Add multilingual extraction and classification support

Author: ndbroadbent

src/classifier/prompt-sections.ts

@@ -58,6 +58,13 @@ DON'T use for: generic ("go to movies"), places (use placeName), bands (use wiki
 export const SHARED_CATEGORIES_SECTION = `CATEGORIES: ${VALID_CATEGORIES.join(', ')}
 ("other" should be used only as a last resort. Only use it if no other category applies.)`
 
+export const SHARED_LANGUAGE_SECTION = `LANGUAGE HANDLING:
+- Chats may be in any language or mix multiple languages in one conversation.
+- Treat translated equivalents of "we should", "let's go", "want to visit", enthusiasm, and agreement the same as English.
+- Output JSON field names exactly as specified. Use concise English for generic activity descriptions when natural.
+- Preserve proper nouns, venue names, media titles, addresses, and city/place names in their original script unless a standard English name is obvious.
+- Do not skip a candidate just because the surrounding context is not English.`
+
 export const SHARED_NORMALIZATION = `NORMALIZATION:
 - Distinct categories: cafe≠restaurant, bar≠restaurant
 - KEEP mediaKey specificity: "glow worm cave" not "cave", "hot air balloon" not "balloon"

src/classifier/prompt.test.ts

@@ -146,6 +146,16 @@ describe('Classifier Prompt', () => {
       expect(prompt).toContain('adult content')
     })
 
+    it('includes multilingual analysis instructions', () => {
+      const candidates = [createCandidate(1, 'あのレストランに行ってみたい')]
+
+      const prompt = buildClassificationPrompt(candidates, TEST_CONTEXT)
+
+      expect(prompt).toContain('Chats may be in any language')
+      expect(prompt).toContain('Preserve proper nouns')
+      expect(prompt).toContain('あのレストランに行ってみたい')
+    })
+
     it('tags agreement candidates with [AGREE]', () => {
       const candidates = [createCandidate(1, 'Sounds great!', [], [], 'agreement')]
 

src/classifier/prompt.ts

@@ -18,6 +18,7 @@ import {
   SHARED_EXAMPLES,
   SHARED_IMAGE_SECTION,
   SHARED_INCLUDE_RULES,
+  SHARED_LANGUAGE_SECTION,
   SHARED_LINK_SECTION,
   SHARED_NORMALIZATION,
   SHARED_SKIP_RULES,
@@ -186,6 +187,8 @@ ${formatted}
 
 ${buildUserContextSection(context.homeCountry, context.timezone)}
 
+${SHARED_LANGUAGE_SECTION}
+
 WHY THESE MESSAGES:
 You're seeing messages pre-filtered by heuristics (regex patterns like "let's go", "we should try") and semantic search (embeddings). We intentionally cast a wide net - you'll see some false positives. Your job is to identify the real activities worth saving.
 
@@ -250,6 +253,8 @@ ${formatted}
 
 ${buildUserContextSection(context.homeCountry, context.timezone)}
 
+${SHARED_LANGUAGE_SECTION}
+
 These are AGREEMENT messages - phrases like "sounds great!", "I'm keen!", "let's do it!". Your job is to find WHAT they are agreeing to by looking at the messages BEFORE the >>> candidate.
 
 The >>> message is the agreement itself. Look at the context BEFORE it to find the activity being agreed to.

src/extraction/embeddings/index.test.ts

@@ -450,6 +450,9 @@ describe('Embeddings Module', () => {
       expect(DEFAULT_ACTIVITY_QUERIES.length).toBeGreaterThan(0)
       expect(DEFAULT_ACTIVITY_QUERIES.some((q) => q.includes('restaurant'))).toBe(true)
       expect(DEFAULT_ACTIVITY_QUERIES.some((q) => q.includes('hiking'))).toBe(true)
+      expect(DEFAULT_ACTIVITY_QUERIES.some((q) => q.includes('deberíamos'))).toBe(true)
+      expect(DEFAULT_ACTIVITY_QUERIES.some((q) => q.includes('行ってみたい'))).toBe(true)
+      expect(DEFAULT_ACTIVITY_QUERIES.some((q) => q.includes('我們應該'))).toBe(true)
     })
   })
 })

src/extraction/embeddings/queries/activity-types.json

@@ -26,7 +26,20 @@
 		"let's get dim sum",
 		"we should get tapas",
 		"let's go to the food market",
-		"we should check out the farmers market"
+		"we should check out the farmers market",
+		"restaurante cafe bar tapas",
+		"restaurante café bar vinho",
+		"restaurant café bar bistro",
+		"restaurant kneipe biergarten",
+		"ristorante pizzeria gelato",
+		"restauracja kawiarnia bar",
+		"ресторан кафе бар",
+		"restoran kafe makanan",
+		"مطعم مقهى طعام",
+		"レストラン カフェ 居酒屋",
+		"식당 카페 술집",
+		"餐厅 咖啡馆 酒吧",
+		"餐廳 咖啡館 酒吧"
 	],
 	"nature_outdoors": [
 		"we should go hiking",
@@ -57,7 +70,22 @@
 		"let's go diving",
 		"we should try kayaking",
 		"let's go paddle boarding",
-		"we should go surfing"
+		"we should go surfing",
+		"senderismo playa montaña parque",
+		"trilha praia cachoeira parque",
+		"randonnée plage montagne parc",
+		"wanderung strand berg park",
+		"escursione spiaggia montagna parco",
+		"wandeling strand natuurpark",
+		"wędrówka plaża góry park",
+		"поход пляж горы парк",
+		"mendaki pantai gunung taman",
+		"yürüyüş plaj dağ park",
+		"شاطئ جبل حديقة",
+		"ハイキング ビーチ 山 公園",
+		"등산 해변 산 공원",
+		"徒步 海滩 山 公园",
+		"徒步 海灘 山 公園"
 	],
 	"entertainment": [
 		"we should go to that concert",
@@ -87,7 +115,22 @@
 		"let's play mini golf",
 		"we should try laser tag",
 		"let's go to go karts",
-		"we should go to the theme park"
+		"we should go to the theme park",
+		"concierto cine festival teatro",
+		"spettacolo cinema festival teatro",
+		"concert cinéma festival théâtre",
+		"konzert kino festival theater",
+		"concerto cinema festival teatro",
+		"concert bioscoop festival theater",
+		"koncert kino festiwal teatr",
+		"концерт кино фестиваль театр",
+		"konser film festival teater",
+		"konser sinema festival tiyatro",
+		"حفلة سينما مهرجان مسرح",
+		"コンサート 映画 フェス",
+		"콘서트 영화 축제 공연",
+		"演唱会 电影 节日 演出",
+		"演唱會 電影 節日 表演"
 	],
 	"culture_arts": [
 		"we should visit the museum",
@@ -118,7 +161,21 @@
 		"let's do a food tour",
 		"we should do the ghost tour",
 		"let's do a pub crawl",
-		"we should do wine tasting"
+		"we should do wine tasting",
+		"museo galería exposición",
+		"museu galeria exposição",
+		"musée galerie exposition",
+		"museum galerie ausstellung",
+		"museo galleria mostra",
+		"muzeum galeria wystawa",
+		"музей галерея выставка",
+		"museum galeri pameran",
+		"müze galeri sergi",
+		"متحف معرض",
+		"美術館 博物館 展覧会",
+		"박물관 미술관 전시",
+		"博物馆 美术馆 展览",
+		"博物館 美術館 展覽"
 	],
 	"travel_accommodation": [
 		"let's book an airbnb",
@@ -136,7 +193,22 @@
 		"we should do a city break",
 		"let's do a staycation",
 		"I found cheap flights",
-		"there's a travel deal"
+		"there's a travel deal",
+		"viaje hotel escapada vacaciones",
+		"viagem hotel férias",
+		"voyage hôtel vacances",
+		"reise hotel urlaub",
+		"viaggio hotel vacanza",
+		"reis hotel vakantie",
+		"podróż hotel wakacje",
+		"поездка отель отпуск",
+		"यात्रा होटल छुट्टी",
+		"perjalanan hotel liburan",
+		"seyahat otel tatil",
+		"رحلة فندق عطلة",
+		"旅行 ホテル 休暇",
+		"여행 호텔 휴가",
+		"旅行 酒店 假期"
 	],
 	"sports_adventure": [
 		"we should go skiing",
@@ -168,7 +240,22 @@
 		"let's go on safari",
 		"we should go whale watching",
 		"let's go dolphin watching",
-		"we should do a wildlife tour"
+		"we should do a wildlife tour",
+		"esquí surf buceo escalada",
+		"esqui surfe mergulho escalada",
+		"ski surf plongée escalade",
+		"ski surfen tauchen klettern",
+		"sci surf immersione arrampicata",
+		"skiën surfen duiken klimmen",
+		"narciarstwo surfing nurkowanie wspinaczka",
+		"лыжи серфинг дайвинг скалолазание",
+		"ski selancar menyelam panjat",
+		"kayak sörf dalış tırmanış",
+		"تزلج غوص تسلق",
+		"スキー サーフィン ダイビング",
+		"스키 서핑 다이빙 클라이밍",
+		"滑雪 冲浪 潜水 攀岩",
+		"滑雪 衝浪 潛水 攀岩"
 	],
 	"social_events": [
 		"let's throw a party",
@@ -189,7 +276,22 @@
 		"we should take a workshop",
 		"let's take a class",
 		"we should take a lesson",
-		"let's sign up for that course"
+		"let's sign up for that course",
+		"fiesta cumpleaños reunión taller",
+		"festa aniversário encontro oficina",
+		"fête anniversaire atelier",
+		"party geburtstag treffen kurs",
+		"festa compleanno laboratorio",
+		"feest verjaardag workshop",
+		"impreza urodziny warsztaty",
+		"вечеринка день рождения мастер-класс",
+		"pesta ulang tahun lokakarya",
+		"parti doğum günü atölye",
+		"حفلة ورشة عيد ميلاد",
+		"パーティー ワークショップ 誕生日",
+		"파티 생일 워크숍",
+		"派对 生日 工作坊",
+		"派對 生日 工作坊"
 	],
 	"shopping": [
 		"let's go to the mall",
@@ -199,7 +301,21 @@
 		"we should go to the market",
 		"let's explore the bazaar",
 		"we should get souvenirs",
-		"let's go to the gift shop"
+		"let's go to the gift shop",
+		"mercado tienda bazar recuerdos",
+		"mercado loja lembranças",
+		"marché boutique souvenirs",
+		"markt laden souvenirs",
+		"mercato negozio souvenir",
+		"rynek sklep pamiątki",
+		"рынок магазин сувениры",
+		"pasar toko suvenir",
+		"pazar mağaza hediyelik",
+		"سوق متجر هدايا",
+		"市場 ショッピング お土産",
+		"시장 쇼핑 기념품",
+		"市场 商店 纪念品",
+		"市場 商店 紀念品"
 	],
 	"family_kids": [
 		"let's go to the playground",
@@ -210,6 +326,20 @@
 		"we should visit the petting zoo",
 		"let's do a farm visit",
 		"we should go to the science museum",
-		"let's try the trampoline park"
+		"let's try the trampoline park",
+		"zoológico acuario parque infantil",
+		"zoológico aquário parque infantil",
+		"zoo aquarium aire de jeux",
+		"zoo aquarium spielplatz",
+		"zoo acquario parco giochi",
+		"zoo akwarium plac zabaw",
+		"зоопарк аквариум детская площадка",
+		"kebun binatang akuarium taman bermain",
+		"hayvanat bahçesi akvaryum oyun alanı",
+		"حديقة حيوان حوض أسماك",
+		"動物園 水族館 遊園地",
+		"동물원 수족관 놀이터",
+		"动物园 水族馆 游乐园",
+		"動物園 水族館 遊樂園"
 	]
 }

src/extraction/embeddings/queries/agreement.json

@@ -6,5 +6,61 @@
 	"that looks awesome",
 	"i'm keen to do that",
 	"i'd be keen",
-	"let's book it"
+	"let's book it",
+	"suena bien",
+	"me apunto",
+	"qué buena pinta",
+	"parece bom",
+	"boa ideia",
+	"eu topo",
+	"klingt gut",
+	"ich bin dabei",
+	"sieht gut aus",
+	"ça a l'air bien",
+	"partant",
+	"bonne idée",
+	"sembra bello",
+	"ci sto",
+	"bella idea",
+	"klinkt goed",
+	"ik ben erbij",
+	"ziet er leuk uit",
+	"låter bra",
+	"jag är på",
+	"ser kul ut",
+	"lyder godt",
+	"jeg er på",
+	"ser sjovt ud",
+	"høres bra",
+	"jeg er med",
+	"ser gøy ut",
+	"brzmi dobrze",
+	"jestem za",
+	"dobry pomysł",
+	"звучит здорово",
+	"я за",
+	"выглядит классно",
+	"अच्छा लगता है",
+	"मैं तैयार हूँ",
+	"बहुत बढ़िया",
+	"kedengarannya bagus",
+	"aku ikut",
+	"ide bagus",
+	"kulağa güzel",
+	"ben varım",
+	"iyi fikir",
+	"يبدو رائع",
+	"أنا معكم",
+	"فكرة جيدة",
+	"よさそう",
+	"いいね",
+	"面白そう",
+	"좋겠다",
+	"좋아",
+	"재밌겠다",
+	"听起来不错",
+	"聽起來不錯",
+	"看起来很好",
+	"看起來很好",
+	"好主意"
 ]

src/extraction/embeddings/queries/query-embeddings.json.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:311165f70eabbe70131dfbd6365fb6b05d07c74df6898c06d6fa809caba55090
-size 3465505
+oid sha256:7cc1019f6519505f38a4815fef9393b3814fe0b5ca8bf46f3906a5c8ccb9a046
+size 6982893