Notes on Essays and Dissertations

How to write an essay

An essay is not just a report, it's an argument. A good essay describes a thesis, presents evidence for and against the thesis and analysis of that evidence.   Analysis of evidence should show how the evidence relates to the thesis as well as pointing out any points that aren't fully supported.  Science is not meant to be propaganda --- a good scientific essay mentions under what conditions the thesis may be wrong, how probable these conditions are, and how they could be recognised.   Most essays end with a conclusion which is a reiteration of the thesis with a summary of the argument to support it.

Every paragraph of an essay is itself a smaller essay --- each one should have its own point. Each sentence should contribute to the paragraph, and each paragraph should contribute to the essay.

Many students develop a bad model of essay writing in school: "First I will learn, then I will tell you what I learned.  I will read something, then summarise or report on what I read."  Good writing is much more than this.   The process of writing  requires organising and communicating your thoughts --- you are your own first reader, and first critic.   Writing, like any kind of teaching or argument, helps you understand the material you are presenting better. Often the process of writing will show you holes in your reasoning, beliefs that haven't been fully supported. These holes lead you to further thought and further research.

How to write a dissertation

A dissertation is more of an essay than a report.  Dissertations contain reports, but their fundamental structure is that of an essay.  They should have a thesis which is being demonstrated.  The reports come in describing extensive evidence and the means by which it was gathered.

Dissertations which describe software development projects are still essays.  Their theses might take the form "An XXX needed to be built, and I have succeeded (or partially succeeded) in building it."  Both halves of this thesis need to be supported: the motivation for the project must be described, and evidence must be presented that you have met the criteria of your project.

The target audience for a dissertation is not just your supervisor --- it is your peers.  Most readers of a dissertation will be other students, other developers or other scientists who are interested in at least some of the areas covered in your dissertation.  Some of these readers will only be interested in one or two paragraphs you wrote about another paper or project.  Some of them will be doing a similar project and will want to understand the decision process that lead to the path your project took.  Some may be trying to replicate, test, extend or build from your work.  These readers will need a full report of your methods, assumptions and results.

How to write a research article

Research articles are a little like dissertations, but they have to be direct, to one point (if you have more points, write more papers) and formulaic. The reason research articles should be formulaic is because scientists need to read many of them, so it should be easy to find exactly what you need to read. The one point of a research article is like the thesis of an essay.

As with a dissertation, it is critical in an article that you provide information so someone else can replicate your work and compare your results to their own. This belongs in a section called Methods. All results and the statistics used to analyse them should be in a Results section, which comes after Methods. Either the Methods section or the introduction to the results section should also explain exactly what experiments you have run and why. The bulk of the Results section describes the procedures and outcomes of usually several experiments precisely. After the Results section is a Discussion section, which describes the implications of your research, including what your results imply about other people's work. This is the one place in the paper where you are permitted to speculate a little, but any speculation should be well supported. The Conclusion is only a summary of the main points of the paper, possibly including a summary of your intended future work. No novel ideas should be introduced in that section.

I have started at the heart of the article, but it is worth mentioning that articles must start with very specific titles that make their contents clear, good concise abstracts, an Introduction that motivates the research by explaining the problem thoroughly and how your research addresses it. There is often next a Background section which goes into more detail on the existing state of the art (see further below the Literature Review), though this information may be just included in the Introduction.

Components of a dissertation

A dissertation needs all of the following components.
Notice that component does not mean chapter. These things may not deserve entire chapters to themselves. Further, the argument, not the components, should dictate how you structure your dissertation. But every dissertation should have all the above components somewhere.

Literature Review

The Review of Literature is a very important part of a dissertation.  Again, many people who read your dissertation may be most interested in your literature review.  Once you have spent months (or years) on a topic, you become an expert.  This makes your perspective on the rest of the field around your own work particularly valuable.
The Literature Review is not listed as a component because it is an integral part of your evidence, and it may potentially be in many different locations.  For example, you would expect some literature review in the Background and in the Implications.   Normally there will be a phase of heavy reading toward the beginning of your research project, then a period of heavy coding, and finally another burst of reading during the final write up.

When and why to reference

Science works much the same way as evolution. Each piece of work inherits most of its bulk from the good, proven ideas in the existing literature, but each experiment adds some novel variation, some new ideas, and then tests those new ideas against the old. A scientific journal article needs to show that you know and understand the existing literature, and that you understand and can prove your contribution to it. The way you show that you know the existing literature is not only by listing it in the back, but by referencing appropriate citations in the text after any claim that relate to previous knowledge. This allows you both to validate your claims (evidence from other people's experiments can be just as central to your arguments as evidence from your own), and also lets the reviewer (and later, the reader) know that you are up-to-date. Often when you read someone's paper, you might think (for example) "Ah, if they believe that, they haven't seen A's newest work." So, if you are contradicting A, you have to say "This is in contrast to A (2005), who is wrong because..." Another common thing a reader might think is that "B has already done this." If it is true B did (and published) something exactly like you did before you did it, then you work won't be worthy of publication. So if you know B's work is like your own, you need to say something like "This is similar to B (1993), but we have made an improvement in this way..." One exception to this rule is replication, that is, when you deliberately set out to test a published result that is not yet widely accepted. In this case it is perfectly allowable to say, "This provides further evidence that B is correct, and our study provides important corroboration for B's theory." This is especially useful if someone else has published saying B is wrong and you can cite them as well. Or if in attempting your replication you discover that B didn't really make it very clear how to achieve their result, then you can report the extra information. There is always a chance you will find out that B made a mistake, and this can be very publishable, but you also have to be very, very careful that you are right before you say that someone else is wrong.

Structure of a project and its dissertation

Best practise for completing a dissertation often will have reading, writing and coding occurring throughout your project.  Simple prototypes or pilot studies should be done very early in the project, because they will help you understand the material you are reading.   This will also help you get a grasp on the difficulty of the problem you are addressing, which allows you to adjust your schedule and expectations accordingly.  Similarly, when during coding or experimenting you encounter difficulties or make discoveries, then you should take time to search the literature for more information on your new problem or perspective.  During the final writing phase, nearly all students discover that they cannot support all of their assumptions or claims, and therefore need to do more reading, run more experiments, or run further benchmarks.  Although some of this should simply be documented in the Future Work section, it is best to leave some time for doing this research.

As I said earlier, the components above do not necessarily correspond directly to chapters.  Often Motivations are included in the Introductory chapter, or Implications are included with either after Results, or with Future Work in a chapter called Discussion.  On the other hand, there might be several chapters covering different phases of the project (e.g. 3 different versions of a program) each of which contain Methods, Results and Implications.   Background might be split between a chapter explaining Motivations and one explaining Implications.  Many examples of dissertations are available from the front office, from your supervisor, or on the Internet.

More general writing advice:

  • Dissertations are formal. Avoid being overly familiar or the excessive use of colour.
  • At the same time, they should be clear and readable.
  • Use the first person where necessary for clarity. Example: "Most researchers consider sleep to be healthy. My own experiments, described in Chapter 3, indicate..."
  • Avoid passive voice. Example: "Systems were built which caused fires (Bain 1986)." Better: "Bain (1986) documents a fire caused by a poorly designed system."
  • Be careful about tense agreement and, more generally, about parallel structure.  Example: "Most developers like editors to be colourful but haven't paid for software that's too expensive."  Better: "Most developers like colourful editors but not expensive software." or "Historically, developers have chosen colourful editors, but have avoided expensive software."
  • Avoid "is that" constructions Example: "The reason I chose potatoes is that..."  Better: "I chose potatoes because..."  
  • Avoid redundancy, but not at the cost of comprehensibility. 
  • Avoid jargon and acronyms, unless they are important.  In that case, define them on first use.  Positive Example:  "One of the most important parts of artificial intelligence (AI) is action selection.  Action selection is the ongoing problem faced by any intelligent system of deciding what to do next."
  • If you want to write very good British English (not necessary!) look at this style guide.
  • Be sure to proofread! 
  • Be sure to spell check!
  • Be sure to find someone else who will proofread for you (offer to proofread for them!)
  • Principles:

  • Be clear about what you are saying. Separate facts, evidence, opinions, theories, and tautologies (definitions), and identify them for what they are. Be sure to identify your own opinions separately from your sources'; always cite when you use something from an external source.
  • Analyses are suggestions, conclusions should be established facts. This is one reason why there are Discussion sections in scientific papers --- to separate what's been claimed from what's been proved.  If you are writing a literature review, conclusions must usually be about the review itself, not about its contents.  For example, you might conclude that the information you reviewed tended to indicate X.
  • Think critically -- this doesn't mean to be negative ("criticise") but to examine information and theories for internal consistency, logic, strength of support. Be sure your conclusions necessarily follow from your premises, otherwise they aren't conclusions.
  • Good writing is part of what makes science and engineering work. Everyone has different information acquired over their lifetimes, so anyone can make a critical discovery of a hole in a literature, or a relation between two previously unconnected ideas or areas of work.

    page author: Joanna Bryson
    last updated: 29 December 2005