Transcript Presentation in English I: Technical Writing Strategies for NonNative English Speakers Michael Houle National Institute of Informatics [email protected] 2015/11/07 M E Houle @ NII.

Presentation in English I:
Technical Writing Strategies for NonNative English Speakers
Michael Houle
National Institute of Informatics
[email protected]
2015/11/07
M E Houle @ NII
1
Has This Happened to You Yet?
 “I need to write a technical report in English… Help!!”
2015/11/07
M E Houle @ NII
2
Has This Happened to You Yet?
 “I need to write a technical report in English… Help!!”
 One possible strategy?
 Get help from a proofreader!
 Try to learn from the corrections made.
 Submit your paper on time, meeting minimum standards of
presentation.
2015/11/07
M E Houle @ NII
3
Has This Happened to You Yet?
 “I need to write a technical report in English… Help!!”
 One possible strategy?
 Get help from a proofreader!
 Try to learn from the corrections made.
 Submit your paper on time, meeting minimum standards of
presentation.
 But, often there is a problem!
 The proofreader is typically not an expert in your field of specialization.
 The proofreader may not understand your writing.
 The corrections made may not match your intent.
 The paper may be technically flawed, or impossible to understand.
2015/11/07
M E Houle @ NII
4
A Better Strategy?
 Imitation.
 Read as many relevant technical papers as possible.
 Imitate their styles and expressions.
• Individual sentences will probably read more naturally.
• Technical vocabulary and language will probably be used correctly.
 If possible, send to a proofreader afterwards.
2015/11/07
M E Houle @ NII
5
A Better Strategy?
 Imitation.
 Read as many relevant technical papers as possible.
 Imitate their styles and expressions.
• Individual sentences will probably read more naturally.
• Technical vocabulary and language will probably be used correctly.
 If possible, send to a proofreader afterwards.
 However, if the paper is not well organized, even this
strategy may fail.
 Clauses, sentences, paragraphs may not connect properly.
 Readers can become lost.
 Confusion  loss of confidence in the writer  give up reading.
 Your proofreader will also become confused!
2015/11/07
M E Houle @ NII
6
Pre-emptive Strategy
 Best strategy for busy people (?):
“pre-emptive writing”.
 Identify those aspects of your own language that can
cause confusion when expressed in English.
 Minimize the problems in the organization of your paper,
so that the flow of ideas is not impeded.
 Learn appropriate expressions and terminology from
other papers or theses.
 Make sure that someone reads your paper and gives you
face-to-face feedback.
2015/11/07
M E Houle @ NII
7
General Organization of
Technical
Papers
 The parts of a typical computer science technical
paper (other disciplines may have their own styles):
 Title
 List of authors
 Abstract
 Introduction
 Main body
 Experimental results (for experimental research)
 Conclusion
 Acknowledgements (when needed)
 References
2015/11/07
M E Houle @ NII
8
General Organization of
Technical
Theses
 The parts of a typical computer science thesis
(other disciplines may have their own styles):
 Title page
 Abstract
 Preface (acknowledgements)
 Table of contents
 Table of figures (optional)
 Introduction chapter
 Main body chapters
 Conclusion chapter
 References
 Index (optional)
2015/11/07
M E Houle @ NII
9
Title
 Goals:
 Entice targeted readers into finding out more.
 Warn off other readers.
 All with the least amount of effort for the reader.
 In other words: keep it short and informative!
 Readers to keep in mind:
 Conference participants.
 Researchers following up references in another paper.
2015/11/07
M E Houle @ NII
10
Abstract
 Goals:
 Succinctly summarize the topic and contributions of the paper.
 Help an expert reader decide whether or not to read the introduction.
 Format:





Can be as short as a single sentence (e.g. mathematical journals).
Can be a half-page or more (e.g. medical journals).
Ideally, 200 – 300 words, with no unnecessary sentences or phrases.
Must be self-contained!
Notation & terminology should be defined if non-standard.
 To keep in mind:
 Experts in the field.
 Conference program committees.
 Databases of paper abstracts.
2015/11/07
M E Houle @ NII
11
Examples of Titles
and Abstracts
Review of seismic source models for underground nuclear
explosions
A number of seismic source models for underground nuclear
explosions have been developed over the past 2 decades. These
models include the spherically symmetric compressional source
model, the wave conversion source model, the tectonic strain release
source model, the spall slapdown source model, and the nearregional source model. These models are reviewed in this study and
are shown to be inconsistent with various geophysical data
associated with underground nuclear explosions. In particular, the
Rayleigh and Love wave signals generated by underground nuclear
explosions have not been explained satisfactorily by any of these
source models. To explain the observed explosion data, it may be
necessary to model the explosion seismic source as a sequence of
mechanisms producing seismic signals. These mechanisms all act
within the first few seconds following the explosion detonation. One
of the most important of these mechanisms is probably explosioninduced thrust faulting.
2015/11/07
M E Houle @ NII
12
Examples of Titles
and Abstracts
Rupture mechanism and source parameters of UmbriaMarche main shocks from strong motion data
A long sequence of earthquakes causing few casualties and considerable damaging in a wide zone struck Central
Italy starting on September 26, 1997. The earthquakes are characterised by normal faulting mechanism, with a NESW (anti-Apenninic direction) tension axis. In this paper we analyse the accelerometric records collected by the
accelerograph stations belonging to the National Accelerograph Network. About 10 stations were triggered by the
main shocks of the sequence. In particular, a small size foreshock and the two mainshocks that occurred on
September 26 (00:33 GMT, MW=5.7 and 09:40 GMT, MW=6.0) have been recorded by two digital 3-C accelerometers
located at near source distances (within 30 km from the faults). These records are relevant to investigate the detail
of the rupture kinematics, due to the close epicentral distance and azimuthal location relative to the fault
orientation and geometry. Using a trial and error approach we modelled the source mechanism through the fit of
the arrival times, the apparent source duration, the main polarization features and the entire waveforms of the
recorded signals, in order to get some insight on the rupture evolution, the location of the fracture origin point and
the fault geometry. Based on this fault kinematic model, inferences on fault slip distribution are obtained by
modelling the S acceleration waveform, comparing the ray theory synthetics with 1-5 Hz band filtered ground
velocity records. The final model shows that the seismic ruptures occurred along two adjacent, sub-parallel, low
angle dipping normal faults. Ruptures both nucleated from the fault bottom and propagated up-dip, showing
different rupture velocity and length. The presence of a transfer zone (barrier) can be suggested by the main
shocks rupture evolution. This transfer zone has probably controlled the amplitude increase of the local stress
released by the first rupture at its NW edge which triggered about 9 hours later the second rupture. The inferred
model was used to compute the predicted ground acceleration in the near source range, using a hybrid statisticaldeterministic approach. A similar trial and error methods has been also applied on the October 14, 1997 15:23
earthquake (MW=5.6). The inferred kinematic model indicates a rupture nucleating from the fault bottom and
propagating up-dip, toward the SE direction. Thus the three main shocks ruptured distinct fault segments, adjacent
and slightly offset from one to another.
2015/11/07
M E Houle @ NII
13
Examples of Titles
and Abstracts
A long sequence of earthquakes causing few casualties and
considerable damaging in a wide zone struck Central Italy starting on
September 26, 1997. The earthquakes are characterised by normal
faulting mechanism, with a NE-SW (anti-Apenninic direction) tension
axis. In this paper we analyse the accelerometric records collected by
the accelerograph stations belonging to the National Accelerograph
Network. About 10 stations were triggered by the main shocks of the
sequence. In particular, a small size foreshock and the two
mainshocks that occurred on September 26 (00:33 GMT, MW=5.7 and
09:40 GMT, MW=6.0) have been recorded by two digital 3-C
accelerometers located at near source distances (within 30 km from
the faults). These records are relevant to investigate the detail of the
rupture kinematics, due to the close epicentral distance and
azimuthal location relative to the fault orientation and geometry.
Using a trial and error approach we modelled the source mechanism
through the fit of the arrival times, the apparent source duration, the
main polarization features and the entire waveforms of the recorded
signals, in order to get some insight on the rupture evolution, the
location of the fracture origin point and the fault geometry. …
2015/11/07
M E Houle @ NII
14
Examples of Titles
and Abstracts
… Based on this fault kinematic model, inferences on fault slip
distribution are obtained by modelling the S acceleration waveform,
comparing the ray theory synthetics with 1-5 Hz band filtered ground
velocity records. The final model shows that the seismic ruptures
occurred along two adjacent, sub-parallel, low angle dipping normal
faults. Ruptures both nucleated from the fault bottom and
propagated up-dip, showing different rupture velocity and length.
The presence of a transfer zone (barrier) can be suggested by the
main shocks rupture evolution. This transfer zone has probably
controlled the amplitude increase of the local stress released by the
first rupture at its NW edge which triggered about 9 hours later the
second rupture. The inferred model was used to compute the
predicted ground acceleration in the near source range, using a
hybrid statistical-deterministic approach. A similar trial and error
methods has been also applied on the October 14, 1997 15:23
earthquake (MW=5.6). The inferred kinematic model indicates a
rupture nucleating from the fault bottom and propagating up-dip,
toward the SE direction. Thus the three main shocks ruptured distinct
fault segments, adjacent and slightly offset from one to another.
2015/11/07
M E Houle @ NII
15
Examples of Titles
and Abstracts
Large Scale Multiple Kernel Learning
While classical kernel-based learning algorithms are based on a single
kernel, in practice it is often desirable to use multiple kernels.
Lanckriet et al. (2004) considered conic combinations of kernel
matrices for classification, leading to a convex quadratically
constrained quadratic program. We show that it can be rewritten as a
semi-infinite linear program that can be efficiently solved by recycling
the standard SVM implementations. ……… In a second part we
discuss general speed up mechanism for SVMs, especially when used
with sparse feature maps as appear for string kernels, allowing us to
train a string kernel SVM on a 10 million real-world splice data set
from computational biology. We integrated multiple kernel learning in
our machine learning toolbox SHOGUN for which the source code is
publicly available at
http://www.fml.tuebingen.mpg.de/raetsch/projects/shogun.
2015/11/07
M E Houle @ NII
16
Examples of Titles
and Abstracts
Learning a Hidden Hypergraph
We consider the problem of learning a hypergraph using edgedetecting queries. In this model, the learner may query whether a set
of vertices induces an edge of the hidden hypergraph or not. We
show that an r-uniform hypergraph with m edges and n vertices is
learnable with O(24rm · poly(r,logn)) queries with high probability.
The queries can be made in O(min(2r (log m+r)2, (log m+r)3))
rounds. We also give an algorithm that learns an almost uniform
hypergraph of dimension r using O(2O((1+Δ/2)r) · m1+Δ/2 · poly(log n))
queries with high probability, where Δ is the difference between the
maximum and the minimum edge sizes. This upper bound matches
our lower bound of Ω((m/(1+Δ/2))1+Δ/2) for this class of hypergraphs
in terms of dependence on m. The queries can also be made in
O((1+Δ) · min(2r (log m+r)2, (log m+r)3)) rounds.
2015/11/07
M E Houle @ NII
17
Introduction
 Goals:
 Provide motivation & background.
 Establish the connection between the paper and the research literature.
 Summarize the topics covered, pointing out original contributions.
 Provide a “road map” to the rest of the paper.
 Allow reader to judge whether reading the full paper would be useful.
 Format:
 Should not be unnecessarily long and technical.
 Should not be short and uninformative.
 Should not be bogged down in technical details or jargon.
 As much as possible, should be understandable and self-contained.
 Readers to keep in mind:
 Graduate students and senior-year undergraduate students in the field.
 Experienced researchers in other fields.
 Conference program committee members and journal reviewers.
2015/11/07
M E Houle @ NII
18
Ordering the Introduction
 An introduction typically covers the following points,
in order:
 What is the context or field for the work?
 What is the problem to be addressed? Why is it important?
 What work has already been done? What are their
strengths and weaknesses?
 What is the approach taken in this paper? What are its
strengths (and weaknesses)?
 What are the outcomes of the work?
 How is the remainder of the paper organized?
2015/11/07
M E Houle @ NII
19
Level of Detail in
the Introduction
 Concepts, terminology and notation:
 Ordinarily, only the minimum should be introduced.
• Discussion of the issues.
• Flow of ideas.
 If many concepts and terms need to be introduced, then
consider a separate subsection.
 A background or preliminaries section can follow the
introduction.
•
•
•
•
Basic results
Details of the work of other papers
Definitions
Etc.
2015/11/07
M E Houle @ NII
20
Conclusion
 A conclusion is more than just a summary of the
paper:
 The contributions can be discussed with the benefit of the
reader being familiar with the details.
 Implications of the work can be discussed – open
problems, future work, other related issues.
 Simply repeating the statements made in the introduction
should be avoided!
 Points that don’t seem to fit well in the introduction can
often be effective when moved to the conclusion.
 Unlike the introduction, the shorter the conclusion, the
better.
 If space is limited, the conclusion can often be omitted.
2015/11/07
M E Houle @ NII
21
Acknowledgements
 Recognition of a specific contribution from someone not
directly involved in the research:




Preparation of data sets.
Permission for use of copyrighted material.
Research grants from funding agencies.
Important advice.
 Style:
 “Acknowledgements” is sometimes written “acknowledgments”.
 In most communities, names are given without titles such as “Prof.”,
“Dr.”, “Ms.”.
 It’s not usually considered appropriate to thank your spouse, God,
pet, computer, etc for a technical paper (although it is very
appropriate for a thesis!).
 In a thesis, the Acknowledgements usually appear in a longer
section after the Abstract and before the Table of Contents.
2015/11/07
M E Houle @ NII
22
References
 Purposes of references:
 To give credit to other researchers when appropriate.
 To indicate to the reader where to go for further information.
 To indicate to the paper reviewer that you (as author) have been
diligent in your review of the research literature.
 Whenever possible, try to avoid:
References that are not relevant to your work.
References to material that is hard to obtain (except for giving credit).
Inconsistencies in the formatting of references.
Referring to unpublished work. If it’s your own work, consider making
it into a technical report.
 Giving a URL as the sole reference. Links eventually end up broken!
 Referring to a “personal communication”.
 Citing a paper as (for example) “Suzuki et al.” when it has only two or
three authors.




2015/11/07
M E Houle @ NII
23
Main Body +
Paper
Introduction
Preliminaries
…
Experiments
Conclusion
Subsections
Paragraphs
 The flow of ideas:




Starts with introduction, ends with conclusion.
Conceptual unit: the paragraph (not the sentence!).
Each paragraph should be associated with exactly one main idea.
The ideas presented should flow smoothly from paragraph to
paragraph and from (sub)section to (sub)section.
2015/11/07
M E Houle @ NII
24
Main Body +
Paper
Introduction
Preliminaries
…
Experiments
Conclusion
Subsections
Paragraphs
 The flow of ideas (continued):
 There should be no abrupt, unforeseen transitions. The reader should
notice no unexplained discontinuities.
 Repetition and redundancy should be minimized or eliminated. Ideally,
the flow of ideas should be linear.
 The most critical errors in presentation and English-language
expression are those with the potential to interfere with the flow of
ideas.
2015/11/07
M E Houle @ NII
25
The Paragraph
 Structure of a good paragraph:
 Contains a sentence that (more than any other)
expresses the main idea of the paragraph.
 This sentence typically appears at the beginning
(introductory style followed by details) or at the end
(details followed by a summary).
 Logical flow from sentence to sentence – disruptions are
avoided whenever possible.
 Connections to or from the preceding and following
paragraphs, in the form of “bridging phrases”.
 Careful management of referents – the relationship
between an object and its pronouns, adjectives, and
descriptive clauses should always be clear.
2015/11/07
M E Houle @ NII
26
Examples of Paragraphs
One influential clustering method that takes the view that simple
similarity tests may not in themselves be sufficient for robust, highquality clustering is ROCK,due to Guha, Rastogi and Shim [4]. ROCK
is a hierarchical clustering algorithm in which the merge criterion
depends on the degree of overlap between neighborhood sets of
cluster items, where the neighborhood is defined according to a
minimum threshold on inter-object similarity (under the assumption
that full similarity corresponds to a value of 1, and total dissimilarity
to a value of 0). Cluster formation is encouraged when pairs of
member items have a high linkage, defined as the number of items
contained in the common intersection of their neighborhoods. The
intuition behind their approach is that whereas the distance values
themselves many not conclusively indicate whether an item is
appropriate for inclusion into a given cluster, the case for inclusion is
strengthened when the neighbors of the item are also present in the
cluster.
2015/11/07
M E Houle @ NII
27
Examples of Paragraphs
On September 26, 1997 Umbria-Marche region (Central
Italy) was struck by two moderate size earthquakes (at
00:33 GMT, MW = 5.7 and at 09:40 GMT, MW = 6.0) which
caused a few causalities and large building damages.
The two main shocks were followed by an intense
postseismic activity that lasted several months and was
characterized by hundreds of aftershocks, some of them
with magnitudes in the range 4.5-5.4. This seismic
sequence filled a gap in the area (Amato et al., 1998).
The focal mechanism for the two main shocks, evaluated
by Ekström et al. (1998) applying the CMT technique on
the long period data, are predominantly normal and the
fault planes appear to be oriented in the typical NW-SE
Apenninic direction.
2015/11/07
M E Houle @ NII
28
Pre-emptive Strategy
 Best strategy for busy people (?):
“pre-emptive writing”.
 Identify those aspects of your own language that can
cause confusion when expressed in English.
 Minimize the problems in the organization of your paper,
so that the flow of ideas is not impeded.
 Learn appropriate expressions and terminology from
other papers or theses.
 Make sure that someone reads your paper and gives you
face-to-face feedback.
2015/11/07
M E Houle @ NII
29
Common Mistakes
by Non-Native Speakers
 Misuse of articles (“a”, “the”, etc):
 Native speakers use “a” and “the” to distinguish between the concrete
and specific on the one hand, and the general and abstract on the
other.
 Almost never done by native speakers.
 Examples:
Method for search suffer from curse of dimensionality.
Or:
Last night I ate a chicken in the back yard.
 Examples of the effect of an error:
 Confusion between your proposed method and solution strategies in
general.
 Confusion between what is being proposed and what is already known.
 The proofreader may not be able to decide how to fix these sorts of
ambiguities!
2015/11/07
M E Houle @ NII
30
Common Mistakes
by Non-Native Speakers
 The ambiguous subject:
 In Japanese it is possible to designate the subject of a
paragraph using the article of speech は.
 In English, the subject must always be explicitly stated.
 Descriptive adjectives or clauses generally refer to the
closest or most recent object, not the assumed subject!
 Native speakers sometimes make these errors too.
 X は Y です:
However, several papers have recently been published which
state that similarity search is getting noisier and useless as the
image/video archives are getting larger, instead, searching the
“identical” image/video is becoming useful [2,3].
2015/11/07
M E Houle @ NII
31
Common Mistakes
by Non-Native Speakers
 Problem modifiers:
Descriptive adjectives or clauses generally refer
to the closest or most recent object, not the
assumed subject!
Native speakers sometimes make these errors
too.
Some errors can lead to serious confusion!
Example:
Existing methods for opinion extraction tend to rely
on relatively simple proximity-based or pattern-based
techniques because they only consider rather clean
review articles as opinion sources.
2015/11/07
M E Houle @ NII
32
Common Mistakes
by Non-Native Speakers
 More problem modifiers (from
http://www.io.com/~hcexres/textbook/gram1.
html#probmod):
 They found out that the walkways had collapsed on
the late evening news. (Was that before or after the
sports news?)
 The committee nearly spent a hundred hours
investigating the accident. (Did they spend even a
minute?)
 The superviser said after the initial planning the indepth study would begin. (Just when did she say
that, and when will the study begin?)
 Having damaged the previous one, a new fuse was
installed in the car. (Who damaged that fuse?)
2015/11/07
M E Houle @ NII
33
Common Mistakes
by Non-Native Speakers
 Problems with parallelism:
 Items in a series must be worded in a compatible manner.
 Variation in style within a sentence can lead to distraction
or confusion.
 Example:
This report is intended for people with some electronics
background but have little or no knowledge of geophysical
prospecting.
 Fix:
This report is intended for people with some electronics
background but with little or no knowledge of geophysical
prospecting.
2015/11/07
M E Houle @ NII
34
Conclusion
 Things you can do to make yourself understood in
your writing.
 Identify those aspects of your own language that can
cause confusion when expressed in English.
 Minimize the problems in the organization of your paper,
so that the flow of ideas is not impeded.
Resource:
日本人の英語 by Mark Petersen, 岩波新書
2015/11/07
M E Houle @ NII
35
Assignment
 Exercise in technical proofreading:
 Assume the responsibilities of a professional technical proofreader!
 Short example draft papers to be distributed in next week’s class.
 Identify and suggest corrections for errors, especially in organization,
grammar, and style.
 For grammatical errors, concentrate on those which lead to a loss of
meaning or misinterpretation. Briefly indicate how the loss of meaning or
misinterpretation arises.
 Web reference:
Online Technical Writing (by David A. McMurrey)
http://www.io.com/~hcexres/textbook/
 Due date: Thursday 25 June 2009 (electronic submission preferred).
 Work will be given one of the following four grades:
A – excellent; B – good; C – pass; D – fail
 Next lectures – examples from:
Online Technical Writing: Power-Revision Techniques
http://www.io.com/~hcexres/textbook/hirevov.html
2015/11/07
M E Houle @ NII
36