Thai Address Parsing

Overview

Thai addresses order components district-to-city the same broad direction as English (house number → subdistrict → district → province → postal code) but use Thai particle words (ตำบล/แขวง, อำเภอ/เขต, จังหวัด) that a naive English-address parser (split on commas, assume last token is a country) will mishandle — Bangkok uses แขวง/เขต instead of ตำบล/อำเภอ, and the parts can appear with or without abbreviations (ต./อ./จ.). This skill gives a regex-based extractor plus the 77-province reference table.

When to use

• แยกที่อยู่ไทยเป็นส่วนประกอบ (ตำบล, อำเภอ, จังหวัด, รหัสไปรษณีย์)
• หารหัสไปรษณีย์ 2 หลักแรก → จังหวัด (candidate lookup, not exact 5-digit resolution)
• Matching a province name across Thai/English spelling and with/without "จังหวัด"/"Province" wrapper
• Validating that a claimed province name is one of the real 77

When NOT to use

• The task needs the exact 5-digit postal code for a specific subdistrict — this skill narrows to province-level candidates from the first 2 digits only; direct the user to Thailand Post's official lookup for the precise code
• The address is non-Thai — use general address parsing instead

Core knowledge

• Thailand has 77 provinces (76 + Bangkok). Bangkok uses แขวง (subdistrict) / เขต (district) terminology; every other province uses ตำบล / อำเภอ.
• Postal codes are 5 digits. The first 2 digits identify a province-level zone in a well-documented sequential assignment (10 = Bangkok area, 20s = eastern seaboard, 30s-40s = northeast, 50s = north, 60s = lower-north/central, 70s = west, 80s-90s = south). The last 3 digits identify the specific post office/district — this skill does not resolve that level.
• Bangkok and Samut Prakan share the 10 prefix; several other neighboring provinces' outlying districts can carry a prefix that "belongs" to an adjacent province's normal range. Treat a prefix match as a candidate list, not a certainty.
• Standard field order in a full Thai address: [เลขที่/บ้านเลขที่] [หมู่ที่] [ถนน] ตำบล/แขวง[X] อำเภอ/เขต[Y] จังหวัด[Z] [รหัสไปรษณีย์].

Common mistakes

1. Assuming ตำบล/อำเภอ apply to Bangkok addresses — Bangkok uses แขวง/เขต instead.
2. Treating the first 2 postal-code digits as uniquely identifying one province — several pairs (notably Bangkok/Samut Prakan) share a prefix.
3. Matching province names with exact string equality only — fails on "จ.เชียงใหม่", "จังหวัดเชียงใหม่", "Chiang Mai Province", and plain "เชียงใหม่" all referring to the same province.
4. Assuming a mostly-empty parse result means "no address" — it more often means the input didn't use the expected Thai particle words (see Known limitations).
5. Fabricating a precise 5-digit postal code from a province name alone — the province-level prefix does not determine the last 3 digits.

Code

parse.py (uses provinces.json, no external dependencies):

• load_provinces() → list of {name_th, name_en, region, postal_prefix}
• find_province(name) → tolerant Thai/English match, handles "จังหวัด"/"จ."/"Province" wrapping
• postal_prefix_to_provinces(code) → candidate province list from the first 2 digits
• parse_thai_address(text) → {postal_code, province, district, subdistrict}

Run python3 parse.py for the self-test.

Known limitations

• parse_thai_address is a structural regex extractor, not a trained address-parsing model. It requires the Thai particle words (ตำบล/แขวง/อำเภอ/เขต/จังหวัด or their abbreviations) to be present in the text; a plain comma-separated address without those words will parse mostly empty.
• postal_prefix_to_provinces resolves to a province-level candidate list from 2 digits only — it cannot give the exact 5-digit code for a specific subdistrict, and a handful of border-area districts use a prefix outside their "home" province's normal range. Verify the exact code against Thailand Post's official lookup before printing it on a real document.
• Province English-name romanization follows one common convention (RTGS-influenced with spaces, e.g. "Buri Ram"); other sources spell some names differently (e.g. "Buriram" one word) — treat spelling variants as equivalent when matching, not as errors.
• provinces.json's region field uses one common 6-region grouping; other classifications (e.g. Royal Institute's 4-region scheme) group some Central provinces as "Lower North" instead — don't treat region as an official government classification.