
Kako smo koristili OpenAI API za prevođenje naše web stranice
- Članci, Priče
- Hugo , OpenAI , Prevođenje , Automatizacija
- 19. ožujka 2025.
- 40 min read
Uvod
Kada smo odlučili učiniti našu web stranicu temeljenu na GoHugo.io višejezičnom, željeli smo učinkovit, skalabilan i isplativ način za generiranje prijevoda. Umjesto da ručno prevodimo svaku stranicu, iskoristili smo OpenAI API za automatizaciju procesa. Ovaj članak objašnjava kako smo integrirali OpenAI API s Hugom, koristeći HugoPlate temu iz Zeon Studija, kako bismo brzo i točno generirali prijevode.
Zašto smo odabrali OpenAI API za prevođenje
Tradicionalne usluge prevođenja često zahtijevaju značajan ručni trud, a automatizirani alati poput Google Prevoditelja, iako korisni, ne pružaju uvijek razinu prilagodbe koju smo trebali. OpenAI API nam je omogućio:
- Automatizaciju prijevoda u velikim količinama
- Prilagodbu stila prijevoda
- Održavanje boljeg nadzora nad kvalitetom
- Besprijekornu integraciju s našom web stranicom temeljnom na Hugu
- Označavanje pojedinačnih stranica za ponovno prevođenje
Postupak korak po korak
1. Priprema Hugo web stranice
Naša stranica već je bila postavljena koristeći HugoPlate temu, koja podržava višejezičnu funkcionalnost. Prvi korak bio je omogućiti podršku za jezike u našoj Hugo config/_default/languages.toml
datoteci:
################ English language ##################
[en]
languageName = "English"
languageCode = "en-us"
contentDir = "content/english"
weight = 1
################ Arabic language ##################
[ar]
languageName = "العربية"
languageCode = "ar"
contentDir = "content/arabic"
languageDirection = 'rtl'
weight = 2
Ova konfiguracija osigurava da Hugo može generirati odvojene jezične verzije našeg sadržaja.
2. Automatizacija prijevoda s OpenAI API
Razvili smo Bash skriptu za automatizaciju prevođenja Markdown datoteka. Ova skripta:
- Čita engleske
.md
datoteke iz izvorne mape. - Koristi OpenAI API za prevođenje teksta uz očuvanje Markdown formatiranja.
- Piše prevedeni sadržaj u odgovarajuće jezične mape.
- Prati status prevođenja koristeći JSON datoteku.
Evo pregleda naše skripte:
#!/bin/bash
# ===========================================
# Hugo Content Translation and Update Script (Sequential Processing & New-Language Cleanup)
# ===========================================
# This script translates Hugo Markdown (.md) files from English to all supported target languages
# sequentially (one file at a time). It updates a JSON status file after processing each file.
# At the end of the run, it checks translation_status.json and removes any language from
# translate_new_language.txt only if every file for that language is marked as "success".
# ===========================================
set -euo pipefail
# --- Simple Logging Function (writes to stderr) ---
log_step() {
echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*" >&2
}
# --- Environment Setup ---
export PATH="/opt/homebrew/opt/coreutils/libexec/gnubin:$PATH"
# (Removed "Script starting." log)
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
log_step "SCRIPT_DIR set to: $SCRIPT_DIR"
if [ -f "$SCRIPT_DIR/.env" ]; then
log_step "Loading environment variables from .env"
set -o allexport
source "$SCRIPT_DIR/.env"
set +o allexport
fi
# Load new languages from translate_new_language.txt (if available)
declare -a NEW_LANGUAGES=()
if [ -f "$SCRIPT_DIR/translate_new_language.txt" ]; then
while IFS= read -r line || [[ -n "$line" ]]; do
NEW_LANGUAGES+=("$line")
done <"$SCRIPT_DIR/translate_new_language.txt"
else
log_step "No new languages file found; proceeding with empty NEW_LANGUAGES."
fi
API_KEY="${OPENAI_API_KEY:-}"
if [ -z "$API_KEY" ]; then
log_step "❌ Error: OPENAI_API_KEY environment variable is not set."
exit 1
fi
# Supported Languages (full list)
SUPPORTED_LANGUAGES=("ar" "bg" "bn" "cs" "da" "de" "el" "es" "fa" "fi" "fr" "ha" "he" "hi" "hr" "hu" "id" "ig" "it" "ja" "ko" "ml" "mr" "ms" "nl" "no" "pa" "pl" "pt" "ro" "ru" "sk" "sn" "so" "sr" "sv" "sw" "ta" "te" "th" "tl" "tr" "uk" "vi" "xh" "yo" "zh" "zu")
STATUS_FILE="$SCRIPT_DIR/translation_status.json"
SRC_DIR="$SCRIPT_DIR/Content/english"
log_step "Source directory: $SRC_DIR"
# Check dependencies
for cmd in jq curl; do
if ! command -v "$cmd" >/dev/null 2>&1; then
log_step "❌ Error: '$cmd' is required. Please install it."
exit 1
fi
done
MAX_RETRIES=5
WAIT_TIME=2 # seconds
# Create/initialize status file if missing
if [ ! -f "$STATUS_FILE" ]; then
echo "{}" >"$STATUS_FILE"
log_step "Initialized status file at: $STATUS_FILE"
fi
# --- Locking for Status Updates ---
lock_status() {
local max_wait=10
local start_time
start_time=$(date +%s)
while ! mkdir "$STATUS_FILE.lockdir" 2>/dev/null; do
sleep 0.01
local now
now=$(date +%s)
if ((now - start_time >= max_wait)); then
log_step "WARNING: Lock wait exceeded ${max_wait}s. Forcibly removing stale lock."
rm -rf "$STATUS_FILE.lockdir"
fi
done
}
unlock_status() {
rmdir "$STATUS_FILE.lockdir"
}
update_status() {
local file_path="$1" lang="$2" status="$3"
lock_status
jq --arg file "$file_path" --arg lang "$lang" --arg status "$status" \
'.[$file][$lang] = $status' "$STATUS_FILE" >"$STATUS_FILE.tmp" && mv "$STATUS_FILE.tmp" "$STATUS_FILE"
unlock_status
}
# --- Translation Function ---
translate_text() {
local text="$1" lang="$2"
local retry_count=0
while [ "$retry_count" -lt "$MAX_RETRIES" ]; do
user_message="Translate the following text to $lang. Preserve all formatting exactly as in the original.
$text"
json_payload=$(jq -n \
--arg system "Translate from English to $lang. Preserve original formatting exactly." \
--arg user_message "$user_message" \
'{
"model": "gpt-4o-mini",
"messages": [
{"role": "system", "content": $system},
{"role": "user", "content": $user_message}
],
"temperature": 0.3
}')
response=$(curl -s https://api.openai.com/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $API_KEY" \
-d "$json_payload")
log_step "📥 Received API response."
local error_type
error_type=$(echo "$response" | jq -r '.error.type // empty')
local error_message
error_message=$(echo "$response" | jq -r '.error.message // empty')
if [ "$error_type" == "insufficient_quota" ]; then
sleep "$WAIT_TIME"
retry_count=$((retry_count + 1))
elif [[ "$error_type" == "rate_limit_reached" || "$error_type" == "server_error" || "$error_type" == "service_unavailable" ]]; then
sleep "$WAIT_TIME"
retry_count=$((retry_count + 1))
elif [ "$error_type" == "invalid_request_error" ]; then
return 1
elif [ -z "$error_type" ]; then
if ! translated_text=$(echo "$response" | jq -r '.choices[0].message.content' 2>/dev/null); then
return 1
fi
if [ "$translated_text" == "null" ] || [ -z "$translated_text" ]; then
return 1
else
translated_text=$(echo "$translated_text" | sed -e 's/^```[[:space:]]*//; s/[[:space:]]*```$//')
echo "$translated_text"
return 0
fi
else
return 1
fi
done
return 1
}
# --- Process a Single File (Sequential Version) ---
process_file() {
local src_file="$1" target_file="$2" lang="$3" rel_src="$4"
# If target file exists and is non-empty, mark status as success.
if [ -s "$target_file" ]; then
update_status "$rel_src" "$lang" "success"
return 0
fi
content=$(<"$src_file")
if [[ "$content" =~ ^(---|\+\+\+)[[:space:]]*$ ]] && [[ "$content" =~ [[:space:]]*(---|\+\+\+\+)[[:space:]]*$ ]]; then
front_matter=$(echo "$content" | sed -n '/^\(---\|\+\+\+\)$/,/^\(---\|\+\+\+\)$/p')
body_content=$(echo "$content" | sed -n '/^\(---\|\+\+\+\)$/,/^\(---\|\+\+\+\)$/d')
else
front_matter=""
body_content="$content"
fi
log_step "Translating [$rel_src] to $lang..."
translated_body=$(translate_text "$body_content" "$lang")
if [ $? -ne 0 ]; then
update_status "$rel_src" "$lang" "failed"
return 1
fi
mkdir -p "$(dirname "$target_file")"
if [ -n "$front_matter" ]; then
echo -e "$front_matter
$translated_body" >"$target_file"
else
echo -e "$translated_body" >"$target_file"
fi
updated_content=$(echo "$content" | sed -E 's/^retranslate:\s*true/retranslate: false/')
echo "$updated_content" >"$src_file"
update_status "$rel_src" "$lang" "success"
}
# --- Main Sequential Processing ---
ALL_SUCCESS=true
for TARGET_LANG in "${SUPPORTED_LANGUAGES[@]}"; do
log_step "Processing language: $TARGET_LANG"
TARGET_DIR="$SCRIPT_DIR/Content/$TARGET_LANG"
while IFS= read -r -d '' src_file; do
rel_src="${src_file#$SCRIPT_DIR/}"
target_file="$TARGET_DIR/${src_file#$SRC_DIR/}"
# If file is marked not to retranslate, check that target file exists and is non-empty.
if ! [[ " ${NEW_LANGUAGES[@]:-} " =~ " ${TARGET_LANG} " ]] && grep -q '^retranslate:\s*false' "$src_file"; then
if [ -s "$target_file" ]; then
update_status "$rel_src" "$TARGET_LANG" "success"
else
update_status "$rel_src" "$TARGET_LANG" "failed"
fi
continue
fi
process_file "$src_file" "$target_file" "$TARGET_LANG" "$rel_src"
done < <(find "$SRC_DIR" -type f -name "*.md" -print0)
done
log_step "Translation run completed."
end_time=$(date +%s)
duration=$((end_time - $(date +%s)))
log_step "Execution Time: $duration seconds"
if [ "$ALL_SUCCESS" = true ]; then
log_step "🎉 Translation completed successfully for all supported languages!"
else
log_step "⚠️ Translation completed with some errors."
fi
# --- Clean Up Fully Translated New Languages ---
if [ -f "$SCRIPT_DIR/translate_new_language.txt" ]; then
log_step "Cleaning up fully translated new languages..."
for lang in "${NEW_LANGUAGES[@]:-}"; do
incomplete=$(jq --arg lang "$lang" 'to_entries[] | select(.value[$lang] != null and (.value[$lang] != "success")) | .key' "$STATUS_FILE")
if [ -z "$incomplete" ]; then
log_step "All translations for new language '$lang' are marked as success. Removing from translate_new_language.txt."
sed -E -i '' "/^[[:space:]]*$lang[[:space:]]*$/d" "$SCRIPT_DIR/translate_new_language.txt"
else
log_step "Language '$lang' still has incomplete translations."
fi
done
fi
3. Upravljanje statusom prevođenja
Kako bismo spriječili redundantna prevođenja i pratili napredak, koristili smo JSON datoteku (translation_status.json
). Skripta ažurira ovu datoteku nakon obrade svakog dokumenta, osiguravajući da se prevodi samo novi ili ažurirani sadržaj.
4. Rukovanje greškama i ograničenjima API-ja
Implementirali smo ponovne pokušaje i rukovanje greškama kako bismo se nosili s ograničenjima, greškama API-ja i problemima s kvotama. Skripta čeka prije ponovnog pokušaja ako OpenAI API vrati grešku poput rate_limit_reached
ili service_unavailable
.
5. Implementacija
Nakon što je prevedeni sadržaj generiran, pokretanje hugo --minify
gradi višejezičnu statičku stranicu, spremnu za implementaciju.
Izazovi i rješenja
1. Točnost prijevoda
Iako su OpenAI-ovi prijevodi općenito bili točni, neki tehnički pojmovi mogu zahtijevati ručni pregled, ali mi smo samo tim od dvoje, pa se nadamo najboljem. Fino smo prilagodili upite kako bismo održali kontekst i ton.
2. Problemi s formatiranjem
Markdown sintaksa ponekad se mijenjala u prijevodu. Kako bismo to popravili, dodali smo logiku za post-procesiranje kako bismo očuvali formatiranje.
3. Optimizacija troškova API-ja
Kako bismo smanjili troškove, implementirali smo keširanje kako bismo izbjegli ponovno prevođenje nepromijenjenog sadržaja.
4. Učinkovito rukovanje ponovnim prevođenjem
Kako bismo ponovno preveli specifične stranice, dodali smo retranslate: true
parametar u front matter. Skripta ponovno prevodi samo stranice označene ovim parametrom. To nam omogućuje ažuriranje prijevoda prema potrebi bez ponovnog prevođenja cijele stranice.
Zaključak
Integracijom OpenAI API-ja s Hugom, automatizirali smo prevođenje naše web stranice dok smo održavali kvalitetu i fleksibilnost. Ovaj pristup uštedio je vrijeme, osigurao dosljednost i omogućio nam da se bez napora skaliramo. Ako tražite način da svoju Hugo stranicu učinite višejezičnom, OpenAI-ov API nudi moćno rješenje.