资源描述
How to Write PapersCharles Ling Chalres Ling Paper Writing Series Today: Foundation of Writing Good Papers Future talks by Yaqin, Harry, and Hongjiang Potential topics How to write survey papers How to integrate research and paper writing Be scientifically rigor but not offensive Statistical tests for empirical experiments How reviewers accept and reject papers Sentences: templates and organization Professional writing: words to use and not Common mistakes/problems Chinese often make Chalres Ling Outline Importance of Paper-writing Skills What Makes a Good Paper Why is it so Hard to Write Good Papers? How to Improve A Template for Good Papers Avoid Run-on Sentences Summary Chalres Ling Importance of Writing Skills Publishing papers is critical for researchers Publishing is hardLow acceptance rateCompeting with good papersReviewers are potential competitors Chalres Ling Outline Importance of Paper-writing Skills What Makes a Good Paper Why is it so Hard to Write Good Papers? How to Improve A Template for Good Papers Avoid Run-on Sentences Summary Chalres Ling What Makes a Good Paper Good paper = good content + writing skillsContent and results: your job Writing skills: Structure, flow, argument, Logical, clear, succinct, consistent, English Standard must be highYour best effortsCompared with others, yours is good/bestDont fool with the reviewers! Chalres Ling Outline Importance of Paper-writing Skills What Makes a Good Paper Why is it so Hard to Write Good Papers? How to Improve A Template for Good Papers Avoid Run-on Sentences Summary Chalres Ling Why Is It hard? Who are our readers/reviewers Differences between thought and language We are the “sales person Revisions: never be perfect Our English is bad, but Nothing can replace experience, but Chalres Ling Outline Importance of Paper-writing Skills What Makes a Good Paper Why is it so Hard to Write Good Papers? How to Improve A Template for Good Papers Avoid Run-on Sentences Summary Chalres Ling How to Improve: In Readers ShoesYou know your work, but not your readers ! Reviewers/Readers: in the same broad area, but have not worked on your problems Define terms, write about motivation and rationales, pose potential questions and answer them yourself What are your contributions? Show your elegance? Dont! Unless for pure math, or when you are really famous Chalres Ling A Reviewers CommentsThis paper attempts to explain the success of Naive Bayes classifiers by showing that Unfortunately, this paper was extremely difficult to follow. In fact, it took several readings before I even understood the papers basic claims. The paper should have stated, at the beginning, that it is focusing on the representational (as opposed to learnability) issue: It would also have been useful to connect this notation with something practical-for example, The arguments and proofs were difficult to follow. They would probably have been easier if the paper had first outlined their structure, before giving the details. Chalres Ling How to Improve: What Reviewers WantAAAI Reviewers (partial) check list: Does the paper introduce a new problem or provide a new solution to an existing one? What is the main result of the paper? Is the result significant? Is the paper technically sound? Does the paper provide an assessment of the strengths and limitations of the techniques/result? Is the paper clearly written so as to accessible to most AI researchers? Does the paper reference appropriate related work? Should the paper be nominated for a prize? Chalres Ling How to Improve: Note the DifferenceDifference 1: Outlining vs details Solutions: Introductory/summary sentences Appendix and footnotes to “hide details Introduction Previous Work Our New Methods Algorithm Complexity Comparisons Conclusions References This type of engine uses a robot or apider (theres an arcane difference between the two that isnt very interesting) which constantly roams the World Wide Web searching for new or updated pages (what a wonderful job eh?) When the robot visits a page it reads the whole thing, including any sub-pages on the same server connected to the page its reading, then it visits all the external links on the pages its read.and so on, you get the idea. A robot can only follow links.It finds pages in two ways: by visiting pages its been told about (ie. on its submission form) or by following links from other pages. So, if your page isnt linked from anywhere else and you havent registered it, theres no way itll ever be found ?itll be invisible. You can actually ensure that most of the deep engines dont index your pages by using special no follow Chalres Ling 1. IntroductionThe paper is organized as follows: In Section 2, we will review2. Review of Previous WorkLearning from noisy data has been extensively studied previously, and can be categorized into several classes. In the first class, entropy method is used. In the second class, Most entropy-based methods assume In sum, most previous work 3. Augmented Nave Bayes In this section, we will discuss an improved version of Nave Bayes. We will first introduce definition and motivations. We will then describe our improved algorithm, and give a proof of its complexity. Finally, 3.1 Notations and Definitions (I am making a point here; a tricky balance) Chalres Ling How to Improve: Note the DifferenceDifference 2: non-linear vs linear process Solutions:Use illustration and flow chartUse connecting wordsA3 XA1A2 B1B1 (A1) This type of engine uses a robot or spider (theres an arcane difference between the two that isnt very interesting) which constantly roams the World Wide Web searching for new or updated pages (what a wonderful job eh?) (A2) When the robot visits a page it reads the whole thing, including any sub-pages on the same server connected to the page its reading, then it visits all the external links on the pages its read.and so on, you get the idea. A robot can only follow links.(A3) It finds pages in two ways: by visiting pages its been told about (ie. on its submission form) or by following links from other pages. (B1) So, if your page isnt linked from anywhere else and you havent registered it, theres no way itll ever be found ?itll be invisible. You can actually ensure that most of the deep engines dont index your pages by using special no follow Chalres Ling Algorithm X has two components: A1 and B1. A1 is an improved version of , which is, in turn, supported by A2. We will describe Xs first component in the order of A1, A2, and A3. Afterwards, we discuss B1 and B2 in details.A1 A2 A3 Recall that X consists of A1 and B1, and we have already described A in details. We will now explain B in detail below.B1 B2 Chalres Ling How to Improve: Sell Our Work Unless it is an invited paper Purpose driven: debating: why are we goodLogical flowProblem X is importantPrevi us work A, B, have been studiedA, B have certain weaknessWe propose our new method DExperiment with D, compare with A, BD is better than A, B (rigorously tested)Why is D better? Why didnt E, F work? Strengths and weaknesses of DFuture work of D Simplest is best: con ey meanings Professional: no careless mistakesProof-readingProper use of ords Chalres Ling How to Improve: Revisions It is tedious to read own paper It is hard to find own mistakes A lot of efforts for small, last improvement SolutionsLet your paper sit (2 x 5 3 times, total 5 times Chalres Ling How to Improve: English Myth: My English is weak, so I have excuse Structure and logic: much more important Your role; proof-reader wont help much Solutions: Top-down organization, outline, logic, flow Use other good p pers as “sample Use “cheap English proof-reader Chalres Ling How to Improve: Experience It does take a lot of efforts to improveDraft, revision, revision, revision, Write many papers to learn But there are ways to speed upThis lecture seriesBooksRecord your own mistakes; eliminate them!Think about writing: 30%; writing: 70% Chalres Ling Solutions - Why Is It hard? Who are our readers/reviewers Differences between thought and language We are the “sales person Revisions: never be perfect Our English is bad, but Nothing can replace experience, but Chalres Ling Outline Importance of Paper-writing Skills What Makes a Good Paper Why is it so Hard to Write Good Papers? How to Improve A Template for Good Papers Avoid Run-on Sentences Summary Chalres Ling A “Template for Good Papers Abstract Introduction Review of Previous Work Our Work Experiments and Comparisons (Relation to Previous Work) Conclusions References Others (Appendix, footnotes, etc.) Chalres Ling Process of Writing a Paper(My own experience)Top-down refinement A 3-4 level outline, adjust many times Start on Introduction, Previous Work, . Last: Summary, Abstract, (rethink) Title Chalres Ling Template - Abstract Purpose: Summary of your work and contributions StyleWhat is the problemWhat is your solution and results (example; note: left-click on example figures to get back) Chalres Ling Template - Introduction Purpose: Background, outline of your paper StyleProblem X is importantPrevious work A, B, have been studiedA, B have certain weakness (be careful here)We propose D, features of D (how much)Experiment with D, compared with A, BD is better than A, B (in certain aspects)Outline of the paper; example Chalres Ling Template: Previous Work Purpose: draw the differences Style Previous work: may split to several classes, e.g. Can review each work in one or several sentences Compare to yours (refer to later sections) May also put it after sections about your work Things to note: Do not miss important ones. Proper use of “e.g. Do not misinterpret; do not overly criticize Emphasize the differences (example, bad example) Chalres Ling Template: Our Work Purpose: describe our work May split to several sections StyleDefinition, notation (need motivation) In the shoes of your readersAlgorithms: pseudo-code; diagram; explanationsAnswer potential questions from readers; e.g.Too much details (such as proof): appendixExceptions: footnote Chalres Ling Template: Experiments Purpose: verifications Style:Experiment design Detailed enough: can others replicate your work Simplest is best: concise, clear Comparisons (Is it scientific? T-test? Example)Discussions (make sense of the results)Draw conclusions Chalres Ling Template: Conclusions Purpose: summary, future work(.), closing StyleQuick summaryFuture workClosing Chalres Ling Template: References Purpose: supporting claims; know well all previous work StyleCitation in text: what needs to be referred? Universally true: no need; opinions: YES Other peoples work: YesConsistent Chalres Ling Template: Others Acknowledgements (example) AppendixProofs: do not stop the flow FootnotesHide details and exceptions Chalres Ling Abstract, Introduction, Summary Similar in nature Differences (length; emphasis)Abstract: short; problem and your workIntroduction: short or long; background, your work, outline of the paperSummary: short/medium; your work, future work Chalres Ling Summary - A Template Abstract Introduction Review of Previous Work Our Work Experiments and Comparisons (Relation to Previous work) Conclusions References Others (Appendix, footnotes, etc.) Chalres Ling Outline Importance of Paper-writing Skills What Makes a Good Paper Why is it so Hard to Write Good Papers? How to Improve A Template for Good Papers Avoid Run-on Sentences Summary Chalres Ling Run-on Sentences Two clauses are connected incorrectly as one sentence (by using a comma) Examples: Most researchers claim that the Nave Bayes can only represent linear functions, they are not quite correct . It results in the problem of estimating the probability of unseen word pairs, that is, the pairs that do not occur in the training set. The complexity is not linear, it is because Why do Chinese often write run-on sentences? Chalres Ling 【 多 維 新 聞 社 12日 電 】 中 文 語 音 辨 識 軟 體 市 場 一 年 後將 更 加 熱 鬧 滾 滾 。 在 中 文 語 音 辨 識 技 術 具 世 界 領 導地 位 的 微 軟 中 國 研 究 院 院 長 李 開 復 表 示 , 微 軟 的 技術 比 現 在 已 商 業 化 的 等 都 更 佳 , 預 計 一 年 後將 其 中 文 話 音 輸 入 軟 體 商 業 化 。 而 此 將 對 廣 大 的 華語 人 士 帶 來 使 用 電 腦 的 方 便 性 。中 國 時 報 十 二 日 報 導 , 微 軟 中 國 研 究 院 是 微 軟 在 全 球的 四 大 先 端 技 術 研 究 中 心 之 一 , 目 前 共 有 八 十 位 研究 人 員 , 網 羅 了 許 多 知 名 的 華 裔 人 才 。 像 卡 內 基 美隆 大 學 電 腦 科 學 博 士 的 李 開 復 就 發 展 出 全 球 第 一 個以 廣 大 詞 彙 、 語 料 作 連 續 語 音 辨 識 系 統 的 人 士 , 今年 才 卅 八 歲 的 他 , 在 九 八 年 被 微 軟 邀 請 回 大 陸 建 置中 國 研 究 院 。 許 多 後 進 者 都 是 以 他 的 架 構 作 語 音 輸入 的 開 發 研 究 。 Chalres Ling Avoid Run-on Sentences Use a “. . Start a new sentence. “, plus a connecting word: “but , “because , Use “; : “; however, Chalres Ling Outline Importance of Paper-writing Skills What Makes a Good Paper Why is it so Hard to Write Good Papers? How to Improve A Template for Good Papers Avoid Run-on Sentences Summary Chalres Ling Summary Paper-writing skills are important Paper-writing skills can be learnedThink about our readers/reviewersThought and languageFlow: logical, convincing, consistentStyle: simple, clear, concise How-to-writing + writing = acceptance!= Promotions= Recognitions Chalres Ling Other Resources Elements of Style, by William, Jr. Strunk, et al. 4th edition (August 1999) The Chicago Manual of Style: The Essential Guide for Writers, Editors, and Publishers (14th Edition) A manual for Writers of Term papers, Theses, and Dissertations, by Kate Turabian Happy Paper Writing!Thank You! Chalres LingTHE END Chalres Ling Chalres Ling Chalres Ling Chalres Ling Chalres Ling Back Chalres Ling Chalres Ling Chalres Ling Chalres Ling Chalres Ling Chalres Ling
展开阅读全文