Bilingual Chatbot

Pakistan is a country where English is not the dominant language to be widely used among the masses. Be it speech or text, most prefer to use Urdu as their medium of communication. A specific area where the rate of frequent communications has increased is the health sector, where, on a daily basis,

Project Title

Bilingual Chatbot

Project Area of Specialization

Artificial Intelligence

Project Summary

Pakistan is a country where English is not the dominant language to be widely used among the masses. Be it speech or text, most prefer to use Urdu as their medium of communication. A specific area where the rate of frequent communications has increased is the health sector, where, on a daily basis, numerous calls to a hospital or clinic are made to inquire about doctors’ availability, appointments’ scheduling and rescheduling and so on. Handling this on calls and in person becomes difficult to manage for the hospital staff and often results in delay to the individual making their inquiries.
This project aims to mitigate the aforementioned problems by giving the user a bilingual health-care chatbot which will work just like a personal hospital/clinic representative, understanding the user’s information needs and answering them just like a real
representative would.

The project will deliver a bilingual chatbot in the form of an application. It will support English and Urdu text and will be primarily for the health sector which implies that queries pertaining to this field will be entertained.

Project Objectives

The project aims to push the current chatbot technology by providing a possibility for the chatbot to understand, detect, and build context for commands and conversations not just in English, but also in Roman Urdu.

It is aimed that the project will be error resistant and responsive of user queries. It is also aimed that the system will correctly manage the conversation’s data in memory and build the require links among its response messages to ensure a smooth conversation flow and experience to the user.

Project Implementation Method

For only data collection purpose we will create a rule based chatbot and filter the data collected by this chatbot after assessing the responses from users and analyzing the words and their frequencies for every query. After the collection of data we will create 4 dynamic neural networks:
1. For communication that holds context of the conversations in Urdu.
2. For communication that holds context of the conversations in English.
3. For handling commands in Urdu.
4. For handling commands in English.
We will create an Urdu dictionary by collecting data from Twitter comments and other available data in Roman Urdu and map all the similar word on a single key to make a standardize Roman Urdu dictionary.
To make this bilingual we will create switching modes that switches from English response to Roman Urdu response and switch the responding neural network on the basis of user’s input. After getting question form the user we will tokenize words and check if all the words exist in English language, if they does then we will generate response in English by activating English neural network model otherwise we will generate response in Roman urdu by activating Roman Urdu neural network model.
We will create a JSON file that contains all the responses, training questions, tags, context_set and context_filter etc. This will provide ease to the user to change the behavior of chatbot by just changing the data in JSON file.

Benefits of the Project

In medical institutions, queries of the general public, patients or relatives of patients are handled either in person or through the phone or email.

The project will provide a much more efficient and less time consuming medium of communication between the public and health institutions whenever their general queries may arise.

The project will also be helpful to the majority population who can not speak or write proper English to communicate with the bot on the internet.

Further research and development on this project will also enable and accelarate research and implementation of Roman Urdu language processing, leading to more robust applications of Roman Urdu detection and general-purpose Roman Urdu chatbots.

The implementation of this project without the use of third-party APIs such as IBM Watson or Google DialogFlow will help ensure that the input and output data collected or generated by the chatbot is not shared either anonymously or unanonymously to any external party besides the user and the hospital.

Technical Details of Final Deliverable

Our final deliverable will be a complete system in which user can communicate with the chatbot through a web interface. The chatbot script will be eployed on a web server along with 4 trained neural network models and JSON files that will carry all the responses, training data and context switching rules. User can ask question in either English or Roman Urdu the response will be generated accordingly by switching between appropriate neural networks.

A training module will also be provided on which user can train models on the basis of new or updated data.

Accuracy test results of models will also be delivered.

Final Deliverable of the Project

Software System

Type of Industry

Health

Technologies

Artificial Intelligence(AI)

Sustainable Development Goals

Industry, Innovation and Infrastructure

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com