confirms the skew: most of the population clusters at the low end, and the outliers pull the average up to 1.83.

This has a direct UX implication. For most users typing a single word into the home page, the app will return exactly one layout. That layout is deterministic and reliable, which is good for product confidence. But if you want multiple layout options to compare — different visual rhythms, different element colors — you need to target the long tail deliberately.

The global median word length is 8 characters, with the 95th percentile at 13 and the 99th at 16. The absolute longest entry is 28 characters — the Dutch word “buitengewonelastenregelingen” — which says something about both Dutch compound-word rules and the tolerance of the element parser for long chains.

German words also trend long, with a median of 9 and a 99th percentile of 19. English words are shorter on average (median 7) but include the famous “supercalifragilistic” at 20 characters. Italian clusters tightly between 7 and 10 characters for the middle 50%, reflecting the regularity of Italian word formation even at longer lengths.

Interestingly, longer words do not automatically produce more variation. A 20-character word can have fewer valid splits than a 10-character word if its letter sequence locks the parser into a single path at each step. The relationship between length and variation is not linear — it depends on the density of ambiguous positions within the word, not the word's total length.

Yes, but they are rare. Across all seven language dictionaries I found 129 unique palindromic words (after normalization, at least three characters long). These words read the same forwards and backwards and still split into valid element symbols — a double constraint that filters out almost everything.

The longest is “reconocer” (9 characters, Spanish). The average palindromic word is just 4.0 characters long — much shorter than the general dictionary median of 8 characters. That compression makes sense: the longer a word gets, the harder it is to maintain mirror symmetry while also landing on valid element boundaries at every position.

Most palindromic words are short and structurally simple — three to five letter loops that happen to tile cleanly. But even the short ones are interesting because they demonstrate the intersection of two unrelated constraints: orthographic symmetry and chemical symbol decomposability. Neither constraint knows about the other, so the words that survive both feel like small accidents of language.