University of Leeds Featured PhD Programmes
Catalysis Hub Featured PhD Programmes
University of Liverpool Featured PhD Programmes
University of Kent Featured PhD Programmes
Bournemouth University Featured PhD Programmes

Deep Learning for Procedural Natural Language Understanding

This project is no longer listed on and may not be available.

Click here to search for PhD studentship opportunities
  • Full or part time
    Dr D Bollegala
  • Application Deadline
    No more applications being accepted
  • Funded PhD Project (Students Worldwide)
    Funded PhD Project (Students Worldwide)

Project Description

Applications are invited for a PhD scholarship in "Deep Learning of Procedural Natural Language Understanding" at the University of Liverpool.

The position will be based in University of Liverpool’s Department of Computer Science and jointly supervised by Dr Danushka Bollegala and Prof Katie Atkinson.

About the project:
Procedural texts contain a sequence of steps that must be followed in a specific order to achieve a defined goal. Some examples of procedural texts are (a) recipes for preparing a particular dish, (b) a user manual for a software or a machine describing the steps required to achieve a particular task, and (c) legal procedures that require parties to follow a set of laws to execute a case. In this PhD project, we will consider the problem of learning the optimal policy for achieving the desired final goal according to some objective function such as minimising the number of steps required or the total cost involved in the process.

Learning from procedural texts has so far been a challenging problem due to several reasons. First, unlike in information retrieval where a document is modelled as a bag-of-words ignoring the ordering of sentences/words in a document, the ordering of information is critical for procedural texts. For example, you will not necessarily get the same dish if you followed any random ordering of the instructions in a recipe. Second, the outcome is unknown until we have followed all steps in the procedure. Therefore, supervision feedback is received only at the end of the process and is not always available for the intermediate steps. For example, in a legal case, you will only receive the judgement at the end of the hearing. In this PhD project, we will explore deep reinforcement learning methods coupled with semantic representation learning to overcome these challenges.

In this PhD project, we will study several challenging research questions and develop solutions such as
(a) How can we accurately represent procedural texts?,
(b) What machine learning techniques are best suited for learning from procedural texts?,
(c) How accurately can we predict the outcome resulting by following the steps in a procedural text?
(d) Can we automatically identify the most critical steps (path) in a procedural text?

We will explore state-of-the-art machine learning techniques including deep reinforcement learning methods, structure prediction, and learning representations.

Prior experience:
Candidates must have a good first degree in Computer Science, or closely related subject, and have excellent programming skills. Some prior experience of natural language processing or machine learning is highly desirable. Previous experience of carrying out research would be a distinct advantage though it not essential.

Applications can be made by following the University of Liverpool’s standard process, details of which are available here:

Applications should list Dr Danushka Bollegala as the potential supervisor and choose the option "Department funded PhD" when asked how you will fund the PhD.

Applications should also be clearly marked as being for the scholarship in "Deep Learning of Procedural Natural Language Understanding"

Funding Notes

The scholarship will be for 3 years at GBP 20,000 per year.
a) If UK/EU this will relate to Full Fees & Maintenance (current fee £4,195)
b) If overseas/international applicants this will relate to Full Fee and a small maintenance (current fee is £18,900).

FindAPhD. Copyright 2005-2019
All rights reserved.