KameliaZaman/

French-to-English-Translation

French to English Machine Translation

French to English language translation using sequence to sequence transformer.
View Demo

About The Project

Built With

Getting Started

Dependencies

• TensorFlow 2.16.1
  conda install -c conda-forge tensorflow
• Keras 2.15.0
  conda install -c conda-forge keras
• Gradio 4.24.0
  conda install -c conda-forge gradio
• NumPy 1.26.4
  conda install -c conda-forge numpy

Alternative: Export Environment

Installation

Usage

Dataset

Model Architecture

Data Preparation

• The parallel corpus containing French and English sentences is preprocessed.
• Text is tokenized and converted into numerical representations suitable for input to the neural network.

Model Training

• The sequence-to-sequence model is constructed, comprising an encoder and decoder.
• Training data is fed into the model, and parameters are optimized using backpropagation and gradient descent algorithms.
  sh

  1def create_model(src_vocab, tar_vocab, src_timesteps, tar_timesteps, n_units): 2 # Create the model 3 model = Sequential() 4 model.add(Embedding(src_vocab_size, n_units, input_length=src_length, mask_zero=True)) 5 model.add(LSTM(n_units)) 6 model.add(RepeatVector(tar_timesteps)) 7 model.add(LSTM(n_units, return_sequences=True)) 8 model.add(TimeDistributed(Dense(tar_vocab, activation='softmax'))) 9 return model 10 11model = create_model(src_vocab_size, tar_vocab_size, src_length, tar_length, 256) 12model.compile(optimizer='adam', loss='categorical_crossentropy') 13 14history = model.fit(trainX, 15 trainY, 16 epochs=20, 17 batch_size=64, 18 validation_split=0.1, 19 verbose=1, 20 callbacks=[ 21 EarlyStopping( 22 monitor='val_loss', 23 patience=10, 24 restore_best_weights=True 25 ) 26 ])

  

Model Evaluation

• The trained model is evaluated on the test set to measure its accuracy.
• Metrics such as BLEU score has been used to quantify the quality of translations.

  

  

Deployment

• Gradio is utilized for deploying the trained model.
• Users can input a French text, and the model will translate it to English.
  sh

  1import string 2import re 3from unicodedata import normalize 4import numpy as np 5from keras.preprocessing.text import Tokenizer 6from keras.preprocessing.sequence import pad_sequences 7from keras.utils import to_categorical 8from keras.models import Sequential,load_model 9from keras.layers import LSTM,Dense,Embedding,RepeatVector,TimeDistributed 10from keras.callbacks import EarlyStopping 11from nltk.translate.bleu_score import corpus_bleu 12import pandas as pd 13from string import punctuation 14import matplotlib.pyplot as plt 15from IPython.display import Markdown, display 16import gradio as gr 17import tensorflow as tf 18from tensorflow.keras.models import load_model 19 20total_sentences = 10000 21 22dataset = pd.read_csv("./eng_-french.csv", nrows = total_sentences) 23 24def clean(string): 25 # Clean the string 26 string = string.replace("\u202f"," ") # Replace no-break space with space 27 string = string.lower() 28 29 # Delete the punctuation and the numbers 30 for p in punctuation + "«»" + "0123456789": 31 string = string.replace(p," ") 32 33 string = re.sub('\s+',' ', string) 34 string = string.strip() 35 36 return string 37 38dataset = dataset.sample(frac=1, random_state=0) 39dataset["English words/sentences"] = dataset["English words/sentences"].apply(lambda x: clean(x)) 40dataset["French words/sentences"] = dataset["French words/sentences"].apply(lambda x: clean(x)) 41 42dataset = dataset.values 43dataset = dataset[:total_sentences] 44 45source_str, target_str = "French", "English" 46idx_src, idx_tar = 1, 0 47 48def create_tokenizer(lines): 49 # fit a tokenizer 50 tokenizer = Tokenizer() 51 tokenizer.fit_on_texts(lines) 52 return tokenizer 53 54def max_len(lines): 55 # max sentence length 56 return max(len(line.split()) for line in lines) 57 58def encode_sequences(tokenizer, length, lines): 59 # encode and pad sequences 60 X = tokenizer.texts_to_sequences(lines) # integer encode sequences 61 X = pad_sequences(X, maxlen=length, padding='post') # pad sequences with 0 values 62 return X 63 64def word_for_id(integer, tokenizer): 65 # map an integer to a word 66 for word, index in tokenizer.word_index.items(): 67 if index == integer: 68 return word 69 return None 70 71def predict_seq(model, tokenizer, source): 72 # generate target from a source sequence 73 prediction = model.predict(source, verbose=0)[0] 74 integers = [np.argmax(vector) for vector in prediction] 75 target = list() 76 for i in integers: 77 word = word_for_id(i, tokenizer) 78 if word is None: 79 break 80 target.append(word) 81 return ' '.join(target) 82 83src_tokenizer = create_tokenizer(dataset[:, idx_src]) 84src_vocab_size = len(src_tokenizer.word_index) + 1 85src_length = max_len(dataset[:, idx_src]) 86tar_tokenizer = create_tokenizer(dataset[:, idx_tar]) 87 88model = load_model('./french_to_english_translator.h5') 89 90def translate_french_english(french_sentence): 91 # Clean the input sentence 92 french_sentence = clean(french_sentence) 93 # Tokenize and pad the input sentence 94 input_sequence = encode_sequences(src_tokenizer, src_length, [french_sentence]) 95 # Generate the translation 96 english_translation = predict_seq(model, tar_tokenizer, input_sequence) 97 return english_translation 98 99gr.Interface( 100 fn=translate_french_english, 101 inputs="text", 102 outputs="text", 103 title="French to English Translator", 104 description="Translate French sentences to English." 105).launch()

  

Contributing

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

License

Contact