Principles of Scientific Communication Writing Scientific Papers PDF

Principles of Scientific Communication Writing Scientific Papers Michel Verleysen Ecole Polytechnique Louvain Université catholique de Louvain, Belgium...

Principles of Scientific Communication Writing Scientific Papers Michel Verleysen Ecole Polytechnique Louvain Université catholique de Louvain, Belgium [email protected] Slides based on material from Axel Van Lamsweerde, Patrick Bertrand, Yves Deville and Xavier Gonze: thanks to them! 1  Who I am I am not a professional of technical communication techniques I aim at sharing principles & experience accumulated as... Researcher Editor-in chief of journal Conference program chair Program committee member Presenter of many technical talks Teacher of many courses 2 Forms of scientific communication Many forms of scientific communication Writing scientific papers Giving technical talks Designing posters But also Writing reviews Interacting during conferences Writing for the Web Writing emails … 3  Writing Scientific Papers Outline General principles Improving paper structure Improving paper clarity Improving style The beginner’s 7 sins 4  General principles Publish… why? Publish… where? Publish… who? for whom? Publish… how? - the paper lifecycle Publish… what? Publish… when? 5  Publish… why? To communicate new findings, new knowledge publication = ultimate result of scientific research research work is never finished until it is published To let the community know about your work ⇒ recognition ⇒ contacts, fruitful collaborations To get useful feedback from peers external, independent anonymous ⇒ frank To embellish your CV (+ CV of colleagues) 6  Publish… where? International journal different quality standards, selectiveness & impact different types of paper research articles, letters, surveys, “Comments on”, magazine articles International conference proceedings different quality standards, selectiveness & impact research paper, experience report, (abstract) 7  Publish… where? (2) Journal vs. conference proceedings journal... more impact (especially long-term impact) more highly rated by selection/promotion committees (much) deeper reviews more space wider target audience (usually) fast-track special issues proceedings... faster process direct contacts & discussions + community awareness sometimes more selective selection of best papers for journal 8  Publish… where? (3) Journal vs. proceedings: choice may depend on specific culture (check yours!) not necessarily exclusive: expanded version of conference paper can be submitted to journal (with spec of differences) Avoid poor-quality, low-impact journals/conferences they reputedly need papers, lack serious reviewing process, are NOT selective low impact on your CV too Check impact factor within your area 9  Publish… where? (4) To decide which conference, check... “Submission topics” of Call for Papers Who is in the Program Committee (appropriate reviewers for your topic?) For good selective confs Make sure the ratio NumberOfAcceptedPapers / NumberOfSubmissions is mentioned in your pub list usually available in PC chair’s foreword in proceedings 10  Publish... who? Every author is responsible for entire content of paper Each author should have contributed in some way Order of authors normally ± reflects weight of contribution... in producing results in writing paper Order may depend on specific cultures (check yours!) Every author must be aware of being an author (!!) Set of authors should be invariant throughout the review process (to avoid conflict-of-interest problems) In case of doubt, discuss issue with authors & colleagues 11  Publish... for whom? Identify your audience; identify who the reader will be, what her background is (selection of the appropriate journal) Write for the reader (or the reviewer), not for you ! Paper = pedagogical explanation of results “you and me together” 12  Publish... for whom? (2) Imagine yourself as the reader Ask yourself questions... is this interesting ? is this comprehensible here ? is this relevant ? what questions are coming to reader’s mind ? Do not speak highly of yourself / your work … leave it to the reader to do that 13  Publish… how? Submission + cover letter (journal) Contact information for corresponding author Relation to your other papers (when necessary) Wait for editorial/PC decision Study reviews do not blame reviewers Revise accordingly Resubmit detailed response to reviewers on how revision addresses concerns If paper accepted correct proofs rapidly (always typo left) (Fill and send back the copyright transfer form) 14  Submit... when? Not too soon... and not too late... Not too often (unless you are genious) ? Advices: Refrain from submitting half-baked ideas (keep them for workshops) Refrain from submitting below the LPU treshold (LPU = Least Publishable Unit) 15  Publish… what? Two types of research contributions Invention of model, method, technique, tool to... Develop, (re-)structure, reuse Analyze Evaluate Measure Understand... Artifact or process Experiment-based discovery of phenomenon, law, structure,... about artifact or process 16  Publish… what? (2) What? versus So what? What? The information of the paper What did you find in your research? So what? The message of the paper What do your findings mean to your audience? 17  Publish… what? (2) Evaluation criteria for research papers: Original contribution Significant... Problem and/or solution Domain-specific qualities: useful, scalable , … Sound results replicability High-quality presentation (we may help only here !) 18  Originality Implications on presentation Specify your objectives & contribution carefully [ abstract, intro, conclusion ] Compare with related work carefully [ paper introduction, special section ] Do not publish the same result twice (exception from conference to journal) Implement your objectives carefully [ paper body ] 19  Significance Implications on presentation Discuss why your problem is significant [ abstract, introduction, conclusion ] Discuss why your solution/result is significant [ introduction, discussion section, conclusion ] What it may be useful for Why/how it replicates, scales up How it applies to other situations … 20  Soundness Implications on presentation Make the paper technically readable sufficient detail & precision to make results verifiable, replicable For experiment-based research papers describe the experimental method carefully so that it can be assessed & replicated do not mix results (data) & their interpretation 21  Good presentation High cohesion One paper, one result Don’t try to say too much …... but don’t try to say too little (cf. LPU problem: Least Publishable Unit) Self-contained paper put anything needed to understand results [ background section ] Crystal-clear presentation no literature nor poetry no “Agatha Christie” sort of writing 22  Good presentation (2) Good structure is essential… Goal-oriented structure (tree) objective ← sub-objective Rigid, generic structure for experimental papers Introduction Method Results and Discussion 23  Good presentation (3) Some golden rules... Say what you are going to do before doing it Avoid mere description of work done no “we-did-this” paper Avoid the standard mistakes of novice writers (cf. below) 24  General principles: conclusion Publish... original, significant, sound work well presented in good journals & conferences with good people at good time Repeated publication of weak papers may severely damage your reputation… 25  Writing Scientific Papers Outline General principles Improving paper structure Improving paper clarity Improving style The beginner’s 7 sins 26  Avoid flat structure Paper ≠ sequence of descriptive statement Paper = tree structure goals (logical sections) subgoals (logical subsections) present top-down (explanation), not bottom-up (discovery) 27  Use one and only one structuring scheme Avoid unordered structure within section section ≠ unordered list of items different ordering schemes: by importance, naturalness, diagram, process steps, … use one and only one ordering scheme in the same section Example (bad) 1. Step 1:... (step description)... 2. Step 2:... (step description).... 3. Section heading... blah blah... 28 Reader should always know what’s going on Why are we here? Where are we coming from? Where are we going to? ⇓ Present logical structure [ intro ] Say what you are going to do before doing it Make current context obvious [ transition between sections ] 29  Example Say what you are going to do before doing it Example (Bad) 4. Synthesizing attack graphs 4.1. Computing agent operations... blah blah... 4.2. Computing single goal decompositions... blah blah... 4.3. Computing entire refinement trees... blah blah... 4.4. Synthesis as agent weakening... blah blah... 30  Example Say what you are going to do before doing it Example (Good) 4. Synthesizing attack graphs To synthesize an attack, we need to... (Section 4.1), then to... (Section 4.2)... Next,... (Section 4.3). Finally, Section 4.4. shows that.... 4.1. Computing agent operations... blah blah... 4.2. Computing single goal decompositions... blah blah... 4.3. Computing entire refinement trees... blah blah... 4.4. Synthesis as agent weakening... blah blah... 31  Overall paper structure Title, authors Abstract (critical!) Keywords & phrases Introduction (critical!) (Background section) Paper body (Related Work section) Conclusion (critical!) Acknowledgement References (Appendices) 32  Title Critical because... used by many, many potential readers ⇒ decision to read abstract (and then paper?) or not used by search engines for paper retrieval inappropriate title may miss target readership Tip: short but informative Not too short: need for specificity Not too long: ≠ abstract Avoid content-free info, word wasting Avoid abbreviations, jargon, ref to other paper 33  Authors Authors take intellectual responsibility of paper results Sensitive issue - source of arguments & conflicts Each author should have made important contributions Order of names normally reflects weight of contribution in producing results, in writing paper but community-specific Everyone listed should know she is listed as an author Tip: in case of doubt/problem, discuss it among the authors (and other colleagues) 34  Abstract Critical because... used by editor to select appropriate reviewers used by reviewers as entry point Þ very first opinion used by readers ⇒ decision to read or not used by review journals to publicize paper used by search engines for paper retrieval free access in the web pages of the journal Goal: give a sense of what paper is about 35  Abstract (2) Conflicting requirements... comprehensible self-contained (published independently) short (» 250 words) precise, to the point convincing ⇒ best compromise needed Write it over and over again ! 36  Abstract (3) Typical structure Introduce context & motivation (1 paragraph) what is the problem, why it is important Summarize contribution (1-2 paragraphs) what is the solution/result, why it is important how was the result obtained (approach, method) what are the implications Zooming from general to specifics 37  Abstract (4) Tips… cf. phone call analogy (or lift conversation)... Provide substance, no noise No salespeech Avoid non-standard abbreviations Avoid references to literature NO indication of the structure of the paper 38  Abstract (5) Tips … The opening para should be your best para its first sentence your best sentence [Knuth] if paper starts bad: reader gets resigned to fight with your prose if paper starts smoothly: reader gets hooked Example E.g. worst way to start: “An X is Y” Bad: “An effective technique for X is to Y” “A well-known problem is to X” Good: “Y is an effective technique for X, because…” 39  Keywords & phrases Used by bibliographic systems for paper classification & retrieval even sometimes for reviewer selection ! Requirements: suggestive accurate standard, commonly used terms Common ordering: from specific to general 40  Introduction Critical because reviewers & readers … make their opinion from it (especially if little time available for reading paper) generally, get back to it Goal: give a more extensive sense of what paper is about 41  Introduction (2) Requirements... Comprehensible (if possible: to broader audience) Flowing smoothly, zooming from general to specifics Convincing No claim left unsubstantiated in the paper No duplication of sentences from abstract Get regularly back to intro for revision ! 42  Introduction (3) Typical GOOD structure 1. Background picture Context of the problem addressed Overview of relevant work in the area (incl. yours if relevant) Definition of the problem, motivation for addressing it (possible analogies with similar problems) 2. Objective Purpose of the paper 43  Introduction (4) Typical GOOD structure 1. Background picture 2. Objective 3. Overview Basic ideas on how objective is achieved Contribution, main results, important messages Assumptions, range of applicability Approach followed Implication of results 4. Organization Brief summary of content of each subsequent section to make logical architecture of paper visible 44  Background (if needed) Goal Make paper self-contained Define what you will be using throughout the paper ⇓ Summary of previous results on which paper is based Definition of basic concepts, global notations, special terms, abbreviations, etc. 45  Body Goal-driven ⇒ tree structure goals/subgoals ⇒ logical sections/subsections reasonable granularity reasonable depth (2-3 levels) bad 3.4.2.5. Title of a section (... followed by 3-line subsection...) 46  Body (2) For each section: Transition para [if not done at end of previous section] start with brief context restoration: what has been done, what remains to be done Specification of objective of this section [ If needed: definition of local terminology & notations ] Material to achieve current objective → introduction of finer subobjectives Transition para [if not done at beginning of next section] end with brief context restoration: what has been done, what remains to be done 47  Related Work Goal Precise, comparative discussion of related results WHAT-WHO, respective pros/cons of your/their contribution At much greater level of detail than Introduction Tips Show personal synthesis : do never copy from other papers ! Aggregate logically by topic: avoid rambling among unrelated papers ! Be comprehensive, accurate Be honest in comparative evaluation --reviewers WILL read it ! 48  Conclusion Critical because reviewers & readers … May read it more carefully than paper body (especially if little time available for reading paper) Generally, get back to it Goal: provide final picture 49  Conclusion (2) Requirements: cf. Introduction... comprehensible (if possible: to broader audience) flowing smoothly --from recap to new perspectives honest assessment no unsubstantiated claim, no overgeneralization or extrapolation to bigger picture no salespeech no duplication of sentences from abstract or intro Get regularly back to conclusion for revision ! 50  Conclusion (3) Tips Mark of good summary is revelation [van Leunen]... “Remember this, reader? And that? Well, here’s how they fit together” 51  Conclusion (4) Typical GOOD structure 1. Summary of paper Key contribution, results, ideas, messages... Why they are significant 2. Perspective More specific comparison with relevant work [if no Related Work section] 3. Critical assessment The pros: strengths of results, benefits The cons: weaknesses of results, limitations  4. Open issues & future work 52  Acknowledgement Goal Ack grant, fellowship, funding source (often required by contract) Thank people advisors, colleagues, technicians, programmers, … who contributed in some way seminal ideas, guidance, suggestions during work, inspiring discussions, comments on drafts, help in experiment, pointers to literature, … Tips Be fair but reasonable… and not embarrassing ! Specificity is best cure for clichés 53  References Goal For reader: opens door to further information independent judgement For writer: keeps you honest shows your contribution Tips Be meticulous & comprehensive… but use all refs listed Avoid secondary or not publically available material Meet journal/conference requirements if permitted: use informative identifiers e.g. [Ver99] 54  Appendices Goal: to prevent... Readers from being lost in subsidiary details Smooth paper flowing from being disrupted Paper structure from getting hidden Typical use Technical details such as proofs of theorem, algorithms, experimental results Algorithms (if this is not THE contribution) Highly detailed figures or tables 55  Structuring experimental papers: the IMRaD structure Specialization of what was presented so far IMRaD = rigid structure, easy roadmap for authors – editors – reviewers – readers - What question/problem was studied ? Answer = Introduction - How was the problem studied ? Answer = Methods - What were the results ? Answer = Results - What do the findings mean ? Answer = Discussion 56  IMRaD: the Introduction section Goal Why was the study undertaken? What was the research question, the tested hypothesis or the purpose of the research? 57  IMRaD: the Methods section Goal When, where, and how was the study done? What materials were used or who was included in the study groups ? Describe & justify the experimental design to make the experiments repeatable by peers reproducibility = basis of Science Rule Must provide enough details (otherwise rejected outright by reviewers) If new, unpublished method: give all needed details 58  IMRaD: the Methods section Tip chronological presentation (with sub headings) Cf. cookbook recipes : How ? How much? No results description in this section yet ! 59  IMRaD: the Results section Goal What answer was found to the research question? What did the study find? Was the tested hypothesis true? Rule Core of the paper Presentation of the data, but predigested: only representative data, not all ”The fool collects facts, the wise selects them" No method description anymore ! Not interpretation of data yet ! (next section) No references 60  IMRaD: the Results section (2) Tips Be crystal clear: paper will stand or fall on this section If n variables tested... show in Table or Graphs only those affecting the reaction for the others: say you did not find out under the experimental conditions absence of evidence is not evidence of absence Avoid redundancy Most common fault : in-text repetition of what is clear in Figures or Table Help the reader look at the important features Avoid verbiage Bad : It is clearly shown in Figure X that … 61  IMRAD: the Discussion section Goal What might the answer imply and why does it matter? How does it fit in with what other researchers have found? What are the perspectives for future research? Show the relationships among observed facts, the meaning of results Harder part to define & to write another cause of rejection ! Often, far too long 62  IMRAD: the Discussion section (2) Tips Try to present the principles, relationships, generalization shown by the results not a reformulation or recap of the results ! Point out any exception or any lack of correlation Define unsettled points Show how your results and interpretations agree (or contrast) with previously published work Don't be shy - discuss the theoretical implications of your work as well as any possible practical application State your conclusions as clear as possible Summarize your evidence for each conclusion 63  Writing Scientific Papers Outline General principles Improving paper structure Improving paper clarity Improving style The beginner’s 7 sins 64  Improving paper clarity The golden rule Avoid unnecessary jargon, formalism, details,... Specifics Headings Transitions Examples Figures Metaphors Definitions Notations, formulas Formatting Programs, proofs, experiments 65  Headings Requirements... Concise Specific Suggestive Beware typos… Example bad: “Some tips for making the content of this paper more clear” good: “Improving paper clarity” 66  Transitions Sometimes needed within section Goal: take care of reader What does she know so far ? What should be expected next & why ? 67  Figures Goal Provide visual overview of result, model, process (steps) Tips Choose suggestive or standard graphical symbols & icons Complement with accurate caption + explanation in text Use numerical values & units that make sense Make sure that the semantics of graphical symbols (boxes, arrows, etc) is... well-defined consistent throughout paper 68  Examples Goal Clarify abstract concepts concrete explanation is most effective Tips Choose interesting examples reduced versions of real situation, not artificial simplified to ease understanding, but not trivial generalizable to convince reader Do not keep moving from one example to the other use same running example throughout paper 69  Definitions Goal Technical accuracy Conciseness [cf. procedure call, macro] Tips Define every concept, term, variable before its first use Do not believe the reader has YOUR notion in mind (unless classical concepts) Reformulate a complex definition just given, in a complementary, more intuitive way to reinforce reader’s understanding Illustrate by an example 70  ◊α≅ ΦΧ∆ Notations The golden rule Favor lightweight notation Tips Avoid subscripting when not necessary e.g. use set operators if no indexing needed Example Bad: Let X = {X1, …, Xn} need to speak of elements Xi and Xj all the time if subsets are needed: 2-level subscripting needed! {Xi1, …, Xim} Good: Let x, y ∈ X and U, V ⊆ X No explicit enumeration 71 ◊α≅ ΦΧ∆ Notations (2) Tips Use notations consistently, follow uniform conventions e.g. for choosing variable name uppercase letters for sets, lowercase for elements Never same notation for different things Never same thing with different notations 72 ◊α≅ ΦΧ∆ Formulas Tips Readers / reviewers tend to skip formulas at first reading Avoid mere listing of sequence of formulas: tie them together with running commentary text should flow smoothly when replacing each formula by “blah” Example: As we know that:... , we may replace X by Z in the left-hand side to obtain:... , from which we get, using (Z):... Give name or number to formula / theorem if needed for further reference 73  Formatting Tips Use indentation to make visible the (complex) structure of a sentence, theorem, formula, program, … Good: Theorem (theorem name). Let x, y ∈ X... If … then if... then … Pagination: avoid cutting formula, program, … by use of table, figure, … Keep (sub)heading with next paragraph 74  Specific problems Presenting an algorithm / program Carefully present … the specifications: WHAT it does good: GIVEN … FIND... the reasoning followed to build it Provide abstract version (if needed, or in Appendix) Avoid comments following sequence of instructions 75  Writing Scientific Papers Outline General principles Improving paper structure Improving paper clarity Improving style The beginner’s 7 sins 76  Some rules Golden rule: Keep sentences short Cut long sentences (2-3 lines) & introduce redundancy Avoid... "also" in consecutive sentences, "thus" when causal connection is obvious heavy padding e.g. "because of the fact that" Text is like music, sentences have rhythm read, re-read your prose change wording if it does not flow smoothly 77  Some rules (2) Do not use superlatives of praise... for your own work (cf. supra) for others: explain why it is “interesting”, “remarkable” Avoid “I” unless your person is relevant Use active form with other subject BAD: I checked this on that to show... GOOD: checking this on that shows... or passive form (use moderately) GOOD: This was checked on that. As a result,... or “we” = you and me together 78  Use of verb tenses Rule for experimental papers... present tense for established knowledge past tense in Methods and Results sections because reference to work you did 79  Incomplete lists Rules of usage item-1, item-2, etc. to point out that the list is not exhaustive item-1, item-2, … when the reader can infer the rest general-item, e.g., specific-instances to point out that there might be more instances than those referred to explicitly (often used in discussion of related work) Tips avoid overdose of etc., e.g. and … 80  Names & numbers Capitalize names such as Theorem 1, Algorithm X, Method Y, Figure 3, Table II Small numbers should be spelled out unless used as names Bad: The method requires 2 passes Good: Algorithm 2 is illustrated in Fig. 1; it requires 17 iterations. The count was increased by 2. There are two reasons for this. 81  Citations Goals for reader: open door to... further information independent judgment for writer keep you honest highlight your contribution 82  Citations (2) Tips Citations should embed specific information typically: who, when, what one way to specify who, when is: [Ver99] but often other systems are required, e.g. GOOD: Verleysen has shown that poorly written papers get rejected [39, 41] Avoid long lists of pointers they necessarily convey unspecific information BAD: Some work has been done to address this problem [6,19, 38, 24, 47, 3, 19]. Do not disrupt sentence flow BAD: The XX method [6,19,38] cannot be used to verify [8, 28, 31] this property [2, 8, 28]. 83  Quotations Goal Avoid plagiarism Problem Quotations make text more hybrid & awkward Tip Paraphrase, and make explicit reference to source GOOD: As Bertrand pointed out, poorly written papers get rejected [39, 41] GOOD: This definition, borrowed from , stresses the role of... 84  Footnotes Footnotes are like parentheses disrupt smooth text flowing result from remorse & poor structuring (cf. below: 7 sins of technical writer) ⇒ avoid them wherever possible Footnotes widely used in other domains (e.g. social sciences) 85  Acronyms Convenient for compound terms occurring repeatedly Reader cannot remember more than a few (beside standard ones) ⇒ avoid AOD (Acronym OverDose) 86  Texts with maths Different formulas must be separated by words (cf. supra: readability when replacement by “blah”) Symbols in different formulas must be separated by words BAD: Consider S(p,q,r) q < p. GOOD: Consider S(p,q,r), where q < p. Don't start a sentence with a symbol n BAD: x - a has n distinct zeroes. n GOOD: The polynomial x - a has n distinct zeroes. 87  Texts with maths (2) The sentence preceding a theorem (algorithm, etc.) should be complete BAD: We now have the following Theorem. H (x) is continuous. GOOD: We can now prove the following result Theorem. The function H (x) is continuous. The statement of a theorem should be self-contained and should be motivated Do not use ⇒, ∀, ∃ within text (unless paper on logics) replace them by words: implies, for all/every, there exists 88  Writing Scientific Papers Outline General principles Improving paper structure Improving paper clarity Improving style The beginner’s 7 sins 89  The beginner’s 7 sins Typical flaws often made by novice writers sometimes by non-novice as well ! Check them out when re-reading your prose Adapted from B. Meyer, “On Formalism in Specifications”, IEEE Software, Jan. 1985. Differents kinds of sins mortal ⇒ fatal consequences venial ⇒ unfortunate consequences 90  Mortal sins Inadequacy: text element not adequately stating some feature of your approach Omission: important feature of your approach not stated by any text element Contradiction: text elements stating some feature of your approach in an incompatible way Ambiguity: text element allowing some feature of your approach to be interpreted in different ways 91  Venial sins Noise: text element yielding no information on any feature of your approach Variant: uncontrolled redundancy “... on feature already stated” Forward reference: text element making use of features of your approach not defined yet Remorse: text element stating some feature of your approach lately / incidentally (cf. use of parentheses) Remorse & forward reference often come together: caused by poor structuring of text (important features not explained first) 92  Other sins In more specific contexts Overspecification: text element stating some feature not of your problem, but of your solution define problem first, then solution Wishful thinking: text element stating some (speculated) feature of your approach that cannot be verified 93  Some useful references English Communication for Scientists. Scitable, Nature Education, 2013. http://www.nature.com/scitable/ebooks/english-communication-for- scientists-14053993/contents Lyn Dupré, Bugs in Writing. Addison-Wesley, 1998. (Entertaining !) R.A. Day, How to Write and Publish a Scientific Paper. Cambridge University Press, 1989. D. Knuth, T. Larrabee, P.M. Roberts, Mathematical Writing. Report STAN-CS- 88-1193, Department of Computer Science, Stanford University, 1988 S. Schwartz, Towards Better Scientific Writing. 1982. D. Solow, How to read and do proofs, Wiley, 1990. M.C. van Leunen, A Handbook for Scholars. Knopf, 1978. San Francisco Edit Newsletter, http://www.sfedit.net/newsletters.htm (Last visited 25.02.2014) 94  Some useful references English Communication for Scientists. Scitable, Nature Education, 2013. http://www.nature.com/scitable/ebooks/english-communication-for- scientists-14053993/contents Lyn Dupré, Bugs in Writing. Addison-Wesley, 1998. (Entertaining !) R.A. Day, How to Write and Publish a Scientific Paper. Cambridge University Press, 1989. D. Knuth, T. Larrabee, P.M. Roberts, Mathematical Writing. Report STAN-CS- 88-1193, Department of Computer Science, Stanford University, 1988 S. Schwartz, Towards Better Scientific Writing. 1982. D. Solow, How to read and do proofs, Wiley, 1990. M.C. van Leunen, A Handbook for Scholars. Knopf, 1978. 95