Our Take
Sietsema's path from theoretical linguistics to dictionary work to spelling bee authority shows how deep specialization in language—across a dozen studied languages—creates unreplicable expertise that no generalist can outsource.
Why it matters
At a moment when large language models are trained on word corpora and pronunciation data, the human judgment of someone who understands the phonological, etymological, and historical layers of English matters more, not less. The Scripps Bee depends on exactly this kind of granular linguistic knowledge.
Do this week
Linguists and lexicographers: document your citation practices and phonological decision-making in writing before institutional knowledge walks out the door; this is how you preserve what LLMs cannot yet replicate.
From Poe to the Spelling Bee
Brian Sietsema discovered his life's calling at age nine, reading Edgar Allan Poe's "The Unparalleled Adventure of One Hans Pfaall." A single unfamiliar word—akimbo—stumped his parents, teachers, and school library dictionaries. He didn't find the answer until college. That gap between curiosity and access stuck with him.
Today, Sietsema is the official pronouncer and etymologist for the Scripps National Spelling Bee. He answers questions about word pronunciation, etymology, and historical usage for competitors, judges, and broadcast audiences. The role spans 15+ years and expanded in 2018 to include earlier rounds when the competition was enlarged.
His path to that desk was circuitous. At the University of Michigan, he entered engineering but switched to religion studies, loading his schedule with biblical languages (ancient Hebrew, ancient Greek, modern Hebrew) plus Dutch, Swedish, modern Arabic, and German. A fifth-year detour into linguistics led him to MIT, where he worked under Morris Halle, a leader in generative grammar. Sietsema's dissertation examined how metrical patterns in language predict tone placement in four Bantu languages spoken in Tanzania.
After MIT (1989), he took a one-year visiting position at Michigan in phonology, then landed at Merriam-Webster as a pronunciation editor. There he updated pronunciations as spoken usage shifted (fluoride's vowel stress changed across the century), recorded voice actors reading dictionary entries, and identified new words for inclusion. He introduced the International Phonetic Alphabet into Merriam-Webster's mass-market publications before it became standard in American dictionaries.
While at Merriam-Webster, Sietsema converted to Greek Orthodox Christianity, married fellow linguist Katherine Chapekis (who was hired at the dictionary as a definer and researcher), and eventually attended seminary part-time. He earned a master of divinity degree and was ordained a priest, serving as Father Mark at Holy Trinity Greek Orthodox Church in Lansing, Michigan.
Language Knowledge Doesn't Scale Horizontally
Sietsema's role at Scripps is not a job that generalizes. Spelling bee pronunciation requires not just phonetic precision but historical knowledge. When he confirms how a word is said, he must know why it's said that way, what competing pronunciations exist in different regions, and when pronunciation has shifted over time. This is descriptive linguistics in real time, on broadcast, under pressure.
A language model trained on text corpora and pronunciation databases can emit a plausible sound. It cannot reliably explain the etymology of a word, defend why one pronunciation is preferred over a regional variant, or acknowledge uncertainty the way a human lexicographer does. Sietsema can say: "This word has two accepted pronunciations, both documented in major dictionaries, and here's why the second one has become more common in the last 20 years."
That nuance matters at Scripps because the competition values not just correctness but transparency. Competitors and judges need to trust the pronouncer. A system that cannot explain its reasoning, or that changes its answer based on retraining, would undermine the competition's credibility.
Sietsema's background in metrical phonology, biblical languages, and historical dictionary work created a depth of judgment that is expensive to replicate and impossible to compress into a benchmark.
What Gets Preserved and What Gets Lost
Sietsema's story raises a quiet question about institutional knowledge in language work. He spent four years at MIT working through the abstractions of generative grammar, then 15+ years at Merriam-Webster building intuition about how English actually moves. That dual training—theory plus corpus work—is rare now.
If you work in lexicography, pronunciation, or linguistic annotation for AI training data, document your decision-making explicitly. Write down why you accepted one variant over another. Explain which reference works you consulted and why. Make the human judgment visible. When large language models need ground truth, they need to see how humans arrived at it, not just the output.
For organizations building speech systems or training pronunciation data, hire someone with Sietsema's profile: deep linguistic training plus years spent in descriptive work, listening to actual usage. Vendor-trained phoneticians can help. But they should sit alongside someone who has spent significant time in a dictionary or corpus, observing how real speakers use language across regions and generations.