This tutorial could also be called “things I wish I had known when I was writing my diploma thesis”. This guideline is intended to ease writing your master's thesis (and hopefully, to produce better theses and successful defenses, as a result) and what better way is there than provide you with a spectrum of information I gained being on both sides of the trench. I'm going to navigate you through the process of your master's thesis assignment, writing and defense. The guideline is mostly fitted to a typical experimental NLP master's thesis (at ÚFAL) but I'm sure you can tweak it to other situation once you get the general idea.

I don't think I can emphasize enough: start EARLY. Seriously. You will make your life much easier if you start programming, measuring and writing (that especially!) in good time. Most opponents can recognize hasty writing hurriedly finished over the last week before deadline. As for the question what is the minimum time in which I can write a thesis, let's pretend I never heard the question.

The exact dates of everything that happens at the Faculty of Mathematics and Physics, such as beginnings and ends of semesters, deadlines for courses and exams enrollments, deadlines for theses submissions and theses exams, is strictly bound by an official Academic Calendar of the faculty. Look out for the exact deadlines of the required actions in the current version.

With that in mind, let's say that a typical master's curriculum takes two academic years (four semesters). Master's thesis writing is officially a three-semester course, consisting of three subsequent (!) semesters each of which you must subscribe to in the Study Information System (SIS), in the appointed deadline. ML: ALE to není oficiámní požadavek a často si studenti zapíší všechny tři kurzy až najednou na konci čtvrtého semestru (což je zejména důležité pro LCT studenty).

That generates roughly the following course of action:

• 1st year winter semester: Starting your first semester as a master's student. Before the end of this semester, start looking around for a possible topic of your thesis. There are two ways to select your topic:
  1. You can go to the module Thesis (Selection of subject) in the Study Information System (SIS) which lists an offer of yet unsubscribed theses. If any of them catches your eye, contact the supervisor by e-mail.
  2. If you have a cool topic in mind, you can also approach a prospective supervisor with your idea and ask if they would be willing to supervise you.
  3. You can also approach a prospective advisor if they would be willing to supervise you and then discuss a potential topic.
• 1st year summer semester: Having agreed with you supervisor, enroll to the first of the three master's thesis courses in the Study Information System (SIS). Most of the communication with your supervisor is conveyed via email, with preferably regular, but sometimes occasional meetings in person (replace person by zoom in covid times). In this semester, you will likely get acquainted with the topic: You will be assigned some scientific papers to read, probably learn new technologies, get access to computation grid, etc. At the end of this semester, your supervisor will grant you credit for the first master's thesis course.
• 2nd year winter semester: Again, don't forget to enroll for the second of the master's thesis courses. In this semester, some serious coding usually happens. If you haven't started in the previous semester, you will probably design experiments, guided by your supervisor. You will implement code for these experiments, run it on the computational grid. If your thesis is more practical one, you may implement an application. This semester is very busy and if your topic was chosen well, it is usually (at least mildly) fun :) At the end of semester, you will be granted credit for the course.
• 2nd year summer semester: Finishing the experiments, polishing your application. But mostly, the writing. I will say more about the actual writing later in the following sections. Again, hopefully, you will be credited and will pass the third and last of the master's thesis courses at the end of this semester.

ML: However, this ideal timetable is not applicable in all case, especially, e.g., for LCT students who spend one year abroad - LCT students typically select the topic at the beginning of their second year, i.e. during October (note that those doing their second year at another university still must find a co-supervisor from Prague.

• End of study: Prior to being admitted to final exams, one has to have successfully cleared any other pending duties of the master's study, such as passed the mandatory exams and collected credits in sufficient amount, according to the specifications of your study programme. The master's programme is finished by taking a final exam, which consists of two parts:
  1. Master's study defense (obhajoba in Czech): requires submitting the electronic (PDF) and the paper (hardcover) version of the thesis in the appointed deadline, enrolling for the exam in the Study Information System (SIS) (again, the in the official deadline) and obviously, showing up for the defence. More about the actual thesis defense later.
  2. Master's state final examinations (státnice in Czech): requires registering for the examination in the required deadline and showing up for the examination. Also more on the examination later.

So each of both parts is offered separately in the Study Information System (SIS), each of them must be enrolled in the required deadline and they can be taken jointly or separately in three possible terms throughout the year: summer (June), autumn (September) and winter (February). Please note that the deadline for registration differs for each of the terms. There is a preferred way, when everything went well, and that is the summer final examination after the 2nd year's summer semester, taking both the defense and examination, which places the registration for both examinations and the submission of the thesis somewhere to May.

“One thing I would like to do better next time (and advise students to do better in their first time) is time management: The programming part should be finished in a half time assigned to bachelor's thesis (even in case it is the main part of the master's thesis). In the end, it will delay and there will be a lot to correct and improve. The experimenting, evaluation, text writing and corrections (corrections!!!) will take more time than one expects. And even if it is not the case - then it's great, at least one is not stressed out from submitting at the last time and will have some spare time.”'

The very fist question I usually get is “How much do I have to write?”. As far as I know, there is no official number anywehere, so I'm just going to speak from experience: An experimental NLP master's thesis most probably non-rejectable on the basis of “too few text” if it consists of absolute minimum of 40-50 A4 pages of content, not including the front page, acknowledgements, table of contents, bibliography & appendices. That is, the Conclusion Section should appear on page 40-50.

Most scientific contributions were achieved in a team and it is therefore customary to write publications in plural, as in “We present”, “We implemented” and “We conclude that…”. If you are a single author of the work, though, opinions differ on the choice of singular (“I”) and plural (“We”). Some even say a third person should be used as in “The researcher found out that…” It is generally advisable to avoid the singular/plural choice wherever you can by using impersonal passive voice, such as “The research shows that…”, “The findings suggest that…”. The active voice singular (“I”) would then be kept for sentences in which you underscore your particular contribution: “I implemented”, “I measured”. Even the use of “We” throughout the thesis in all places where you are clearly referring to you as a single author, is acceptable. Using a plural “we” instead of singular “I” in scientific writings is supposedly called ''pluralis auctoris'' and Wikipedia says it is more common towards the East (hence the Wiki page is only in a few languages, including Czech, but excluding English). That being said, the choice is really yours.

An typical experimental NLP thesis usually consists of these parts:

• Oficially required formalities such as front page, declaration of independent work, dedication, abstract, etc.
• Introduction
• Theoretical Background/Related Work
• Methodology
• Results
• Discussion (sometimes merged with Results)
• Conclusions
• Bibliography
• Attachments

I describe each section in detail on a separate page.

This is an area which you can get right and get some plus points for very low cost. Correct typesetting and formatting can be sitted through with a little of diligence and patience even if one is no Einstein. You won't get complaints for your thesis not being rocket science, but you can get a lot of complaints for poor presentation of your work. All of this can be avoided, if you:

• Read the official Guidelines for Writing a Master Thesis.
• Download the official LaTeX template and take this as an opportunity to learn LaTeX (or brush up your LaTeX skills). Please please be it LaTeX and not a WYSIWYG editor (Word, Libreoffice, …). It's just so easy to get beautifully typesetted text and mathematical formulas with LaTeX and so easy to ruin things otherwise. Also, get BibTeX for managing your references. You'll find yourself having to resist less urge to through it all away when you have to reindex your references in Word for the 50th time. Your future self will thank your current self if you put just a little work into setting up your thesis in LaTeX.
• Read my advice on typesetting, with good practices and frequent errors.

(Petricek, 2016)¹⁾

Plagiarism is a big NO-NO in science. It's so big you can get in serious trouble if caught. The trouble is, many people are not quite sure what plagiarism is and what it isn't. Sometimes, there is a grey area, but sometimes, the borderline is pretty clear.

Generally, plagiarism is presenting someone else's work/idea/text/source code as your own. Specifically, anything that appears in your thesis and is not referenced or is not general knowledge, is being presented by you as your work/idea/text/source code, unless properly referenced. Some very obvious examples of things that should be referenced are:

• big inventions with names: People who invented them should get credit. The first time you mention anything with a name, you should consider adding a reference. E.g., HMM, NNs, RNNs, LSTM, word embeddings (word2vec, FastText), contextualized word embeddings (BERT, …), etc.
• general ideas without names: The first attempts at solving some problem in a way which is similar to yours, or not similar, but was simply the first.
• copied material: Any time you copy anything more than a sentence, it should be referenced. Also definitions, adopted figures and tables.
• related publications: publications closely related to your work. Many master theses are incremental changes to one influential publication - which, obviously, must be referenced.
• source code: Give credit to those who implemented source code you use or build on. It is not necessary to reference Python, but one should reference specialized software (like TensorFlow) and especially the source code you modified. Most source code authors give exact instructions in README about how they wish to be referenced. Usually, by citing a publication.
• data: It is highly unlikely you did not need any data for your thesis. If you did, the data must be referenced. Most datasets are described in an accompanying publication. Some (older ones) are not, then at least give the url in footnote.

Every time you copy/reproduce a sentence/definition/figure/table, the reference must be repeated. It is not OK to state in the beginning of the section “And from now on, I shall draw from publication XY” and then go on three pages freely mixing your text and sentences from XY. With some exceptions, it is also not OK to copy entire paragraphs or pages, even if they are properly referenced, as you are supposed to write you own thesis, not copy someone else's work. It is however allright if you introduce/explain an idea with a proper reference and then discuss the idea in the next three paragraphs without referencing it again and again. If your discussion, however, replicates someone else's opinion, you should reference it, e.g. “Our data support hypothesis ABC, as well as results of REF-to-XY.”

Less obvious and not always necessary references may be required by:

• statements like “It is (generally) considered that …”, “It is (generally) thought that …” are tricky. By whom it is considered? Ideally they would be supported by references to publications.

In scientific writing, there is a rigid, accustomed way of proper referencing. TODO write about reference norms.

Don't fall in the PDF/A trap! The electronic system requires all submitted PDF in PDF/A format and there is an automatic check for PDF/A. Allow yourself enough time to find out how to convert your PDF into PDF/A (like, not in the last two hours before deadline midnight). Some good advice on PDF/A, unfortunately only in Czech, can be found here.

As has already been said, the final examination consists of two parts:

1. Master's study defence (obhajoba in Czech): requires submitting the electronic (PDF) and the paper (hardcover) version of the thesis in the appointed deadline, enrolling for the exam in the study information system (SIS) (again, the in the official deadline) and obviously, showing up for the defence. More about the actual thesis defence later.
2. Master's state final examinations (státnice in Czech): requires registering for the examination in the required deadline and showing up for the examination. Also more on the examination later.

So each of both parts is offered separately in the study information system (SIS), each of them must be enrolled in the required deadline and they can be taken jointly or separately in three possible terms throughout the year: summer (June), autumn (September) and winter (February). Please note that the deadline for registration differs for each of the terms. Actually, there is a preferred way, when everything went well, and that is the summer final examination after the 2nd year's summer semester, taking both the defence and examination, placing the registration for both examinations and the submission of the thesis somewhere to May.

In fact, the defence and the final state examination are usually organized in two different days with span about a week. One day accommodates all defences, the other on all final state examinations. The reason for this is purely organizational: different committees have to meet for each event.