app.py

import gradio as gr
from transformers import pipeline
import spacy
from gradio_client import Client
import re

# Initialize models
nlp = spacy.load("en_core_web_sm")
spell_checker = pipeline("text2text-generation", model="oliverguhr/spelling-correction-english-base")

def preprocess_capitalization(text: str) -> str:
    """Preprocess input text to handle capitalization rules."""
    words = text.split(" ")
    processed_words = []
    
    for word in words:
        if re.match(r"^[A-Z]+$", word):
            processed_words.append(word)  # Leave acronyms unchanged
        elif re.search(r"[A-Z]", word) and re.search(r"[a-z]", word):
            processed_words.append(word[0].upper() + word[1:].lower())  # Correct capitalization
        else:
            processed_words.append(word)  # Leave other words unchanged
    
    return " ".join(processed_words)

def preprocess_text(text: str, is_suggestion_applied: bool = False):
    """Process text and return corrections with position information."""
    result = {
        "spell_suggestions": [],
        "other_suggestions": [],
        "entities": [],
        "tags": []
    }

    # Only generate suggestions if no suggestion has been applied
    if not is_suggestion_applied:
        # Apply capitalization preprocessing (spell suggestions)
        capitalized_text = preprocess_capitalization(text)
        if capitalized_text != text:
            result["spell_suggestions"].append({
                "original": text,
                "corrected": capitalized_text
            })
            text = capitalized_text  # Update text for further processing

        # Transformer spell check (other suggestions)
        spell_checked = spell_checker(text, max_length=512)[0]['generated_text']
        if spell_checked != text:
            result["other_suggestions"].append({
                "original": text,
                "corrected": spell_checked
            })

    # Translate the text (after preprocessing if first pass, or as-is if suggestion applied)
    client = Client("Frenchizer/space_21")
    try:
        translation = client.predict(text)
    except Exception as e:
        translation = f"Error: {str(e)}"

    # Add entities and tags
    doc = nlp(text)
    result["entities"] = [{"text": ent.text, "label": ent.label_} for ent in doc.ents]
    result["tags"] = [token.text for token in doc if token.text.startswith(('#', '@'))]

    return translation, result

def preprocess_and_forward(text: str, is_suggestion_applied: bool = False):
    """Process text and forward to translation service."""
    translation, preprocessing_result = preprocess_text(text, is_suggestion_applied)
    return translation, preprocessing_result

# Gradio interface
with gr.Blocks() as demo:
    input_text = gr.Textbox(label="Input Text")
    is_suggestion_applied = gr.Checkbox(label="Suggestion Applied", value=False, visible=False)  # Hidden flag
    output_text = gr.Textbox(label="Output Text")
    preprocess_button = gr.Button("Process")
    preprocess_button.click(fn=preprocess_and_forward, inputs=[input_text, is_suggestion_applied], outputs=[output_text])

if __name__ == "__main__":
    demo.launch()