Natural Language Processing and Artificial Intelligence

 

Challenges

Imagine visiting the web-site of a bank and trying to transfer money from one account to another - not every site makes this easy to do, especially say if forex transactions are involved. Instead, imagine sending an email to the bank with the same request; a response comes back either confirming the action or asking a few extra questions - one may not even notice that it could be a computer generated response. Again there are alternative menu-driven interfaces that are a standard mechanism to interact with business systems, such interfaces can become rather cumbersome for users of large applications, especially for users who know the kind of data they want from the system, but do not know which menus to traverse in order to get the data. The Natural Language Processing team is developing systems that allow the users to easily interact with the business systems using the power of Natural Language.

Artificial Intelligence
Traditional AI involves humans in knowledge capture mode followed by automated reasoning using the captured knowledge represented in some form (logic, rules, Bayesian/neural networks, etc.) Many problems require multiple reasoning techniques to work together since knowledge is widespread and cannot be all collated. Further, building knowledge-based systems for data-intensive domains can be rather complex as voluminous data needs to be processed and reasoning on each data item can be rather time-consuming. At times in such domains, some data items by themselves may mean nothing unless they are analyzed in a larger context. Can systems and humans work jointly to solve a problem by contributing parts of a solution? So the problem which we are interested in can be stated as follows: Given a situation where multiple modes of reasoning and multiple modes of collaboration between human agents need to work together in order to solve a problem, what is the formal framework for the reasoning engines and the human agents to collaborate and produce effective solutions?

 

Solution Overview

The Natural Language Processing and Artificial Intelligence research in TCS currently focuses on natural-language based interfaces with business applications, reasoning with uncertainty in data, and multi-modal reasoning for data-intensive domains.

There is substantial expertise in knowledge engineering, knowledge-based systems and reasoning engines. Several real-world systems have been developed and implemented, such as decision support systems, scheduling systems, planning systems and rule and case based inference engines. We conduct research in areas such as multi-modal reasoning, natural language interfaces, knowledge-base verification and validation, rule-based reasoning, knowledge engineering, and Bayesian reasoning. TCS has over 300 man-years of experience in building and integrating AI systems.

We are creating a framework to build conversational interactions into applications, exploiting AI techniques as well as semantic computing technologies. The current focus is on building natural language interfaces to menu-driven business application systems. Architecture of a text-based natural language conversational interface, called NATAS, has been developed for menu-driven systems. NATAS permits the user to carry out a dialog with the system in order to fetch relevant data and carry out various tasks of the system. The architecture uses semantic web based ontology of the domain, to aid in the retrieval of the relevant data and concepts from the system. The idea is to carry out a conversation with the user, in order to drill down to what the user actually wants and then identify the application task(s) that would carry out the user’s requirement. Thus, instead of the user going through the process of translating his requirement into the exact set of system commands, in a conversational interface, a user specifies at a high-level the task that needs to be done and then the system carries out a dialog to map it to the exact set of commands of the system. The natural language interface interprets the text and calls appropriate APIs of the application to accomplish the requested tasks. The main advantage of such a system is that the user is free to enter any information that he has, in the raw form. It is the job of the system to process that and get whatever else is required. One way to tackle this, is to abstract the voluminous data to a higher level for effective reasoning to take place. The domains of surveillance and situation assessment in various possible application scenarios are of interest. We are investigating the mechanisms for multimodal reasoning for such domains.

Natural Language Interface to Business Applications