Negative space — why training every component sharpens each one

A useful intuition guides Mailwoman's corpuscorpusThe BIO-labeled training data used to train Mailwoman's neural classifier. Assembled from real sources (OpenAddresses, National Address Database) and synthetic shards (boundary stress, order variants, negative space). Managed by @mailwoman/corpus.-coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. work: as we add trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. signal for the address components we've been missing (unit, the streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. affixes, and more), the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' gets better at the components we already handle — not just the new ones. The reason is that a sequence labeller learns a tag by its boundaries, so teaching it what a tokentokenOne word or subword in the tokenized input. For the neural classifier, tokens come from SentencePiece (subword units); for the rule classifiers, tokens are whitespace- and punctuation-separated words. isn't is how it learns what the neighbours are. (One caveat that the campaign earned the hard way: this logic applies to tags the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' must learn from context, not to every starvedstarvedA tag with too little training representation to learn — near-zero F1 — because the corpus adapter never emits examples of it (intersection tags sat at 0% until an intersection synthesizer existed). tag — see Two kinds of starved tag at the end.)

This has names at four altitudes:

• Philosophy calls it via negativa — defining a thing by what it is not.
• Art calls it negative spacenegative spaceAddress formats the neural model was never trained on because they're absent from the base training corpus — PO boxes, CEDEX, intersections, units. The parity campaign targets negative space through synthetic shards and corpus augmentation. — you draw the vase by drawing the gap.
• Linguistics is Saussure's whole thesis: a sign has no meaning in isolation, only value through difference. "StreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels." means streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. partly because it isn't "unitunitA subdivision of a building — apartment, suite, floor — that refines a street address. Mailwoman's unit component; a designator plus identifier forms a subpremise." or "localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.."
• Machine learningmachine learning (ML). Building systems that learn patterns from examples instead of following hand-written rules. Mailwoman's neural classifier is trained on millions of labeled addresses rather than programmed with parsing rules. calls it discriminative learning — the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' fits the decision boundary between classes, not a description of each class in isolation.

The catch-all failure it cures

The softmaxsoftmaxThe function that converts a vector of logits into a probability distribution summing to 1, applied after priors and biases are added to the emission logits. at each tokentokenOne word or subword in the tokenized input. For the neural classifier, tokens come from SentencePiece (subword units); for the rule classifiers, tokens are whitespace- and punctuation-separated words. position is a competition: every tag's probability is normalised against all the others, and the mass has to land somewhere. When a real category has no trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. signal, the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' never learns a detector for it — so tokenstokenOne word or subword in the tokenized input. For the neural classifier, tokens come from SentencePiece (subword units); for the rule classifiers, tokens are whitespace- and punctuation-separated words. that belong to that category get dumped into the nearest tag it did learn. That tag becomes a catch-all (or "dumping-ground") labelcomponent tagOne of the 33 labels in Mailwoman's address schema — street, locality, region, postcode, house_number, unit, po_box, country, venue, intersection, and others. Each parsed span carries exactly one component tag..

This is not hypothetical for us. The v0-parity assessment caught it directly:

• 123 Main St Apt 456 Oakland CA → the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' tags Apt 456 as streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels., because unit was never a class it could reach for.
• 1 Main St Pittsburg PA → streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. absorbs the citylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy., because the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' has never been shown a clean example of what legitimately comes after a streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels., so it doesn't know where street ends.

street is acting as a garbage collector for everything on the road line. The moment unit becomes a trained category, those tokenstokenOne word or subword in the tokenized input. For the neural classifier, tokens come from SentencePiece (subword units); for the rule classifiers, tokens are whitespace- and punctuation-separated words. have a home: their probability mass moves off street and onto unit, and street gets sharper as a side effect — we improved a tag by teaching its neighbours.

The caveat

This is usually netneural networkA model made of layers of simple numeric units whose connection strengths (weights) are learned from data. The transformer encoder at Mailwoman's core is a neural network.-positive but not guaranteed monotonic. Adding a class is free when the old labelcomponent tagOne of the 33 labels in Mailwoman's address schema — street, locality, region, postcode, house_number, unit, po_box, country, venue, intersection, and others. Each parsed span carries exactly one component tag. was wrongly swallowing a distinct category (our case). It can backfire when two categories are inherently ambiguous (you trade one confusion for another), or through capacity and labelcomponent tagOne of the 33 labels in Mailwoman's address schema — street, locality, region, postcode, house_number, unit, po_box, country, venue, intersection, and others. Each parsed span carries exactly one component tag.-noise effects — Mailwoman has scar tissue here, where adding a localelocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. interfered with an existing one on some retrains. So negative spacenegative spaceAddress formats the neural model was never trained on because they're absent from the base training corpus — PO boxes, CEDEX, intersections, units. The parity campaign targets negative space through synthetic shards and corpus augmentation. is a strong hypothesis, not a theorem.

Which is exactly why coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. work is gated on measurability first. Before generating corpuscorpusThe BIO-labeled training data used to train Mailwoman's neural classifier. Assembled from real sources (OpenAddresses, National Address Database) and synthetic shards (boundary stress, order variants, negative space). Managed by @mailwoman/corpus. rows for a starvedstarvedA tag with too little training representation to learn — near-zero F1 — because the corpus adapter never emits examples of it (intersection tags sat at 0% until an intersection synthesizer existed). tag, the evalevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. set must carry enough held-out examples of it that its F1 is signal, not noise from a handful of rows (see eval discipline and the val-set stratification work). The coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. evalevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. is the apparatus that lets the numbers say whether negative spacenegative spaceAddress formats the neural model was never trained on because they're absent from the base training corpus — PO boxes, CEDEX, intersections, units. The parity campaign targets negative space through synthetic shards and corpus augmentation. held for our modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' — whether covering unit lifted street/locality precisionprecisionOf the spans the model labeled as a given tag, the fraction it got right. High precision means few false positives. Paired with recall to compute F1., and whether anything regressed — rather than assuming it.

Two kinds of starved tag — and why the model handles both

Negative spacenegative spaceAddress formats the neural model was never trained on because they're absent from the base training corpus — PO boxes, CEDEX, intersections, units. The parity campaign targets negative space through synthetic shards and corpus augmentation. tells you which tags are taxing their neighbours; it doesn't tell you the only way to cover them is more rows of the same shape. The parity campaignparity campaignThe systematic effort to close the gap between Mailwoman's neural parser and the legacy rule-based v0 parser on every address component tag. Tracked through per-tag parity tables and gated by honest-eval probes. surfaced two shapes of starvedstarvedA tag with too little training representation to learn — near-zero F1 — because the corpus adapter never emits examples of it (intersection tags sat at 0% until an intersection synthesizer existed). tag — but the conclusion is that the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' stays the authority for both; the difference is only what kind of help it needs.

• Open-vocabularyvocabularyThe fixed set of tokens a tokenizer can produce. Mailwoman's SentencePiece vocabulary is tens of thousands of subword pieces, with byte fallback for anything outside it., boundary-defined tags — unit, street_prefix, street_suffix, locality. Their members are open-ended and their extent depends on context, so the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' has to learn the boundary. A balanced coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. shardshardA partial output file of the corpus build, written in Parquet format. The training pipeline streams shards row by row. is the whole fix, and this is where the negative-space effect compounds: covering unit (0 → 92%) lifted US street by about three points.
• Closed-vocabularyvocabularyThe fixed set of tokens a tokenizer can produce. Mailwoman's SentencePiece vocabulary is tens of thousands of subword pieces, with byte fallback for anything outside it. tags — country, po_box, cedex. A finite, enumerable surface set in a predictable slot. The tempting shortcut is a deterministic lookup, and an early pass took it. That was wrong, for an instructive reason.

A first attempt trained country on a shardshardA partial output file of the corpus build, written in Parquet format. The training pipeline streams shards row by row. that was almost entirely trailing-countrycountryThe top-level address component (an ISO country). Closed-vocabulary, so it is best handled by a deterministic matcher feeding a proposal rather than a retrained model head. rows, so the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' over-learned the cheap correlation "trailing tokentokenOne word or subword in the tokenized input. For the neural classifier, tokens come from SentencePiece (subword units); for the rule classifiers, tokens are whitespace- and punctuation-separated words. ⇒ countrycountryThe top-level address component (an ISO country). Closed-vocabulary, so it is best handled by a deterministic matcher feeding a proposal rather than a retrained model head." and started promoting cities and regionsregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. to nationhood (49 F1, 23% precisionprecisionOf the spans the model labeled as a given tag, the fraction it got right. High precision means few false positives. Paired with recall to compute F1.). A flat lookup scored 100 — but only on an evalevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. with no homographshomonymyOne surface, many referents: 'Georgia' is a US state in 'Atlanta, Georgia' and a country in 'Tbilisi, Georgia.' Handled by disambiguation, split across two stages — the parser resolves the tag from in-string context; the resolver late-binds the referent with geographic context.. The homographhomonymyOne surface, many referents: 'Georgia' is a US state in 'Atlanta, Georgia' and a country in 'Tbilisi, Georgia.' Handled by disambiguation, split across two stages — the parser resolves the tag from in-string context; the resolver late-binds the referent with geographic context. case ("Atlanta, Georgia" → regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. vs "Tbilisi, Georgia" → countrycountryThe top-level address component (an ISO country). Closed-vocabulary, so it is best handled by a deterministic matcher feeding a proposal rather than a retrained model head.) is exactly where a lookup can't win without a hand-coded guard, and resolving it from context is the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.''s whole reason to exist. So the over-firing was a trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input.- distribution artifact, not a modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' limitation, and "deterministic wins" was a mirage measured on the easy cases.

The corrected rule: cover every starvedstarvedA tag with too little training representation to learn — near-zero F1 — because the corpus adapter never emits examples of it (intersection tags sat at 0% until an intersection synthesizer existed). tag by trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input.; let the lexicon inform the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.', never override it. Open-vocabvocabularyThe fixed set of tokens a tokenizer can produce. Mailwoman's SentencePiece vocabulary is tens of thousands of subword pieces, with byte fallback for anything outside it. and closed-vocabvocabularyThe fixed set of tokens a tokenizer can produce. Mailwoman's SentencePiece vocabulary is tens of thousands of subword pieces, with byte fallback for anything outside it.-without-homographshomonymyOne surface, many referents: 'Georgia' is a US state in 'Atlanta, Georgia' and a country in 'Tbilisi, Georgia.' Handled by disambiguation, split across two stages — the parser resolves the tag from in-string context; the resolver late-binds the referent with geographic context. (po_box, cedex) just need a balanced coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. shardshardA partial output file of the corpus build, written in Parquet format. The training pipeline streams shards row by row.. A closed-vocabvocabularyThe fixed set of tokens a tokenizer can produce. Mailwoman's SentencePiece vocabulary is tens of thousands of subword pieces, with byte fallback for anything outside it. tag with a large set and homographshomonymyOne surface, many referents: 'Georgia' is a US state in 'Atlanta, Georgia' and a country in 'Tbilisi, Georgia.' Handled by disambiguation, split across two stages — the parser resolves the tag from in-string context; the resolver late-binds the referent with geographic context. (country) additionally gets the gazetteergazetteerA geographical index that maps place names and postcodes to real-world coordinates. Mailwoman uses a custom-built Who's On First (WOF) SQLite database as its gazetteer — the 'atlas' half of the grammar/atlas architecture. as a soft anchor featurefeatureAn input signal a model conditions on. Beyond the raw tokens, Mailwoman feeds soft features — gazetteer-membership channels and the postcode anchor — that inform predictions without overriding them. — a match flag plus a "known-homographhomonymyOne surface, many referents: 'Georgia' is a US state in 'Atlanta, Georgia' and a country in 'Tbilisi, Georgia.' Handled by disambiguation, split across two stages — the parser resolves the tag from in-string context; the resolver late-binds the referent with geographic context." bit — fed into the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' the same way the postcode anchoranchor inferenceA technique where structured knowledge (postcode locations, gazetteer place names) is injected into the model as soft input features — not as deterministic overrides. The model still decides the final labels, but the anchor signal biases it toward correct admin tags. is, so the modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' decides with the lexicon as evidence. The matcher becomes a featurefeatureAn input signal a model conditions on. Beyond the raw tokens, Mailwoman feeds soft features — gazetteer-membership channels and the postcode anchor — that inform predictions without overriding them.-source, not a post-parseaddress parsingThe process of decomposing a free-text postal address string into structured components — house number, street name, locality, region, postcode, and country — so a geocoder can resolve them to coordinates. verdict. Full rationale, data recipe, and the evalevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error.-first plan: Closed-vocab fields: model-first.