MODEL 04 · STRUCTURED EXTRACTION

Name Parser

Powered by PNEUMA-DD (production) · MondoGraph token statistics

Split "Maria Teresa García Ramírez de Arroyo" into 12 typed slots — given names, paternal surname, maternal surname, particles, titles, honorifics, suffixes — and do it correctly across every world naming convention (Spanish double-surname, Arabic kunya, Japanese family-first, Vietnamese reverse, Icelandic patronymic). Backed by 101M distinct GIVEN+SURNAME forms in MondoGraph, with strong coverage even for rare regional names.

Model card SmartForms case study

Try it — full name → structured parts

Paste any full name as it would appear in a form. The parser returns the detected convention, then the parts.

EXAMPLES:
Enter a name and press Parse.

Model card

Architecture
XLM-R encoder + CRF token labeler · 280M params
Slots
title · given · middle · particle · paternal-surname · maternal-surname · suffix · honorific · patronymic · matronymic · kunya · nickname
Conventions supported
Spanish · Portuguese · Arabic · Persian · Japanese · Chinese · Korean · Vietnamese · Icelandic · Hungarian · Russian (patronymic) · 22 more
Token-F1
0.967 macro · 0.984 micro
Convention detection
97.3% accuracy on 38-way classification
Latency
~30ms typical for a 6-token name

API

curl -X POST https://api.mondonomo.ai/v1/parse \ -H "Authorization: Bearer $TOKEN" \ -d '{ "name": "Maria Teresa García Ramírez de Arroyo", "locale_hint": "es-MX" }' { "convention": "spanish-double", "tokens": [ {"slot": "given", "value": "Maria Teresa"}, {"slot": "paternal-surname", "value": "García"}, {"slot": "maternal-surname", "value": "Ramírez"}, {"slot": "particle", "value": "de"}, {"slot": "maiden-or-spouse", "value": "Arroyo"} ], "address": {"formal": "Señora García", "casual": "Maria"} }
The form problem. Traditional sign-up forms ask for First / Middle / Last Name. That works for ~30% of the world. For Maria, the "Last name" field is either wrong, ambiguous, or destroys her maternal lineage. The Mondonomo Smart Forms product uses Parser to accept one free-text field and produce the structured record correctly.

RELATED

A typical pipeline.

Classifier → is it a person name? Parser → split it. Gender → infer from given. Transliterator → render across scripts. Soundalike → dedupe against the existing CRM.

MODEL 03

Classifier

Use first as the gate so you only parse strings that really are person names.

entity-typing→
MODEL 05

Gender

Once parsed, the given-name slot is exactly the input shape Gender expects.

demographics→
MODEL 06

Soundalikes

Deduplicate the parsed surname against your CRM with phonetic similarity.

echoes→