The entire English sentence corpus runs on 138 unique words — and that is the most interesting thing about element sentences

Vocabulary reuse is totalTokens / uniqueTokens. A reuse of 1.0 means every word in the corpus appears exactly once — maximum diversity, zero repetition. A reuse of 14× means you are cycling through the same small set relentlessly. The metric captures how much the corpus leans on a few workhorses versus spreading load across a broad vocabulary.

High reuse (French at 14.38×) signals that the language has found a handful of elementizable function words and leans on them hard. The words "un" (325 occurrences) and "une" (243 occurrences) together account for roughly 28% of all word uses in the French sentence corpus. That is an astonishing concentration — two words doing more than a quarter of all the work.

Low reuse (Dutch at 1.67×, Welsh at 1.65×) means the corpus is more lexically diverse. But it also means each word is less battle-tested. If you are trying to build a new sentence in Dutch, you have fewer proven building blocks to work with. You are essentially starting from scratch each time, whereas a French speaker can assemble a new sentence by recombining the same proven template.

English sits in a sweet spot: enough reuse (3.49×) that the core function words are well-established, but enough diversity (138 unique words) that the sentences do not all sound identical. It is also worth noting that the English dictionary has 13,337 elementizable words, but the sentences use only 1% of them. The rest sit unused — not because they cannot be elementized, but because they are hard to combine with other elementizable words into grammatically valid sentences.

French has 494 sentence entries — more than five times the Welsh corpus (82) and more than five times the Dutch (89). But it uses only 143 unique words, barely more than English's 138. How?

The answer is structural. French has two elementizable articles ("un" decomposes as U + N; "une" as U + Ne) that are grammatically required before most nouns. The generator discovered that pairing one of these articles with a rotation of elementizable nouns, verbs, and names produces hundreds of valid sentences from the same template. It is efficiency through grammar — the language's own article system makes it easier to produce valid combinations.

Italian tells the opposite story: 97 sentences but 134 unique words, giving a reuse of only 1.88×. Italian sentences tend to be very short (2.6 words on average) and use each word only once or twice across the entire corpus. There is no dominant template — each sentence is more or less unique.

“What does it take to make a sentence?” In this corpus, the answer is not “a noun and a verb.” The answer is “a stable grammatical bridge.” You need a subject that already works, a bridge verb that the corpus has proven safe, and a determiner or short preposition that can carry you into the content word without breaking the line.

That is why English keeps returning to patterns like name verb determiner adjective noun and French keeps returning to name determiner noun verb. The corpus is not merely choosing words. It is discovering the smallest grammatical machines that can survive the element filter.

If I were giving the shortest useful advice, it would be this: start with a name, add a known bridge verb, and finish with an elementizable noun or adjective. The data does not just suggest that pattern. It repeats it over and over.

That is also why many successful sentences feel slightly formal or stylized. The corpus optimizes for survivable structure, not spontaneity. Once you understand the structure, though, you can work with it instead of fighting it.

In English, "is" appears 62 times and "one" appears 58 times. Together, those two words account for roughly 25% of all word uses across 95 sentences. Two words doing a quarter of the work — that shapes everything about how element sentences read.

This creates a stylistic signature. Element sentences in English gravitate toward declarative structures: "X is Y" patterns dominate because "is" is elementizable (I + S), "one" is elementizable (O + Ne), and "this" is elementizable (Th + I + S). You end up with sentences like "agnes is serene" — grammatically fine, but stylistically constrained in a way you can feel even before you know the rule. The sentence could say something surprising, but the vocabulary is not cooperating.

In French, the compression is even more extreme. The article "un" (U + N) appears 325 times. The feminine "une" (U + Ne) appears 243 times. These two words together represent 28% of all French sentence words. Every other word in the language is fighting for the remaining share. When more than a quarter of your word budget is consumed by articles, the creative space for the rest of the sentence is surprisingly narrow.

German shows a different pattern entirely. The pronoun "er" (Er) leads with 56 uses, followed by "es" (Es) at 42 and "hat" (H + At) at 39. German does not have a single overwhelming anchor — it distributes load across several short, common words. That is reflected in its lower vocabulary reuse (2.66×). The German corpus has less repetition but also less predictability.

Then there is the gap between function words and content words. In English, after "is" (62×), "one" (58×), "this" (32×), and "has" (31×), the next words are "that" (21×) and "serious" (12×). Notice the drop: from 62 uses to 12 uses. The top four function words have 183 combined appearances; the next four have only 51. Function words are not just leading — they are running away with the race.

Look at the English top-15 list again. The content words — "serious" (12×), "person" (10×), "crisis" (8×), "book" (8×) — appear far less frequently than the function words. This is partly natural frequency bias (function words are always more common), but it is also a reflection of the constraint: there are more elementizable function words than elementizable content words, because content words tend to be longer and therefore have more opportunities to contain an unmatchable letter sequence.

The practical consequence is that element sentences tend to have strong syntactic skeletons ("is", "has", "one", "this") but limited semantic variety. You can reliably form the frame of a sentence. Filling it with interesting content is the hard part.

"Casino" is a loanword that most European languages absorbed without significant spelling changes — and it happens to decompose cleanly into C + As + I + N + O in the parser's preferred path (with alternative splits like Ca + Si + No also valid). Meanwhile "in" (I + N) is a preposition or adverb that works across Germanic and Romance languages alike. These are not cognates in the linguistic sense; they are orthographic coincidences that survive the element filter independently in each language.

I find this genuinely interesting because it is a kind of convergent evolution. Five different language pipelines, each with their own dictionaries and generators, independently arrived at the same handful of words. The element constraint creates a shared vocabulary that transcends linguistic boundaries — not through translation, but through the accident of compatible letter sequences.

The three-language group is the most diverse. It includes loanwords ("bar", "crash", "fiasco"), shared function words ("no", "un", "was"), proper names ("eric", "laura", "francisco"), a botanical term ("flora"), and a single-letter word ("o" — used as a vocative in Welsh, a conjunction in Spanish, and a vowel-article in Italian).

Each one survived independently in three different language pipelines, which is a low-probability event given how many words fail the element filter entirely. If you estimate that a given word has maybe a 15–20% chance of being elementizable, the probability of the same word passing in three independent languages is already quite low. The words that do make it through tend to be short, common, and orthographically simple — exactly the properties that also make them elementizable.

In the global word corpus, the top five elements are O (41%), I (41%), N (40.1%), S (35.6%), and C (30.5%). Those percentages are computed across 85,911 dictionary words. Single-letter symbols dominate because they can fill any single-character gap in a word.

In the English sentence corpus, I and S are tied at the top (12.7%), followed closely by O (12.6%). The gap between top elements is much smaller than in the word corpus — the sentence distribution is flatter, because a small recycled word set levels out the element frequencies rather than letting the broad statistical tendencies of the dictionary drive them.

The most interesting sentence-specific element is Neon (Ne): it ranks 7th in English sentences at 4.1% but barely registers in the global word corpus. Why? Because "one" (O + Ne) is the second most common English sentence word at 58 occurrences, and it pulls Neon into the sentence distribution almost single-handedly. This is a case where a single high-frequency word reshapes the element profile of an entire corpus.

Similarly, Thorium (Th) at 3.3% rides on "this" (32×) and "that" (21×). In German, Germanium (Ge) appears at 5.2% because "gestern" and the "ge-" prefix are common in the German sentence vocabulary. And in French, Uranium (U) leads the element chart at 9.9% — because "un" (U + N) is the most common word.

The pattern is consistent: in the sentence corpus, element frequency is a downstream effect of word frequency. Whichever few words the generator leans on hardest, those words' constituent elements get amplified proportionally. This is a fundamentally different relationship than the word corpus, where element frequency reflects the statistical properties of tens of thousands of dictionary entries.

If you pick a random English word from the general vocabulary, the chance it is elementizable is high — the word corpus has a 100% elementizable rate because it is pre-filtered. But that is a curated dictionary. In unconstrained natural language, many common words fail the element test: "the" works (Th + E), but "that" works (Th + At), while "what" does not (W + H + At leaves nothing for the remaining characters... actually it might — the point is that many everyday words fail).

Now consider a five-word sentence. Even if each word independently has a 70% chance of being elementizable, the probability that all five succeed is 0.7⁵ = 16.8%. At 80% per word, you get 32.8%. At 90%, you get 59%. The multiplication is merciless — and real natural language has far more words that fail the filter than succeed.

This is why the corpus exists at all. The generator does not pick words randomly; it draws from a pre-verified vocabulary. And even then, the resulting sentences lean heavily on a small proven word set rather than exploring the full elementizable vocabulary. The data confirms this: 138 unique words across 95 sentences, when English alone has a dictionary of 13,337 elementizable words.

That is a utilization rate of roughly 1% of the available vocabulary. The sentences use about one in every hundred elementizable English words. The rest of the dictionary sits unused — not because those words cannot be elementized, but because they are hard to combine with other elementizable words into grammatically valid sentences. The constraint is not on individual words; it is on the combinations.