app.py

import streamlit as st
import sacrebleu
from bert_score import score as bert_score
import jieba

# Function to calculate BLEU score
def calculate_bleu(translations, references):
    return sacrebleu.corpus_bleu(translations, [references]).score

# Function to calculate TER score
def calculate_ter(translations, references):
    return sacrebleu.corpus_ter(translations, [references]).score

# Function to calculate CHRF score
def calculate_chrf(translations, references):
    return sacrebleu.corpus_chrf(translations, [references]).score

# Function to calculate BERTScore
def calculate_bertscore(translations, references, lang):
    P, R, F1 = bert_score(translations, references, lang=lang)
    return F1.mean().item()

# Streamlit app
st.title("Machine Translation Quality Evaluation")
st.write("Input the translated text and the reference translation to compute BLEU, TER, CHRF, and BERTScore metrics.")

# List of supported languages
languages = {
    "English": "en",
    "Chinese": "zh",
    "French": "fr",
    "German": "de",
    "Spanish": "es",
    "Russian": "ru",
    "Japanese": "ja",
    "Korean": "ko",
    "Arabic": "ar",
    "Italian": "it",
    "Dutch": "nl",
    "Portuguese": "pt",
    "Turkish": "tr",
    "Polish": "pl",
    "Czech": "cs",
    "Swedish": "sv",
    "Danish": "da",
    "Finnish": "fi",
    "Greek": "el",
    "Hungarian": "hu",
    "Indonesian": "id",
    "Norwegian": "no",
    "Romanian": "ro",
    "Thai": "th",
    "Vietnamese": "vi",
    "Hebrew": "he",
    "Hindi": "hi",
    "Bengali": "bn",
    "Tamil": "ta",
    "Urdu": "ur",
    "Other": "other"
}

# Language selection
source_lang = st.selectbox("Select Source Language", list(languages.keys()))
target_lang = st.selectbox("Select Target Language", list(languages.keys()))

# Input fields for custom language codes if "Other" is selected
source_lang_code = st.text_input("Enter Source Language Code (ISO 639-1):", value=languages[source_lang]) if source_lang == "Other" else languages[source_lang]
target_lang_code = st.text_input("Enter Target Language Code (ISO 639-1):", value=languages[target_lang]) if target_lang == "Other" else languages[target_lang]

# Input fields for translations and references
translation_input = st.text_area("Translated Text", height=200)
reference_input = st.text_area("Reference Translation", height=200)

# Evaluate button
if st.button("Evaluate"):
    if translation_input and reference_input:
        translations = [translation_input.strip()]
        references = [reference_input.strip()]

        # Handle tokenization if necessary (e.g., for Chinese)
        if source_lang_code == "zh" or target_lang_code == "zh":
            translations = [' '.join(jieba.cut(text)) for text in translations]
            references = [' '.join(jieba.cut(text)) for text in references]

        bleu_score = calculate_bleu(translations, references)
        ter_score = calculate_ter(translations, references)
        chrf_score = calculate_chrf(translations, references)
        bertscore = calculate_bertscore(translations, references, target_lang_code)

        st.write(f"**BLEU Score:** {bleu_score:.2f}")
        st.write(f"**TER Score:** {ter_score:.2f}")
        st.write(f"**CHRF Score:** {chrf_score:.2f}")
        st.write(f"**BERTScore:** {bertscore:.2f}")
    else:
        st.error("Please provide both translated text and reference translation.")