Assignment 2: CMC
PART 1
INTRODUCTION
1.1 BACKGROUND OF STUDY
One of the compulsory subjects to be
completed by the Bachelor of English Language and Literature (BENL) students in
International Islamic University Malaysia is the Computer Applications in
Language Studies (COMPAPP). This subject comprises of a number of topics and one of them is CMC or
Computer-Mediated Communication. This topic aims to promote the technology
especially the computers which can be used as a medium to communicate as well
as spread the knowledge instead of using the conventional way of communication.
The online forum is one of the ways in
communicating with everyone around the globe. This type of asynchronous communication is still a good
way of communicating today. Even though there are many other social networking
sites available on the Internet, it is still a choice among the netizens. They
share almost every single thing related to life, as per instance, the food,
clothes, places to travel and much more.
Hence, when they share and chat almost
everything, they will be surely using different types of speech acts in
conveying the statements and the meanings of their sayings. These speech acts
must be having the contextual cues that influence them to be placed in the
conversations. Thus, this study aims to identify the
elements of speech acts which are frequently used in online forum as well as to
understand the contextual cues that influence the choice of speech acts
employed by the participants in online forum.
1.2 STATEMENT OF THE PROBLEM
Making a statement may be the paradigmatic use of
language, but there are all sorts of other things we can do with words. We can
make requests, ask questions, give orders, make promises, give thanks, offer
apologies, and so on. Moreover, almost any speech act is really the performance
of several acts at once, distinguished by different aspects of the speaker's
intention: there is the act of saying something, what one does in saying it,
such as requesting or promising, and how one is trying to affect one's
audience.
Hence, there are several elements of speech acts that
are being stressed, they are representatives,
directives, commissives, expressions, declarations. Thus, this study then will help in identifying those elements of speech acts which
are frequently used in online forum. All
of these speech acts are being identified on an online forum, Quora. Quora
comprises of a variety of topics and the chosen topic is the language exchange
online. A 1000-word corpus has been extracted from the whole conversation in
the forum and it is then be analyzed based on the elements of speech acts
mentioned before.
The difficulties encountered by the students who
use online forum in language learning process triggers this study to be
conducted where when the analysis has been done, it will finally help the
students to understand the contextual cues that influence the choice of speech
acts employed by the participants in the online forum. By observing their
statements and the way they express them will definitely show us some distinguished
features between all of them.
1.3 RESEARCH OBJECTIVES
The study aims to achieve the following
objectives:
1.
To
identify the elements of speech acts which are frequently used in online forum.
2.
To
understand the contextual cues that influence the choice of speech acts
employed by the participants in online forum.
1.4 RESEARCH QUESTIONS
1.
What
are the elements of speech acts which are frequently used in online forum?
2.
What
are the contextual cues that influence the choice of speech acts employed by
the participants in online forum?
1.5 DESCRIPTION OF FRAMEWORK
The
framework which we have chosen to use is speech acts. Speech acts are utterances which can be used to perform
actions. For instance, apologizing, suggesting and threatening. The
classification of speech acts which is used in this research is according to
Searle’s classification of speech acts. There are five elements of speech acts.
The first one is assertive which is the
speech acts that state what the speaker believes to be true. For example, “shared talk offers chat room which is good for entry learners to
start chatting”. This shows that the speaker believes that the application, shared talk is good for learners to start
chatting. However, this may not be true as it is only the opinion of the
speaker. The second one is directives, speech act which the speaker uses when
he wants something from another person.
For instance, “If you want to check out more language learning and my journey,
peek over at my little blog called Lingualism”.
The speaker suggests to the other person to take a look at his blog if he wants
to know in details. The third one is commissives, a speech act that is used when the speaker intends to do something in
the future. For example, “I am going to Germany…” This shows that the speaker
intended to go to Germany in the future. The fourth
one is expressive which are speech act that is used to show the emotions
of the speaker. For example, “I also remember my brother telling me of various
gifts he got from his Chinese tandem. I was a bit envious :)”. This portrays
the speaker feelings in which he feels envy towards his brother. The last one
is declarations in which the speech act used would cause a change in something.
This type of speech act cannot be found in our corpus.
1.6 VALUE OF CORPUS
Since the corpus
used is an online forum, Quora, it really helps in gaining and sharing
knowledge regarding language exchange online. This is because it involves
everyone from around the globe. Hence, this connection at the same time will
enhance the language learning among the users. For instance, an American, who
is a native speaker of English perhaps then communicates with a non-native
speaker from Korea instead of learning English, the Korean may then share his
or her own mother tongue with the American.
Besides that, the
exploitation of language sites may be occurred by communicating through an
online forum. This online forum can be a language site towards the netizens who
speak different languages and dialects. The corpus can really show us the usage
of language among different people of the world and how they are exchanged
through the conversation. From the conversations, we can eventually learn and
value some of the words from different languages. Hence, this creates an
awareness in learning and exchanging the language with others through the
Internet.
1.7 SIGNIFICANCE OF STUDY
This study will contribute to students to
make any intervention based on the findings derived from this study to know the contextual cues that influence the choice of
speech acts employed by the participants in the online forum. This will eventually help them in their
learning process by understanding the contextual cues in various speech acts.
It stresses the usage of speech acts on online forum Quora, which comprise of
several elements which are representatives,
directives, commissives, expressions, declarations while also being beneficial to linguists
who are also in the field of digital technologies and curriculum planners and
designers. Furthermore, this study is beneficial to specialists in teacher
training and those in charge of their training in Computer-mediated
communication (CMC) by emphasizing that trained teachers use digital
technologies more effectively through the language exchange online.
1.8 METHODOLOGY
1. Qualitative
This method involves the description and
the explanation on the topic which is the elements of speech acts on online
forum. Since every element carries every meaning, an analysis is done to know
and understand the contextual
cues that influence the choice of speech acts employed by the participants in
online forum.
2. Content-based Analysis
Data for
the study was the actual instances of written messages collected from a public
online discussion board forum, Quora. The analysis is a
content-based analysis on the corpus of 1137 words that is chosen on Quora. This
particular forum website discusses issues pertaining to every aspect of life including food, travel, learning
and much more. The general topic of the
corpus is about the language learning. Data was purposively selected texts from
the forum which was used to answer the research questions being investigated.
Therefore during the process of coding and tagging, utterances that made up of
a single word, a phrase, a sentence, or a paragraph was tagged according to the
language function they were performing such as to express an opinion, to the question, to make a suggestion and so on. With
that, the starting point for analyzing the
data is to categorize the text-based
utterances according to Searle’s (1976) Speech Acts taxonomy to explore the
interactive language function of the messages.
PART 2
LITERATURE REVIEW 1
Title of research
·
Language
Function and Knowledge Construction in Online Discussion Board Forums.
Authors
·
Alice Shanthi,
Lee Kean Wah, Denis Lajium, and Xavier Thayalan.
URL
·
https://www.academia.edu/12785713/Language_Function_and_Knowledge_Construction_in_Online_Discussion_Board_Forums
Purpose of study
·
To determine how participants in asynchronous
discussion board forum use language to share and elicit information, knowledge
and experience unique to a Malaysian setting.
Significance of the study
·
This study will aid educators and academicians
in the pedagogical aspect in using discussion boards in the teaching and
learning process.
Research questions
·
What types of language functions are mostly
used while communicating online through discussion board forums?
·
Which phases of knowledge construction is
evident in discussion board forum postings/messages?
Methodology
·
Qualitative research
·
Corpus – data for the study was the actual
instances of written messages collected from a public online discussion board
forum set in Malaysia.
·
Framework: Searle’s (1976) Speech Acts
(Assertive, Directive, Commissive, Expressive, and Declaration)
·
Method
1.
The text-based utterances are categorised according to Searle’s (1976) Speech
Acts taxonomy to explore the interactive language function of the messages.
2.
Based on these categories, the data was recoded.
3.
Then, the data was tagged again to study the
gradual process of co-construction of knowledge according to descriptors
indicated by Gunawardena, Lowe and Anderson’s (1997) Interaction Analysis Model
(IAM).
Findings
·
Assertive speech acts were most frequently
present in the online interaction followed by directives, expressive, and
commissive. No declarative speech acts were found in the corpus.
·
The data used for this study have evidence of
the different phases of knowledge construction; sharing and comparing opinion
(44%), the discovery and exploration of inconsistency ideas (33.9%),
negotiation of meaning co-construction of knowledge (14.5%), testing and
modification of proposed synthesis (4.4%), and agreement statement (3.2%),
therefore proving that new knowledge is indeed constructed and shared in the
online forums.
Discussion
·
Assertive speech acts are the speaker’s
utterances that are merely stating his/her mind and it often described as an
act to express the speaker’s belief and attention.
·
Directive speech acts were used especially by
those who have better knowledge of the subject matter to provide members who
needed information with helpful instructions either to overcome their problem
or new knowledge for better understanding of the subject-matter at hand.
·
Expressive speech acts were used not only to
inform other members of their personal opinions, but they also give a glimpse
of their emotional state.
·
Members who used commissive speech acts
revealed their future plans.
·
As for the phases of knowledge, for phase I, it
is natural process that they shared and
exchanged their experiences which helped and guided the forum members to have a
better understanding of the subject matter as they shared a common interest.
·
For phase II, when members experienced conflict
and inconsistency in ideas, they had to negotiate meaning, making it possible
for higher levels of knowledge construction to happen.
·
For phase III, forum activity has enabled some
members to try to achieve greater
understanding of the knowledge constructed.
·
For phase IV and V, the level of knowledge
construction shows evidence of
accommodation of new knowledge (or its synthesis) on the part of the
participants.
LITERATURE REVIEW 2
Title of research
·
An Analysis of
Expressive Speech Acts in Online Task-Oriented Interaction by University
Students
Author
·
Marta
Carretero, Carmen Maiz-Arevalo and M. Angeles Martinez
URL
Purpose of Study
·
To study
whether Expressives equally frequent across the three sub-corpora, and they
similarly distributed in terms of sub-types such as Apologies, Thankings,
Compliments, and so forth.
·
To study
whether the contextual variables with a bearing on the choices made by
participants.
Research Questions
·
Are Expressives
equally frequent across the three sub-corpora, and are they similarly
distributed in terms of sub-types such as Apologies, Thankings, Compliments,
and so forth?
·
If this is not the case, which are the
contextual variables with a bearing on the choices made by participants?
Significance of Study
·
The analysis on
the relative frequency of occurrence of different subtypes of Expressives
across the three subcorpora.
·
The influence
of certain contextual variables have a strong bearing on the Expressives
employed by each group.
Methodology
·
Quantitative
·
Corpus:
83 university students belonging to one of the following groups: 64
undergraduate students taking an optional course on English Discourse, 9
undergraduate students from an evening group taking an obligatory course on
Pragmatics and 10 post-graduate students following the Master’s Seminar on
English Linguistics.
Each group of participants was subdivided into smaller groups of
three or four students, randomly created by Virtual Campus itself.
·
Procedures
Framework
Speech
Acts (focusing on Expressive)
Method
1.
These smaller
groups had to carry out one or two collective assignments.
2.
They were asked
to do these collaborative exercises online, by means of an e-forum.
3.
Once the
activity was over, participants gave their written consent.
Findings
·
The analysis
uncovers two main similarities:
1.
A predominance
of other-oriented over self-oriented speech acts
2.
The high degree
of conventionalization found in the most recurrent subtypes: Compliments,
Greetings, Thankings, and Apologies.
·
The analysis
also showed remarkable differences in terms of frequency of use, concrete
linguistic realizations of individual subtypes, and the use of typographic
marks.
LITERATURE REVIEW 3
Title of research
·
Classifying
Sentences as Speech Acts in Message Board Posts.
Author
·
Ashequl Qadir
and Ellen Riloff
Journal of publication
·
Proceedings of
the 2011 Conference on Empirical Methods in Natural Language Processing
URL
·
https://www.cs.utah.edu/~riloff/pdfs/emnlp11-speechacts.pdf
Purpose of study
·
to distinguish
between expository sentences and speech act sentences in message board posts
·
to classify
speech act sentences into four types: Commissives, Directives, Expressives, and
Representatives
Statement of Problem
·
The text genres
(weblogs and social media sites) offer new challenges for natural language
processing (NLP)
Research Questions
·
How to
distinguish between expository sentences and speech act sentences in message
board posts?
·
How to classify
speech act sentences into four types: Commissives, Directives, Expressives, and
Representatives?
Significance of Study
·
Information
extraction systems could benefit from filtering speech act sentences so that
facts are only extracted from the expository text.
·
Identifying Directive sentences could be
used to summarize the questions being asked in a forum over a period of time.
·
Representative sentences could
be extracted to highlight the conclusions and beliefs of domain experts in
response to a question.
Methodology
·
Quantitative
·
Corpus:
Randomly selected 150 Veterinary
Information Network (VIN) message board threads from this collection on the
three topics: cardiology, endocrinology, and feline internal medicine.
·
Framework:
Searle’s Speech Acts (Commissives,
Directives, Expressives, and Representatives)
·
Procedures:
1.
We did basic
cleaning, removing html tags and tokenizing numbers.
2.
Two human
annotators were told to assign one or more speech act classes to each sentence.
3.
For our first
experiment, we created a speech act filtering classifier to distinguish
sentences that contain one or more speech acts from sentences that do
not contain any speech acts.
4.
Our next set of
experiments focused on labelling sentences with the four specific speech act
classes: Commissive, Directive. Expressive, and Representative.
Findings
·
We achieved
good results for speech act filtering and the identification of Directive and
Expressive speech act sentences.
·
We found that
Representative and Commissive speech acts are much more difficult to identify,
although the performance of our Commissive classifier substantially improved
with the addition of lexical, syntactic, and semantic features.
PART 3
No
|
Elements of Speech Acts
|
Frequency
|
1
|
Representatives
|
10
|
2
|
Directives
|
7
|
3
|
Commissives
|
3
|
4
|
Expressives
|
5
|
5
|
Declarations
|
0
|
Total
|
25
|
|
Summary of Speech Acts
FINDINGS
Based on the table
of summary of the data from the corpus, it can be seen that assertives speech
acts occurs repeatedly and most often used by the users in the discussion where
it takes place ten times from the beginning until the end of the discussion. For
instance, in corpus that we have chosen, a user said “Still, it’s a great tool
to use no matter what level you are at” and another user makes a statement “The
last product that I have used has been the app HelloTalk, which has probably
been the most productive and most used tool I have used so far”. From these two
examples, it can be seen that the users try to share their opinions about
certain websites that they have been using.
The second frequent speech acts used in online forum is directives
speech acts which happens about seven times during the discussion. For this
part, it can be said that the users use directives speech acts when their aims
are to give instruction or command to other users. Based on the analysis of the
corpus, a user said “If you want to check out more language learning and my
journey, peek over at my little blog called Lingualism”. What the user tries to
convey from his statement is that he wants the other users to feel free
visiting his blog.
As for expressives
speech acts, it appears quite frequently in an online forum with the frequency
of 5 occurrences throughout the corpus. By using expressive speech acts, the
participants intend to express their feelings and emotions in the online forum.
For example, one of the participant expresses their feelings by saying, “thanks
so much” to express gratitude towards the suggestion made by other
participants. Other participant also express his emotion by saying, “I was a
bit envious” to show his envy towards his brother.
As for commissives
speech acts, it also appears in an online forum with the frequency of 3
occurrences throughout the corpus. By using commissives speech acts, the
participants show their intention to do something in the future. For instance,
one of the participant said that, “I am going to Germany” in order to tell his
future planning of going to Germany.
As for the last speech acts, declarations, it does not occur in the
online forum. This shows that there are no utterances uttered which would
causes changes happen to the world.
In conclusion, the
most frequent speech acts is assertives in which the participants share their
thoughts and opinions. Then, it is followed by directives in which the
participants express what they want. They also express their emotions and feelings
through the online forum. Finally, they also show what they intend to do later
in the future.
DISCUSSION
Discussion on Research Question 1: What are the
elements of speech acts which are frequently used in online forum?
The elements of speech acts
which are frequently used in an online forum are assertive. The reason why
assertive is the most speech acts used is because the online forum is a place where
people leaves their opinions and suggestions. Assertive is kind of speech acts
that states what the speaker believes to be the case or not. Thus, most of the
people in the conversation leave their opinions on their experiences
doing language exchanges online while some of them leave some suggestions on
the best language sites that can be used in learning English language.
Directive speech act is also frequently used in this conversation as it is a
type of speech act that speakers use to get someone to do something. As can be
seen from the corpus, Directive is used when a speaker wants the other people
to give opinions or suggestions on which language sites are better in learning
the language. This is because there are many language sites that can be found
in today's time but some of them might not be useful to the learners.
Besides that, Expressive is also frequently
used in this conversation. Expressive is the speech act that states what the
speaker feel. While reading the comments and feedbacks from the readers, they
tend to express what they feel on the suggestions and opinions given by the
people in the online forum. Other than that, Commissive which helps the
speakers to commit some future action is also used in this conversation. This is due to the reason that after reading
the suggestions and opinions from other people, they are interested in trying
the language sites suggested by others. Thus, some of the people in this
conversation express what they intend to which they replied something that they
will do in future. Declaration does not exist in this conversation as there is
nothing to be changed due to the fact that they are just sharing their
experiences on online language exchanges throughout the conversation.
Discussion on Research Question 2: What are the
contextual cues that influence the choice of speech acts employed by the
participants in online forum?
First
and foremost, for the assertives, the example in the corpus is when a
participant in the online forum responded to the question regarding the
experience of doing language exchange online. He explained his experience as
well as the sites that he used in exchanging language. The assertives element
can be seen when he said “It’s a great introduction to online language
exchange”. This shows that he believed that the suggested site which is
sharedtalk.com is really a great site and he viewed his opinion by saying so.
The contextual cue that triggers this speech act is the experience of the
language exchange and the site that he used itself. This is because the
participant is eager to share his experience using the online forum sites in
order to exchange the language.
Besides
that, for the directives, the example is when a participant introduced herself
and wanted to join others as their friend. This can be seen when she said: “I
want to be your friend”. The contextual cue that involves in this speech act is
the previous conversations of other participants which then trigger the
affected participant to say so. This may be because of the favor of joining the
active conversations among the participants. The heat may be felt by her that
finally made her start the conversation and requested to be their friend.
Apart
from that, the commisives can be found in the corpus when a participant said
“probably will do a review!” This shows that the participant may be unsure
about doing the review, even though he does have the intention. The contextual
cue for this is when he came up with an update about the new website that he
found interesting. It means that he might not want to elaborate on the new
website instead of just telling the other participants about the existence of
the website.
Last
but not least, the expressives can be vividly seen when another participant
said “Yay!” that shows the happiness and joyfulness in him. He talked and
shared about his experience in a language exchange site. The contextual cue
here is his experience of buying a group ticket where the group was introduced
by his German partner with a huge discount. This influence him to say “yay” to
show that he was happy and delightful.
In relating both speech acts and contextual cues
with the online forum, it is known that users in online forum have different
experiences and encounter different situations. Thus, their experiences trigger
and influence them to use particular speech acts in their utterances in the
online forum. In other words, the discussions that take place in the online
forum have something to do with what the users have experienced before.
Different people would experience different things. Hence, online forums is a
medium for them to exchange opinions, ideas and thoughts among them.
PART 5
Raw Data
What is your experience of doing language exchange online?
Any positive experience and challenges, would be great to share.
In short, online language exchange has left me with tons of
experience, albeit much of it repetitive, but quite a bit deep as well. It's
also left me with plenty acquaintances and a very small number of people I talk
to regularly, which has been pretty great actually.
I first started using sharedtalk.com. Run by Rosetta Stone and slightly outdated, it's a great introduction to online language exchange. Using it made me a great conversationalist and gave me the ability to not only filter what forms of a language I wanted, but connect with learners my same age. I've been exchanging with one guy since September 2013, and it all started with a small conversation in which he made me laugh on sharedtalk, but the site is also kind of a hit or miss. The good thing is, you have a lot to choose from.
After sharedtalk I proceeded to Italki.com . The site really offered great tools such as peer editing and asking open questions, and great assets such as the community itself to help me out. Though, I did not get into it very much and did not end up making any good exchange partners. Still, it's a great tool to use no matter what level you're at, just be prepared for an excess of notifications depending on who you are, but none that are "off topic."
After Italki I then moved to Interpals. I discovered the youth community to have a much greater presence here, but of course that also brought a few problems. It's perfectly possible to find a great partner. There are great tools to use to narrow down to exactly what it is you want. A problem is that some users are not very serious about their endeavors or are quite "off topic." Never the less, I had many good mid-length running conversations there before I phased out a little. I might come back at some point, though.
After searching for a short while, I finally found a good way to practice in high quantity verbally. Verbling.com was the answer and it's a great tool even if you use it free. It may take a little bit to get over some beginner anxiety, but it's overall great. The only possible complaint about this is that it's possible in big groups for maybe 2 or 3 people to do all the talking, but that's practically bound to happen no matter what. It's just how groups work and you can specifically regulate the group size or even do a one on one anyway. No matter what, even just listening, you'll get a lot better with not only speaking anxiety, but verbal fluency as well.
The last product that I've used has been the app HelloTalk, which has probably been the most productive and most used tool I've used so far. I've called it a "gift from the language gods." It really does offer a great amount of useful tools. On HelloTalk, once can narrow users down even by city. You can also block users, which can prove useful. Content is regulated, so spam is nonexistent. You have verbal and textual tools ranging from voice-mail like chatting and modifiable speech synthesizers, to special windows to show textual corrections and the ability to translate any message sent immediately in the window just by tapping it. I've developed some great connections here and had some great conversations that would have been slightly more difficult otherwise.
Without all these tools, there is no way I'd be capable of speaking probably to even half the abilities I have now. Depending on how you use them, you can really, really excel or feel like you're not moving much at all. If you want to check out more language learning and my journey, peek over at my little blog called Lingualism.
Update:
New website I've found GoSpeaky.com, really great probably will do a review!
I first started using sharedtalk.com. Run by Rosetta Stone and slightly outdated, it's a great introduction to online language exchange. Using it made me a great conversationalist and gave me the ability to not only filter what forms of a language I wanted, but connect with learners my same age. I've been exchanging with one guy since September 2013, and it all started with a small conversation in which he made me laugh on sharedtalk, but the site is also kind of a hit or miss. The good thing is, you have a lot to choose from.
After sharedtalk I proceeded to Italki.com . The site really offered great tools such as peer editing and asking open questions, and great assets such as the community itself to help me out. Though, I did not get into it very much and did not end up making any good exchange partners. Still, it's a great tool to use no matter what level you're at, just be prepared for an excess of notifications depending on who you are, but none that are "off topic."
After Italki I then moved to Interpals. I discovered the youth community to have a much greater presence here, but of course that also brought a few problems. It's perfectly possible to find a great partner. There are great tools to use to narrow down to exactly what it is you want. A problem is that some users are not very serious about their endeavors or are quite "off topic." Never the less, I had many good mid-length running conversations there before I phased out a little. I might come back at some point, though.
After searching for a short while, I finally found a good way to practice in high quantity verbally. Verbling.com was the answer and it's a great tool even if you use it free. It may take a little bit to get over some beginner anxiety, but it's overall great. The only possible complaint about this is that it's possible in big groups for maybe 2 or 3 people to do all the talking, but that's practically bound to happen no matter what. It's just how groups work and you can specifically regulate the group size or even do a one on one anyway. No matter what, even just listening, you'll get a lot better with not only speaking anxiety, but verbal fluency as well.
The last product that I've used has been the app HelloTalk, which has probably been the most productive and most used tool I've used so far. I've called it a "gift from the language gods." It really does offer a great amount of useful tools. On HelloTalk, once can narrow users down even by city. You can also block users, which can prove useful. Content is regulated, so spam is nonexistent. You have verbal and textual tools ranging from voice-mail like chatting and modifiable speech synthesizers, to special windows to show textual corrections and the ability to translate any message sent immediately in the window just by tapping it. I've developed some great connections here and had some great conversations that would have been slightly more difficult otherwise.
Without all these tools, there is no way I'd be capable of speaking probably to even half the abilities I have now. Depending on how you use them, you can really, really excel or feel like you're not moving much at all. If you want to check out more language learning and my journey, peek over at my little blog called Lingualism.
Update:
New website I've found GoSpeaky.com, really great probably will do a review!
Excellent personal reflection and tips, JD,
thanks so much for sharing.
I am also curious to learn more about your experience of using purely commercial sites like Verbling for free, even decided to ask a question on Quora about that.
I am also curious to learn more about your experience of using purely commercial sites like Verbling for free, even decided to ask a question on Quora about that.
very informative
How are you,
My name is miss Lucy and i want
to be your friend.and so that i we tell you about me please contact me (lucy_mouna@yahoo.in) ok
For me, the biggest challenge has been to find users that
actually do start language
exchange (meaning start talking on Skype).
Many users I have connected to are very keen on exchanging Skype IDs and e-mails, but after that they disappear and never actually start talking.
I was wondering if there any solution to this problem.
Is there an effective way for learners of English to overcome the false "fear of speaking English"?
Many users I have connected to are very keen on exchanging Skype IDs and e-mails, but after that they disappear and never actually start talking.
I was wondering if there any solution to this problem.
Is there an effective way for learners of English to overcome the false "fear of speaking English"?
each platform has advantage and disadvantage. sharedtalk offers
chat room which is good for entry learners to start chatting, and live voice
chat for advanced learners for 1 to 1 conversations. Italki offers tools for
writing essays and getting corrected, and kind of forum for users for mutual
help. but live chat is not available there.
TALKEER is what i want to recommend here. TALKEER.com
TALKEER is what i want to recommend here. TALKEER.com
offers live voice and video
chat, essay writing and correction, and messaging, and personal profile and
album. Tutoring and learning functions go paralel with socializings.
Now I’m using Bilingua: Your Language Exchange & Learning
Companion,
it helps me to find French native speakers with whom i can conversate on the
topics i’m interested in.
Josef, Founder of a language exchange
site GetTandem.com.
I only have a good experience with
tandems. Lucky me!
I give
you one that just crossed my mind. I mentioned my German tandem partner
that I am going to Germany to visit a friend and that I will need to buy a
train ticket. Not only did he suggested to use shared rides (mitfahrgelegenheit) to save money, he
immediately found me a group, called the guy and arranged everything so when I
found myself on the platform someone just approached me to take me on a group
ticket and I went for 5 instead of 20 euros! Yay!
And things like this just happen. I also remember my brother
telling me of various gifts he got from his Chinese tandem. I was a bit envious
:).
Sun Shine, Majoring in English
Language and Literature
Hello. I have never experienced any language exchange online
before but one of my friends has suggested to use WeSpeke. It is a language site that
enables the user to communicate with the English speakers across the globe. I
have seen her communicating with English speaker and yes, it is quite
interesting. It is just like Skype, but if you want to try something new, I
would recommend WeSpeke to you. :)
Analyzed Data
REFERENCES
References
Ashequl Qadir, & Riloff, E.
(2011). Classifying Sentences as Speech Acts in Message Board Posts. Proceedings
of the 2011 Conference on Empirical Methods in Natural Language Processing.
Retrieved from https://www.cs.utah.edu/~riloff/pdfs/emnlp11-speechacts.pdf.
Carretero, M.,
Maíz-Arévalo, C., & Martínez, M. Á. (2015). An Analysis of Expressive
Speech Acts in Online Task-oriented Interaction by University Students.
Procedia - Social and Behavioral Sciences, 173, 186-190. Retrieved from http://ac.els-cdn.com/S1877042815013609/1-s2.0-S1877042815013609-main.pdf?_tid=6f9aa796-b59c-11e6-97ee-00000aacb362&acdnat=1480359597_c5a56d1a1237fa7ef6752cac09b4ea93
Shanti, A., Wah, L. K., Lajum, D., & Thayalan, X. (2015). Language
Function and Knowledge Construction in Online Discussion Board Forums. Frontiers
of Language and Teaching, 6. Retrieved from https://www.academia.edu/12785713/Language_Function_and_Knowledge_Construction_in_Online_Discussion_Board_Forums
Yule,
G. (1998). Pragmatics. Oxford: Oxford Univ. Press.
Article 1
Frontiers of Language and
Teaching …...………………………. Volume 6 (2015)
94
Language Function and Knowledge Construction in Online Discussion Board Forums
Alice Shanthi Universiti Teknologi MARA
Lee Kean Wah Universiti Malaysia Sabah
Denis Lajium
Universiti Malaysia Sabah
Xavier Thayalan Universiti Teknologi MARA
Corresponding Author’s Email: alice_shanthi@yahoo.com.my
Abstract One form of online communication categorised as
asynchronous is online discussion board forums. This paper presents the
findings of a study on the interactive language functions and the phases of
knowledge construction in asynchronous computer-mediated discourse (CMD) in an
online discussion board forum in Malaysia. Data for the study was collected
using purposive non-random data samples motivated by theme. It was found that
the members of the online discussion forum used more assertive speech acts such
as explaining, giving suggestions, agreeing, supporting, and answering to
queries while interacting online. The second part of the study revealed that
the most common phase of knowledge constructing was the act of sharing and
comparing of opinion, as well as the discovery and exploration of dissonance or
inconsistency among ideas, concepts or statements.
Keywords: Computer Mediated Communication, Computer Mediated
Discourse, Asynchronous Computer-Mediated Discourse, Online interaction
Introduction The advent of
the Internet has revolutionized the manner by which people communicate
virtually. The growth of Internet
service providers and the increasing number of virtual communication tools are
testimonies of the popularity of virtual communication. It is said that people are attracted to
communicate virtually because it reduces the constraints of time and distance,
when people share information, experience and feelings with one another (van
Varik & van Oostendorp, 2013).
Additionally, Internet communication especially asynchronous does not
require people to reply to messages instantly but allows them to make a more
thought out reply or answer. This allows
people who are engaged in asynchronous communication to pull in resources,
knowledge and expertise in an online discussion forum. As such asynchronous communication helps
people to engage in fruitful discussions such as overcoming problems through
the knowledge acquired from the online discussion forums. Discussion board
forums, newsgroups and email are some of the most common asynchronous mode of
virtual communication. According to
Annand ( 2011), discussion board forum is an interactive channel which allows
users to be active and engage in a two-way communication. Furthermore, it is an inexpensive
Frontiers of Language and
Teaching …...………………………. Volume 6 (2015)
95
way of information seeking for increasing efficiency and
productivity (Miller, 2009). Thus it is a good tool for generating dialogue
between and among users, and to solicit feedback from others. Hence, the focus
of this study is discussion board forums that allow people to read and exchange
comments while expressing views on a particular subject. Some of the popular
discussion board forums in Malaysia are Lowyat.NET, mudah.my, cari.com.my and
Webportal Malaysia. By studying the different discourse functions the
researchers aim to determine how participants in asynchronous discussion board
forum use language to share and elicit information, knowledge and experience
unique to a Malaysian setting. Two general research questions addressed in this
study are: 1. what are the types of language function mostly used in discussion
board forums?, and 2. which phases of knowledge construction is mostly evident
in discussion board forums?
Literature Review There are basically two modes of CMC;
asynchronous and synchronous. Online communication that allows for a delay
between message and response, meaning the people interaction need not be online
at the same time is regarded as asynchronous, whereas communication that occur
in real time, meaning the people interacting must be online at the same time is
termed as synchronous communication.
Discussion board forum is one of the most common types of asynchronous
CMC which enables multiple users to engage in discussion with each other; read
and exchange comments beyond real time. It has empowered people from diverse
background to meet and engage in online discussion (Herring, 2004; Paolillo,
2011). In discussion board forums people share information and experiences thus
creating a space where knowledge can be constructed, and shared (Thanasingam, Kit,
& Soong, 2007). The information
shared in online discussion board forums, unlike other forms of online
discussion such as chat rooms or online conferencing, is stored in the form of
messages in the archives of the forums, and they are arranged according to
topic. Since past messages in the form of written text remain on the site along
current messages, and are arranged according to topics, thus it makes it easier
for the researcher to choose data according to the criteria set by the
researcher (Byrne et al., 2013). Hence,
making discourse analysis of participants’ text- based transcripts an
effective technique for researchers to get a better understanding of the
participants’ cognitive processes and of the phases that depict the quality
of knowledge constructed and shared
online (Gunawardena, Lowe, & Anderson, 1997; Wang, 2005; Akayoglu & Altun, 2009). Next looking at prior studies conducted
locally in the academic setting using asynchronous mode of CMC found it to be
of benefit to students. Kaur and Mohamed Amin (2010) investigated the
effectiveness of the use of asynchronous computer-mediated communication in
emails using quantitative method and semi-structured interview and found that
asynchronous computer-mediated communication has the potential in aiding
learners to take charge of their own learning. In another study, Thayalan and
Shanthi (2011), conducted a study to investigate the social presence
experienced by undergraduates in online forums for distance learning, the study
found that interactivity in the discussion boards served the purpose to
maintain contact among the students. The study also found that students
actively read messages posted by others, but posted limited number of messages,
thus limiting the amount of information shared online. Other studies conducted
overseas based on the experience of CMC in online communities by analysis of
their transcripts of online interaction suggest that in addition to the content
of the discussion, the interactive strategies (denoted by the used of speech
acts) used play important
Frontiers of Language and
Teaching …...………………………. Volume 6 (2015)
96
role in determining active participations in online interaction
(Pena-Shaff & Nicholls, 2004; Schallert et al., 2009; Means, Toyama,
Murphy, Bakia, & Jones, 2009). Pertaining to studies conducted to
investigate CMC and of its use for knowledge construction, Fitzpatrick (2010), cited a few studies that
show favourable results such as studies conducted by Aviv, Erlich, Ravid, and
Geva, 2003; Hawkey, 2003; Hiltz, Coppola, Rotter, and Turoff, 2000; Curtis and
Lawson, 2001; McConnell, 2000; Thomas, 2002. These studies found evidences for
higher order thinking and knowledge building through collaborative learning
that happened through online interaction in web forums or online discussion
boards. However, Paulus and Phipps,
(2008), in their study found that students engaged in asynchronous discussion
board as part of their course fulfilment did not go beyond surface- level
discussion, and so questioned whether deep, meaningful discussions are even
possible in asynchronous learning environments. Another study that came out
unfavourable to asynchronous CMC was conducted by Lester and Paulus, (2011),
their study found that the online interaction lacked “quality”. They stated
that the lack of “quality talk” will be a notable problem because only when
members of a virtual community actively give and comment on each other’s ideas
can knowledge be constructed and shared in online learning. Problems with getting
good participation for online discussion board forums from members was also
encountered by Griffith (2009), who examined computer-mediated communication
discussions in educational environments for evidence of learning, and found
that attempts to use a voluntary asynchronous discussion forum among student
members resulted in little to no participation.
In short, besides the physical aspects of discussion board forums such
as the number of participants, speed of internet and excess to computers and so
on, the role of language in how it is used to perform actions (speech acts)
plays an important role in getting participants to continue posting messages,
and thus continue to share, elicit and exchanging information. This study hopes
to look into the discourse function of the messages themselves to have a better
understanding of the participants’ use of language to communicate and construct
knowledge online.
Method Data for the study
was the actual instances of written messages collected from a public online
discussion board forum set in Malaysia. This particular forum website discusses
issues pertaining to everyday Malaysian life. Data was purposively selected
texts from the forum which was used to answer the research questions being
investigated. According to Herring (2004), qualitative analysis of text based
Computer Mediated Discourse (CMD) is usually based on individual themes as the
unit for analysis, rather than the physical linguistic units (e.g., word,
sentence, or paragraph). Therefore during the process of coding and tagging,
utterances that made up of a single word, a phrase, a sentence, or a paragraph
was tagged according to the language function they were performing such as to
express an opinion, to question, to make a suggestion and so on. With that, the starting point for analysing
the data is to categorise the text-based utterances according to Searle’s (1976)
Speech Acts taxonomy to explore the interactive language function of the
messages. Based on these categories, the data was recoded and tagged again to
study the gradual process of co-construction of knowledge according to
descriptors indicated by Gunawardena, Lowe and Anderson’s (1997) Interaction
Analysis Model (IAM). Messages from the
three online discussion board forums which comprise of different interest
groups (IG) were analysed as stated in Table 1.
Frontiers of Language and
Teaching …...………………………. Volume 6 (2015)
97
Table 1: Data set selected for the study Interest
group Topic Messages No. of words Fast and Furious
Proton Saga FLX Very High fuel Consumption 92 4145 Finance, Business and
Investment House Geneva Malaysia V2 130 4786 Computer Technical Support Folding@Malaysia
needs your help! 62 3099
Total
284
11530
Findings and Discussion
Research Question 1: What types of language functions are mostly used
while communicating online through discussion board forums? Messages from the three different interest
groups were coded and analysed to study the function of language used to
communicate online, following Searle’s (1976) category of speech act analysis:
assertive, directive, commissive, expressive and declaratives. This yielded a
total of 492 speech acts (refer to Table 2). In total, almost half of the
language function used to communicate by members from the different interest
groups was assertive (47.6%) in nature, roughly 32% was directive, and
expressive stood at 17 %, and finally, almost five per cent of participants’
speech acts consisted of commissive acts. No declarative acts were found in
this sample.
Table 2: Functions of Utterance According Speech Acts Types of Speech Acts IG1 – Fast and Furious
IG2 - Finance, Business and Investment House IG3- Computer Technical
Support
TOTAL
%
Assertive 83 103 48 234 47.6 Directive 61 69 26 156 31.7 Commissive
7 9 5 21 4.3 Expressive 28 41 12 81 16.5 Declaration 0 0 0 0 0 IG – interest
group
The study found that while communicating online in discussion board
forums the function or purpose of language used were more assertive in nature.
Assertives are primarily used to share information with other members of the
group by explaining, describing, stating
opinion, reflecting, disputing, making
predictions and so on. They are mainly statements that are neither true or
false, accurate or inaccurate (Searle, 1976), but rather these are the
speaker’s utterances that are merely stating his/her mind. An assertive act is
often described as an act to express the speaker’s belief and intention.
Frontiers of Language and
Teaching …...………………………. Volume 6 (2015)
98
The study also found that directive speech acts also play an
important role in virtual community members’ discourse. Directive speech act
that was commonly used in the forum was questioning. This action was used in
order to elicit direct responses from those seeking information or help. As the
directive speech acts focus on getting the receiver to do something (Searle,
1976), besides the action of questioning, this study found that directives such
as suggesting, requesting or asking, inviting, insisting and so on, were used
by members. These actions were used especially by those who have better
knowledge of the subject matter to provide members who needed information with
helpful instructions either to overcome their problem or new knowledge for
better understanding of the subject-matter at hand. Expressive speech acts were
also relatively frequent in the discussion board messages, comprising 16.5% of
the speech act. Through the display of emotions and feelings (e.g., " haha
i can't feed my car 97 fuel, i even have problem feeding myself every
month", "ai yo yo.... this poor guy!", “STOP MILKING SYMPATHY
AND ACCEPT YOU LOSS QUIETLY!!!!!!!!), participants not only inform other
members of their personal opinions, but they also give a glimpse of their
emotional state (e.g., inspired, happy, sad, angry, stressed). Next, by posting
commissive based messages, members performed acts such as promising, refusing,
offering and/or volunteering to help other members in the discussion board
forum. Members of the forum revealed their future plans, mostly based on the
new information/knowledge gathered from the discussion (e.g., "ok i will
change to lighter oil for my next service"). In conclusion, by using Searle’s speech acts,
the taxonomy has provided this study important insight into how messages from
discussion boards were built linguistically.
This study found that in the process of discussion the members used
mainly used assertive speech acts to share information. They also asked
questions in order to get information and at the same time get other members to
respond to them with their personal experiences and knowledge so that the other
members in the virtual community can share their knowledge and experience.
Research Question 2: Which phases of knowledge construction is
evident in discussion board forum postings/messages? The data selected to
answer research question two is the same as that which was selected to answer
research question one (refer to Table 1). As shown in Table 3, 109 (44 %)
comments were categorized as sharing and comparing of opinions (Phase I level).
84 (33.9 %) stating disagreements, asking and answering questions (Phase II level),
36 (14.5 %) displaying negotiation of meaning and co-constructing knowledge
(Phase III level), 11 (4.4%) messages showed evidence that participants’
perception have changed as a result of the interaction in the discussion board
(Phase IV level), and finally 8 (3.2%) refers to messages that show evidence of
accommodation of new knowledge (or its synthesis) on the part of the
participants of the discussion board forums.
Frontiers of Language and
Teaching …...………………………. Volume 6 (2015)
99
Table 3: Phases of knowledge construction in online discussion
board forums Phases of knowledge construction IG1 IG2
IG3 TOTAL % Ph I- Sharing and
comparing of opinion 29 59 21 109 44.0 Ph II - The discovery and exploration of
dissonance or inconsistency among ideas, concepts or statements 34 36 14 84
33.9 Ph III - Negotiation of meaning co- construction of knowledge 13 17 6 36
14.5 Ph IV - Testing and modification of proposed synthesis or co-construction
6 2 0 11 4.4 Ph V - Agreement statement (application of newly constructed
meaning. 6 4 1 8 3.2 Total 88 118 37 248
IG1 – Fast and Furious IG2 - Finance, Business and Investment House IG3-
Computer Technical Support
The findings signify that the most common activity for constructing
and sharing knowledge was exchanging ideas, opinions and experiences (44%). As
most members shared a common background/interest it seems natural that they
shared and exchanged their experiences, resources and/or information which
helped and guided the forum members to have a better understanding of the
subject-matter they were discussing, and in the process they constructed and
shared new knowledge Next, 33.9% of the comments posted in the discussion
forums were clarification comments (level II). When members experienced conflict
and inconsistency in ideas, they had to negotiate meaning, making it possible
for higher levels of knowledge construction to happen. In fact in IG1 there
were more phases II level of knowledge construction compared to phases I,
suggesting this group of people were constructing new knowledge by asking and
answering questions to clarify the source and extent of disagreement. As such,
suggesting that the online forum has been effective in engaging members of the
interest group to critically review their peers’ feedback on the subject-matter
being discussed. Members also at times counter-argued and sometimes criticised
or provoked reactions, these actions raised the opportunities for further
discussions and exchange of ideas. Phase
3 level comments, though small in number (14.5%) suggests that the forum
activity has enabled some members to try to achieve greater understanding of
the knowledge constructed. Through exercising higher mental functions such as
negotiating or clarifying (level II), they have tried to process and construct
more accurate feedback on the subject-matter (level III). The findings on
levels of knowledge constructed also suggest that discussion forums promote the
construction of critical feedback. These findings support Thanasingam, Kit, and
Soong's (2007), claim that tools such as discussion forums facilitate knowledge
construction through collaboration. An
almost similar lower percentage of phase IV and V level of knowledge
construction were also detected in the messages taken from the discussion board
forums. There were 11(4.4%) comments of phase 4, and 8 (3.2%) comments that
were observing knowledge construction at
phase 5. These messages show evidence of accommodation of new knowledge
(or its synthesis) on the part of the participants.
Frontiers of Language and
Teaching …...………………………. Volume 6 (2015)
100
Conclusion In regards to the interactive language functions of the
language used in the discussion board forums, it was observed that assertive
speech acts were most frequently present in the online interaction followed by
directives. From this it can be concluded, with respect to the first research
question, that speech acts in which the members of the virtual community
constructed and shared knowledge, used more assertive acts such as explaining,
giving suggestions or opinion, agreeing, reporting or stating, supporting,
conclusions, complaining (indirectly-expression of dissatisfaction) and
answering to queries. Second, they also used directive speech acts such as to
question, to ask, to advice, and/or to instruct other members of the virtual
community in order to construct and share new knowledge. This study also showed
that forums used as data for this study have evidence of the different phases
of knowledge construction, therefore proving that knowledge is indeed
constructed and shared in the online forums.
The findings of this study will aid educators and academicians in the
pedagogical aspect in using discussion boards in the teaching and learning
process. It is hoped the findings of
this study can be extended to the learning environment because over the years
the use of internet technology in classroom has gained popularity, and this can
be seen in the rapid growth in research into computer mediated discourse (Fitzpatrick
& Donnelly, 2010).
References
Akayoglu, S., & Altun, A.
(2009). The Functions of Negotiation of Meaning in Text-Based CMC. In Handbook
of Research on ELearning Methodologies for Language Acquisition (pp. 291–306).
IGI Global. Retrieved from http://www.mendeley.com/research/functions-negotiation-meaning-text-based-cmc/
Annand, D. (2011). Social Presence
within the Community of Inquiry Framework. The International Review of Research
in Open and Distance Learning, 12(5), 40–56. Byrne, C. L., Nei, D. S., Barrett,
J. D., Hughes, M. G., Davis, J. L., Griffith, J. a., … Mumford, M. D. (2013).
Online Ideology: A Comparison of Website Communication and Media Use. Journal
of Computer-Mediated Communication, 18(2), 25–39. doi:10.1111/jcc4.12003
Fitzpatrick, N., & Donnelly, R. (2010). Do
You See What I Mean ? Computer-Mediated Discourse Analysis Do You See What I
Mean ? Computer- mediated Discourse Analysis (pp. 0–17). IGI Global.
doi:10.4018/978-1-61520-879-1.ch004
Herring, S. C. (2004). Content
Analysis for New Media : Rethinking the Paradigm (pp. 47– 66). Bloomington.
Lester, J. N., & Paulus, T. M. (2011). Accountability and public displays
of knowing in an undergraduate computer-mediated communication context. Discourse
Studies, 13(6), 671–686. doi:10.1177/1461445611421361
Means, B., Toyama, Y., Murphy, R., Bakia, M.,
& Jones, K. (2009). Evaluation of Evidence- Based Practices in Online
Learning. Structure, 15(20), 94. Retrieved from http://newrepo.alt.ac.uk/629/
Paolillo, J. C. (2011). “ Conversational ” Codeswitching on Usenet and Internet
Relay Chat. Language@Internet, 8, article 3.
Paulus, T. M., & Phipps, G.
(2008). Approaches to case analyses in synchronous and asynchronous
environments. Journal of Computer-Mediated Communication, 13(2), 459–484.
doi:10.1111/j.1083-6101.2008.00405.x Pena-Shaff, J. B., & Nicholls, C.
(2004). Analyzing student interactions and meaning construction in computer
bulletin board discussions. Computers & Education, 42(3), 243–265. doi:10.1016/j.compedu.2003.08.003
Frontiers of Language and Teaching …...………………………. Volume 6 (2015)
Schallert, D. L., Chiang, Y. V.,
Park, Y., Jordan, M. E., Lee, H., Janne Cheng, A.-C., … Song, K. (2009). Being
polite while fulfilling different discourse functions in online classroom
discussions. Computers & Education, 53(3), 713–725.
doi:10.1016/j.compedu.2009.04.009 Searle, J. R. (1976). A classification of
illocutionary acts ’. In Language in society (Vol. 5, pp. 1–23). Cambridge
University Press.
Thanasingam, S., Kit, S., &
Soong, A. (2007). Interaction patterns and knowledge construction using
synchronous discussion forums and video to develop oral skills, 1002–1008.
Thayalan, X., & Shanthi, A. (2011).
Qualitative Assessment of Social Presence in Online Forums. In IEEE Colloquium
on Humanities, Science and Engineering research (pp. 438–440). Van Varik, F. J.
M., & van Oostendorp, H. (2013). Enhancing Online Community Activity: Development
and validation of the CA framework. Journal of Computer-Mediated Communication,
n/a–n/a. doi:10.1111/jcc4.12020
Wang, H. (2005). A Qualitative
Exploration of the Social Interaction in an Online Learning Community Haidong
Wang, 1, 79–88.
Article 2
An analysis of expressive speech acts in online task-oriented
interaction by university students
Marta Carreteroa
*, Carmen Maíz-Arévaloa
, M. Ángeles Martíneza
a
Universidad Complutense de Madrid, Facultad de Filología, Madrid,
28040, Spain
Abstract
This study explores the use of Expressive speech acts in a corpus
of online interaction involving three groups of university students in the area
of English Linguistics. The analysis focuses on the relative frequency of
occurrence of different subtypes of Expressives across the three subcorpora.
The influence of certain contextual variables such as multiculturality, age,
linguistic proficiency and group size seems to have a strong bearing on the
Expressives employed by each group.
© 2015 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of Universidad Pablo de Olavide.
Keywords: expressive speech acts; online collaborative writing;
multiculturality.
1. Online collaborative writing
Online collaborative writing is the term used to refer to the
computer-mediated joint production of a text by two or more authors with shared
ownership of the product (Storch, 2011). The use of online collaboration for pedagogical
purposes is connected to collaborative learning theories (Dillenbourg, 1999),
in turn deeply linked to socio-cultural and interactionist views of the
learning process (Piaget, 1928; Vygotsky, 1978). Among the many benefits of
collaborative learning we could mention stronger learner motivation and
improved social dynamics (Neumann & Hood, 2009, p. 383), as well as higher
involvement (Cole, 2009) and enhanced learner autonomy and control over the
learning process (Blake, 2011, p. 25; Leeming & Danino, 2012, p. 54). Blended
learning environments, now frequent in higher education settings using a
virtual campus, are those that combine face-to-face and computer-mediated
interaction. From the point of view of discourse organization, online
* Corresponding author. Tel.:+3491-394-53-83. Fax: +3491-394-57-62.
E-mail address: mcarrete@ucm.es
© 2015 The Authors. Published by Elsevier Ltd. This is an open
access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of Universidad Pablo de Olavide.
Marta Carretero et al. /
Procedia - Social and Behavioral Sciences 173 ( 2015 ) 186 – 190 187 written
interaction differs from face-to-face communication in several main respects.
One is related to the asynchronous nature of computer-mediated communication
(Herring et al., 2013). Secondly, online interaction cannot rely on many of the
multimodal resources used in face-to-face settings, such as eye-to-eye contact,
prosodic features, gestures, or body language (Herring et al., 2013), and this
endows the ongoing written production with a strong dependence on linguistic
organization, particularly when the conveyance of emotion is concerned.
Finally, many computer modes –wikis, e-forums, or blogs –imply the permanent
recording of the interaction in the form of a history log which allows
privileged access by analysts to the complete transcription of the linguistic
production of the participants.
These specific features of online communication are particularly
relevant to the present study, which focuses on the online written
collaboration of three groups of undergraduate and post-graduate university
students interacting in pedagogical e-forums for the subjects Discourse and
Text (D&T), Pragmatics (Pr), and Seminar on English Linguistics (SL), at
the Complutense University of Madrid, Spain. Although the language used is
strongly taskoriented, the analysis of the e-forum logs and their resulting
three written sub-corpora reveals a high presence of Expressives that seem to
perform the communicative function of making up for the absence of face-to-face
resources, in terms of smoothing transactional and task-oriented communication,
and building rapport among participants. The research questions are the
following:
a) Are Expressives equally frequent across the three sub-corpora,
and are they similarly distributed in terms of
sub-types such as Apologies, Thankings, Compliments, and so forth?
b) If this is not the case, which are the contextual variables with
a bearing on the choices made by participants?
2. Expressive speech acts
Expressives are one of the basic speech act types proposed in
Searle’s (1976) seminal classification, together with Representatives,
Directives, Commissives and Declaratives. Searle gives Apologizing,
Congratulating and Thanking as examples of Expressives. A preliminary study of
the data uncovered the need for the scope of Expressives to be enlarged, since
many speech acts were considered intuitively as expressive but did not fit into
any of Searle’s types.
Hence, other references were consulted: Bach & Harnish (1979),
Thomas (1995), Verschueren (1999), and especially Weigand (2010), who proposes
a speech act classification based on the notions of belief and desire. We adopted
as a criterial feature the concern with desire, or the predominance of desire
over belief. The resulting corpus-driven taxonomy included Expressives of two
general types: self-centred, pertaining to the speaker / writer’s feelings; and
other-centred, focusing on the addressee’s feelings. Self-centred Expressives
include:
Likings, which express positive emotional reactions (1); Concerns,
which express worries (2); and Wishes, which claim that the truth of the
proposition should (or should not) be the case (3):
1. I really like the classification. (SL)
2. I cannot recognize PCIs nor GCIs... It is difficult to see them...
the easiest are the presuppositions xD (Pr)
3. I wanted to answer to the last part of question two and question
three but I really cannot think any longer. (Pr)
Other-oriented Expressives include Apologies, Compliments and
Thankings, which correspond to Searle’s expressives mentioned above, as well as
other subtypes: Reassurings, which aim at comforting the addressee by diminishing
his/her feeling of guilt (4); and Reproaches, which may be seen as the negative
counterpart of Compliments (5):
4. Don't worry because everything is finished and sent (D&T)
5. I feel like I'm having pretty much of a monologue here…
(D&T)
Finally, our scope of Expressives also includes speech acts of
other kinds that focus on the speaker/writer’s emotional involvement by linguistic
or typographical means, concretely interjections such as oh, exclamation marks,
emphatic do, accumulation of evaluative expressions, repetition of a letter or
of a question mark, capitalization, and the use of emoticons (Yus, 2011).
Utterances containing any of these marks have also been considered as Expressives,
in addition to the subtypes described above. In the corpus, occurrences were
found of reinforced Greetings (6), Assertions (7), Directives (8) and
Commissives (9). Due to its importance in the three subcorpora, the category of
Agreement was split from Assertions at large and conferred the status of an
individual subtype (10).
6. Hello everybody! (D&T)
7. I have finished my part! (D&T)
8. Suggestions would be very welcome!! (Pr)
9. I'm going to try to post my ideas tomorrow! (D&T)
10. I agree with everything you've said :D (SL)
3. Methodology
The data used in the study consists in a 79,699-word long corpus,
made up of three subcorpora containing the eforum written interaction of 83
university students belonging to one of the following groups: 64 undergraduate students
taking an optional course on English Discourse and Text (Subcorpus D&T:
40,226 words) (Martínez,2014); 9 undergraduate students from an evening group
taking an obligatory course on Pragmatics (Subcorpus Pr: 14,119 words)
(Carretero, 2014); and 10 post-graduate students following the Master’s Seminar
on English Linguistics (Subcorpus SL: 25,354 words) (Maíz-Arévalo, 2014). Each
group of participants was subdivided into smaller groups of three or four
students, randomly created by Virtual Campus itself. Each of these smaller
groups had to carry out one or two collective assignments. However, they were
specifically asked not to do these collaborative exercises in the traditional
face-to-face way but online, by means of an e-forum where they could negotiate
and discuss for one week before producing a final written report. None of the
participants was informed a priori of their participation in this research
project, in order to avoid unnaturally biased interactions. However, once the
activity was over, participants gave their written consent. In any case,
pseudonyms were used to preserve their identity.
The unbalanced number and age of participants could not be
controlled for the present research but implied two interesting variables to
take into account when analysing the results. A third variable was the
participants’ level of English. Although quite advanced in general terms, the
undergraduate students’ level ranged from B2 to C1 according to the Common
European Framework of Reference (2001), whilst the postgraduates’ linguistic proficiency
ranged between C1 and C2. A fourth major difference was the high degree of
interculturality present in the Master group, which included Russian, Korean,
Arabic, Polish and Spanish students, as opposed to the undergraduate groups,
consisting mostly of Spaniards.
4. Data analysis
Table 1 presents the Expressive subtypes found in the corpus,
accompanied by the number of tokens (N), together with the corresponding
percentages across the three subcorpora. The analysis uncovers two main
similarities: the first is a predominance of other-oriented over self-oriented speech
acts. This tendency may well be due to the students’ concern with assuring a
good rapport, rather than focusing on their own feelings. Another reason might
be the blended nature of the learning context. These otherfocused Expressives
are enhanced in the data by the use of typographic signs like exclamation marks
or emoticons, as in the Thanking in (11) or the Apology in (12):
(11) Thanks, Anat for offering to put the analysis in the final
document! - (SL)
(12) Hi, sorry for being this late, I've been having problems with
my internet connection at home - (Pr)
Marta Carretero et al. / Procedia
- Social and Behavioral Sciences 173 ( 2015 ) 186 – 190 189
Table 1. Cross-comparative view of results.
Speech acts Corpus Pr % Corpus SL % Corpus D&T %
Apology 25.56 (N=34) 10.72 (N=25) 10.90 (N=52)
Compliment 16.54 (N = 22) 21.00 (N=49) 14.89 (N=71)
Greeting 9.02 (N = 12) 13.73 (N=32) 16.14 (N=77)
Wish 6.77 (N = 9) 3.43 (N=8) 17.20 (N=82)
Thanking 6.77(N = 9) 18.88 (N=44) 19.91 (N=95)
Liking 0.00 (N= 0) 4.29 (N=10) 0.42 (N=2)
Concern 10.53 (N = 14) 1.71 (N = 4) 2.93 (N=14)
Reproach 4.51 (N = 6) 0.85 (N = 2) 5.66 (N=27)
Directive 7.52 (N = 10) 13.30 (N=31) 4.40 (N=21)
Agreement 4.51 (N = 6) 3.00 (N=7) 3.14 (N=15)
Assertion 3.01 (N = 4) 6.86 (N=16) 1.44 (N=7)
Commissive 2.26 (N = 3) 1.71 (N=4) 2.10 (N=10)
Reassuring 3.01 (N = 4) 0.42 (N =1) 0.63 (N=3)
TOTAL 100 (N=133) 100 (N=233) 100 (N=476)
The second similarity lies in the high degree of
conventionalization found in the most recurrent subtypes:Compliments often
contain adjectives such as excellent, fine, good, interesting or perfect;
Greetings, hello, hi or hey; Thankings, thank you or thanks; and Apologies,
sorry. This conventionalization may be accounted for by the formulaic nature of
these expressions as well as the priority given to the performance of the task,
which needed quick and effective rapport building other-centeredness, as in
example (13):
11. Hey guys! (D&T)
In spite of these similarities, the three subcorpora differ in some
respects. For instance, expressions of Concern are overwhelmingly higher in
Subcorpus Pr than in the other two, as can be observed in table 1. Subcorpus
SL, on the other hand, displays a remarkably higher presence of Compliments,
Directives, and Assertions, while ranking below average in Wish, Concern, and
Reproach. Finally, Subcorpus D&T presents above average percentages in Greetings
and Wishes, and is also high in Thanks and Reproaches, but has a relatively
lower presence of Directives, Assertions, and Agreements typographically marked
as Expressives. The reasons for these differences seem to be related to the
four contextual variables mentioned in Section 3: general group size, age,
linguistic proficiency and cultural homogeneity.
The larger size of the morning group (64 students) in contrast to
the rather small evening groups (9 and 10 students, respectively) could have
accounted for the higher number of Reproaches in Subcorpus D&T. In this
group, face-threat might have been perceived with lower intensity, given the
lack of real face-to-face contact, as in (14):
12. I hope that the other two participants of the group say
something, if not… I think we must talk to T[eacher]
(D&T)
Age may have a bearing on the differences between the two
undergraduate subgroups: the evening students in Subcorpus Pr often produce
Concerns and Apologies. By contrast, their morning D&T younger counterparts
seem to favour “wishful thinking”, hence the high frequency of conventionalized
Wishes, as in (15):
13. I hope you can give me an idea and do it together (D&T)
As for linguistic proficiency, it appears to be connected with the
different percentage of Compliments across the three subcorpora: they rank
slightly higher in the master Subcorpus SL (21%), and gradually decrease along
the proficiency scale, with 16.54% in Subcorpus Pr and 14.89% in Subcorpus
D&T.
Finally, interculturality seems to influence the preference for
typographic signs like emoticons in Subcorpus SL, produced by students
belonging to very different nationalities, who resorted to typographic signs to
build rapport, as in example (16):
14. For question 2, I tried to summarize before the table. It seems
logical to put words before the table. ;-) (SL)
In addition, this multicultural group seems particularly fond of
Thanking in its formulaic realization. It could be argued that these Master
students issue thanks on a British English basis, since they were perfectly
aware that English was being used as a lingua franca.
5. Conclusions
The analysis carried out in this paper covered expressive speech
acts in a corpus consisting of e-forum history logs produced by three groups of
students in English linguistics. The study revealed two common features: predominance
of other-oriented over self-oriented Expressives and a high degree of
conventionalization in the linguistic realization of the four most frequent
subtypes (Thankings, Apologies, Greetings and Compliments). The analysis also
showed remarkable differences in terms of frequency of use, concrete linguistic
realizations of individual subtypes, and the use of typographic marks. These
differences may be accounted for by the influence of contextual variables,
namely group size, age, linguistic proficiency and cultural homogeneity.
References
Bach, K., & Harnish, R. M.
(1979). Linguistic communication and speech acts. Cambridge, USA and London,
UK: MIT Press.
Blake, R. J. (2011). Current trends
in online language learning. Annual Review of Applied Linguistics, 31, 19-35.
Carretero, M. (2014). The Virtual
Campus in the teaching of English pragmatics in BA’s last year: a description
of two e-forum-based activities.
In M. Ángeles Martínez (coord.) The
use of institutional e-forums for online collaborative writing activities in
the field of discourse analysis
and English linguistics. Universidad
Complutense de Madrid: Proyectos de Innovación y Mejora de la Calidad Docente
(PIMCD2012-
2013) (pp. 18-29). Madrid: Editorial
del Economista.
Cole, M. (2009). Using wiki
technology to support student engagement: Lessons from the trenches. Computers
and Education, 52, 141-146.
Dillenbourg, P. (1999).
Collaborative-learning: Cognitive and computational approaches. Oxford:
Elsevier.
Herring, S. C., Stein D. &
Virtanen T. (2013). Handbook of pragmatics of computer-mediated communication.
Berlin: Mouton.
Leeming, D. E. & Danino N.
(2012). Breaking barriers: A case study of culture and Facebook usage. Journal
of Modern Languages and
International Studies, 1(1), 52-64.
Maíz-Arévalo, C. (2014). Using
e-forums to develop intercultural pragmatic awareness. In M. Ángeles Martínez
(coord.) The use of institutional
e-forums for online collaborative
writing activities in the field of discourse analysis and English linguistics.
Universidad Complutense de
Madrid: Proyectos de Innovación y
Mejora de la Calidad Docente (PIMCD2012-2013) (pp. 43-53). Madrid: Editorial
del Economista.
Martínez, M. Á. (2014). Teaching
discourse analysis through online collaborative writing in a tertiary
educational setting: a focus on assessment.
In M. Ángeles Martínez (coord.) The
use of institutional e-forums for online collaborative writing activities in
the field of discourse analysis
and English linguistics. Universidad
Complutense de Madrid: Proyectos de Innovación y Mejora de la Calidad Docente
(PIMCD2012-
2013) (pp. 62-78). Madrid: Editorial
del Economista.
Neumann, D. L. & Hood M. (2009).
The effects of using a wiki on student engagement and learning of report
writing skills in a university
statistics course. Australasian
Journal of Educational Technology, 25(5), 382-398.
Piaget, J. (1928). The language and
thought of the child. New York: Harcourt.
Searle, J. R. (1976). A
classification of illocutionary acts. Language in Society, 5(1), 1-23.
Storch, N. (2011). Collaborative
writing in L2 contexts: Processes, outcomes, and future directions. Annual
Review of Applied Linguistics, 31(1),
275-288.
Thomas, J. (1995). Meaning in
interaction: An introduction to pragmatics. London and New York: Longman.
Verschueren, J. (1999).
Understanding pragmatics (vol. 31). London: Arnold.
Vygotsky, L. S. (1978). Mind and
society: The development of higher mental processes. Cambridge, Ma: Harvard
University Press.
Weigand, E. (2010). Dialogue: The
mixed game (vol. 10). Philadelphia: John Benjamins Publishing.
Yus, F. (2011). Cyberpragmatics: Internet-mediated
communication in context . Philadelphia: John Benjamins Publishing.
Article 3
In
Proceedings of the 2011 Conference on Empirical Methods in Natural Language
Processing (EMNLP-2011). Classifying Sentences as Speech Acts in Message Board
Posts
Ashequl
Qadir and Ellen Riloff School of Computing University of Utah Salt Lake City,
UT 84112 {asheq,riloff}@cs.utah.edu
Abstract
This
research studies the text genre of mes- sage board forums, which contain a mix-
ture of expository sentences that present fac- tual information and
conversational sentences that include communicative acts between the writer and
readers. Our goal is to create sentence classifiers that can identify whether a
sentence contains a speech act, and can recognize sentences containing four
different speech act classes: Commissives, Directives, Expressives, and
Representatives. We con- duct experiments using a wide variety of fea- tures,
including lexical and syntactic features, speech act word lists from external
resources, and domain-specific semantic class features. We evaluate our results
on a collection of mes- sage board posts in the domain of veterinary medicine.
1
Introduction
In
the 1990’s, the natural language processing com- munity shifted much of its
attention to corpus-based learning techniques. Since then, most of the text
cor- pora that have been annotated and studied are collec- tions of expository
text (e.g., news articles, scientific literature, etc.). The intent of
expository text is to present or explain information to the reader. In re- cent
years, there has been a growing interest in text genres that originate from Web
sources, such as we- blogs and social media sites (e.g., tweets). These text
genres offer new challenges for NLP, such as the need to handle informal and
loosely grammatical text, but they also pose new opportunities to study
discourse
and pragmatic phenomena that are funda- mentally different in these genres.
Message boards are common on the WWW as a forum where people ask questions and
post com- ments to members of a community. They are typ- ically devoted to a
specific topic or domain, such as finance, genealogy, or Alzheimer’s disease.
Some message boards offer the opportunity to pose ques- tions to domain
experts, while other communities are open to anyone who has an interest in the
topic. From a natural language processing perspective, message board posts are
an interesting hybrid text genre because they consist of both expository text
and conversational text. Most obviously, the conver- sations appear as a
thread, where different people respond to each other’s questions in a sequence
of posts. Studying the conversational threads, however, is not the focus of
this paper. Our research addresses the issue of conversational pragmatics
within indi- vidual message board posts. Most message board posts contain both
exposi- tory sentences as well as speech acts. The person posting a message
(the “writer”) often engages in speech acts with the readers. The writer may
explic- itly greet the readers (“Hi everyone!”), request help from the readers
(“Anyone have a suggestion?”), or commit to a future action (“I promise I will
report back soon.”). But most posts contain factual infor- mation as well, such
as general knowledge or per- sonal history describing a situation, experience,
or predicament. Our research goals are twofold: (1) to distin- guish between
expository sentences and speech act sentences in message board posts, and (2)
to clas-
sify
speech act sentences into four types: Com- missives, Directives, Expressives,
and Representa- tives, following Searle’s original taxonomy (Searle, 1976).
Speech act classification could be useful for many applications. Information
extraction sys- tems could benefit from filtering speech act sen- tences (e.g.,
promises and questions) so that facts are only extracted from the expository
text. Identifying Directive sentences could be used to summarize the questions
being asked in a forum over a period of time. Representative sentences could be
extracted to highlight the conclusions and beliefs of domain experts in
response to a question. In this paper, we present sentence classifiers that can
identify speech act sentences and classify them as Commissive, Directive,
Expressive, and Repre- sentative. First, we explain how each speech act class
is manifested in message board posts, which can be different from how they
occur in spoken dia- logue. Second, we train classifiers to identify speech act
sentences using a variety of lexical, syntactic, and semantic features.
Finally, we evaluate our sys- tem on a collection of message board posts in the
domain of veterinary medicine.
2
Related Work
There
has been relatively little work on applying speech act theory to written text
genres, and most of the previous work has focused on email classi- fication.
Cohen et al. (2004) introduced the notion of “email speech acts” defined as
specific verb-noun pairs following a pre-designed ontology. They ap- proached
the problem as a document classification task. Goldstein and Sabin (2006)
adopted this no- tion of email acts (Cohen et al., 2004) but focused on verb
lexicons to classify them. Carvalho and Cohen (2005) presented a classification
scheme us- ing a dependency network, capturing the sequential correlations with
the context emails using transition probabilities from or to a target email.
Carvalho and Cohen (2006) later employed N-gram sequence fea- tures to
determine which N-grams are meaningfully related to different email speech acts
with a goal towards improving their earlier email classification based on the
writer’s intention. Lampert et al. (2006) performed speech act clas- sification
in email messages following a verbal re-
sponse
modes (VRM) speech act taxonomy. They also provided a comparison of VRM
taxonomy with Searle’s taxonomy (Searle, 1976) of speech act classes. They
evaluated several machine learning al- gorithms using syntactic, morphological,
and lexi- cal features. Mildinhall and Noyes (2008) presented a stochastic
speech act model based on verbal re- sponse modes (VRM) to classify email
intentions. Some research has considered speech act classes in other means of
online conversations. Twitchell and Jr. (2004) and Twitchell et al. (2004)
employed speech act profiling by plotting potential dialogue categories in a
radar graph to classify conversa- tions in instant messages and chat rooms.
Nas- tri et al. (2006) performed an empirical analysis of speech acts in the
away messages of instant mes- senger services to achieve a better understanding
of the communication goals of such services. Ravi and Kim (2007) employed
speech act profiling in online threaded discussions to determine message roles
and to identify threads with questions, answers, and unanswered questions. They
designed their own speech act categories based on their analysis of stu- dent
interactions in discussion threads. The work most closely related to ours is
the re- search of Jeong et al. (2009) on semi-supervised speech act recognition
in both emails and forums. Like our work, their research also classifies indi-
vidual sentences, as opposed to entire documents. However, they trained their
classifier on spoken telephone (SWBD-DAMSL corpus) and meeting (MRDA corpus)
conversations and mapped the la- belled dialog act classes of these corpora to
12 di- alog act classes that they found suitable for email and forum text
genres. These dialog act classes (ad- dressed as speech acts by them) are
somewhat differ- ent from Searle’s original speech act classes. They also used
substantially different types of features than we do, focusing primarily on
syntactic subtree structures.
3
Classifying Speech Acts in Message Board Posts
3.1
Speech Act Class Definitions
Searle’s
(Searle, 1976) early research on speech acts was seminal work in natural
language processing that opened up a new way of thinking about con-
versational
dialogue and communication. Our goal was to try and use Searle’s original
speech act def- initions and categories as the basis for our work to the
greatest extent possible, allowing for some inter- pretation as warranted by
the WWW message board text genre. For the purposes of defining and evaluating
our work, we created detailed annotation guidelines for four of Searle’s speech
act classes that commonly occur in message board posts: Commissives, Direc-
tives, Expressives, and Representatives. We omitted the fifth of Searle’s
original speech act classes, Dec- larations, because we virtually never saw
declara- tive speech acts in our data set.1 The data set used in our study is a
collection of message board posts in the domain of veterinary medicine. We
designed our definitions and guidelines to reflect language use in the text genre
of message board posts, trying to be as domain-independent as possible so that
these defini- tions should also apply to message board texts rep- resenting other
topics. However, we give examples from the veterinary domain to illustrate how
these speech act classes are manifested in our data set. Commissives: A
Commissive speech act oc- curs when the speaker commits to a future course of
action. In conversation, common Commissive speech acts are promises and
threats. In message boards, these types of Commissives are relatively rare.
However, we found many statements where the main purpose was to confirm to the
readers that the writer would perform some action in the future. For example, a
doctor may write “I plan to do surgery on this patient tomorrow” or “I will
post the test results when I get them later today”. We viewed such state- ments
as implicit commitments to the reader about intended actions. We also considered
decisions not to take an action as Commissive speech acts (e.g., “I will not do
surgery on this cat because it would be too risky.”). However, statements
indicating that an action will not occur because of circumstances be- yond the
writer’s control were considered to be fac- tual statements and not speech acts
(e.g., “I cannot do an ultrasound because my machine is broken.”). Directives:
A Directive speech act occurs when
1Searle
defines Declarative speech acts as statements that bring about a change in
status or condition to an object by virtue of the statement itself. For
example, a statement declaring war or a statement that someone is fired.
the
speaker expects the listener to do something as a response. For example, the
speaker may ask a question, make a request, or issue an invitation. Di- rective
speech acts are common in message board posts, especially in the initial post
of each thread when the writer explicitly requests help or advice re- garding a
specific topic. Many Directive sentences are posed as questions, so they are
easy to identify by the presence of a question mark. However, the language in
message board forums is informal and often ungrammatical, so many Directives
are posed as a question but do not end in a question mark (e.g., “What do you
think.”). Furthermore, many Direc- tive speech acts are not stated as a
question but as a request for assistance. For example, a doctor may write “I
need your opinion on what drug to give this patient.” Finally, some sentences
that end in ques- tion marks are rhetorical in nature and do not repre- sent a
Directive speech act, such as “Can you believe that?”. Expressives: An
Expressive speech act occurs in conversation when a speaker expresses his or
her psychological state to the listener. Typical cases are when the speaker
thanks, apologizes, or welcomes the listener. Expressive speech acts are common
in message boards because writers often greet readers at the beginning of a
post (“Hi everyone!”) or ex- press gratitude for help from the readers (“I
really appreciate the suggestions.”). We also found Ex- pressive speech acts in
a variety of other contexts, such as apologies. Representatives: According to
Searle, a Rep- resentative speech act commits the speaker to the truth of an
expressed proposition. It represents the speaker’s belief of something that can
be evaluated to be true or false. These types of speech acts were less common
in our data set, but some cases did ex- ist. In the veterinary domain, we
considered sen- tences to be a Representative speech act when a doctor
explicitly confirmed a diagnosis or expressed their suspicion or hypothesis
about the presence (or absence) of a disease or symptom. For example, if a
doctor writes that “I suspect the patient has pancre- atitis.” then this represents
the doctor’s own propo- sition/belief about what the disease might be. Many
sentences in our data set are stated as fact but could be reasonably inferred
to be speech acts. For example, suppose a doctor writes “The cat has
pancreatitis.”.
It would be reasonable to infer that the doctor writing the post diagnosed the
cat with pancreatitis. And in many cases, that is true. How- ever, we saw many
posts where that inference would have been wrong. For example, the following
sen- tence might say “The cat was diagnosed by a pre- vious vet but brought to
me due to new complica- tions” or “The cat was diagnosed with it 8 years ago as
a kitten in the animal shelter”. Consequently, we were very conservative in
labelling sentences as Representative speech acts. Any sentence presented as
fact was not considered to be a speech act. A sen- tence was only labelled as a
Representative speech act if the writer explicitly expressed his belief.
3.2
Features for Speech Act Classification
To
create speech act classifiers, we designed a vari- ety of lexical, syntactic,
and semantic features. We tried to capture linguistic properties associated
with speech act expressions as well as discourse prop- erties associated with
individual sentences and the message board post as a whole. We also incorpo-
rated speech act word lists that were acquired from external resources, and
used two types of seman- tic features to represent semantic entities associated
with the veterinary domain. Except for the semantic features, all of our
features are domain-independent so should be able to recognize speech act
sentences across different domains. We experimented with domain-specific
semantic features to test our hy- pothesis that Commissive speech acts can be
asso- ciated with domain-specific semantic entities. For the purposes of
analysis, we partition the fea- ture set into three groups: Lexical and
Syntactic (LexSyn) Features, Speech Act Clue Features, and Semantic Features.
Unless otherwise noted, all of the features had binary values indicating the
pres- ence or absence of that feature.
3.2.1
Lexical and Syntactic Features
We
designed a variety of features to capture lexical and syntactic properties of
words and sentences. We described the feature set below, with the features cat-
egorized based on the type of information that they capture. Unigrams: We
created bag-of-word features rep- resenting each unigram in the training set.
Numbers were replaced with a special # token.
Personal
Pronouns: We defined three features to look for the presence of a 1st person
pronoun, 2nd person pronoun, and 3rd person pronoun. We in- cluded the
subjective, objective, and possessive form of each pronoun (e.g., he, him, and
his). Tense: Speech acts such as Commissives can be related to tense. We
created three features to iden- tify verb phrases that occur in the past,
present, or future tense. To recognize tense, we followed the rules defined by
Allen (1995). Tense + Person: We created four features that re- quire the
presence of a first person subjective pro- noun (I, we) within a two word window
on the left of a verb phrase matching one of four tense representa- tions:
past, present, future, and present progressive (a subset of the more general
present tense represen- tation). Modals: One feature indicates whether the sen-
tence contains a modal (may, must, shall, will, might, should, would, could).
Infinitive VP: One feature looks for an infinitive verb phrase (‘to’ followed by
a verb) that is preceded by a first person pronoun (I, we) within a three word
window on the left. This feature tries to capture common Commissive expressions
(e.g., “I definitely plan to do the test tomorrow.”). Plan Phrases: Commissives
are often expressed as a plan, so we created a feature that recognizes four
types of plan expressions: “I am going to”, “I am planning to”, “I plan to”,
and “My plan is to”. Sentence contains Early Punctuation: One fea- ture checks
for the following punctuation marks within the first three tokens of the
sentence: , : ! This feature was designed to recognize greetings, such as:
“Hi,” , or “Hiya everyone !”. Sentence begins with Modal/Verb: One feature
checks if a sentence begins with a modal or verb. The intuition is to capture
interrogative and impera- tive sentences, since they are likely to be
Directives. Sentence begins with WH Question: One fea- ture checks if a
sentence begins with a WH question word (Who, When, Where, What, Which, What,
How). Neighboring Question: One feature checks whether the following sentence
contains a question mark ‘?’. We observed that in message boards, Di- rectives
often occur in clusters.
Sentence
Position: Four binary features repre- sent the relative position of the
sentence in the post. One feature indicates whether it is the first sentence,
one feature indicates whether it is the last sentence, one feature indicates
whether it is the second to last sentence, and one feature indicates whether
the sen- tence occurs in the bottom 25% of the message. The motivation for
these features is that Expressives of- ten occur at the beginning and end of
the post, and Directives tend to occur toward the end.
Number
of Verbs: One feature represents the number of verbs in the sentence using four
possible values: 0, 1, 2, >2. Some speech acts classes (e.g., Expressives)
may occur with no verbs, and rarely occur in long, complex sentences.
3.2.2
Speech Act Word Clues
We
collected speech act word lists (mostly verbs) from two external sources. In
Searle’s original pa- per (Searle, 1976), he listed words that he consid- ered
to be indicative of speech acts. We discarded a few that we considered to be
overly general, and we added a few additional words. We also collected a list
of speech act verbs published in (Wierzbicka, 1987). The details for these
speech act clue lists are given below. Our system recognized all derivations of
these words.
Searle
Keywords: We created one feature for each speech act class. The Representative
keywords were: (hypothesize, insist, boast, complain, con- clude, deduce,
diagnose, and claim). We discarded 3 words from Searle’s list (suggest, call,
believe) and added 2 new words, assume and suspect. The Direc- tive keywords
were: (ask, order, command, request, beg, plead, pray, entreat, invite, permit,
advise, dare, defy, challenge). We added the word please. The Expressives
keywords were: (thank, apolo- gize, congratulate, condole, deplore, welcome).
We added the words appreciate and sorry. Searle did not provide any hint on
possible indicator words for Commissives, so we manually defined five likely
Commissive keywords: (plan, commit, promise, to- morrow, later).
Wierzbicka
Verbs: We created one feature that included 228 speech act verbs listed in the
book “English speech act verbs: a semantic dictionary”
(Wierzbicka,
1987)2.
3.2.3
Semantic Features All of the previous features are domain- independent and
should be useful for identifying speech acts sentences across many domains.
How- ever, we hypothesized that semantic entities may correlate with speech
acts within a particular do- main. For example, consider medical domains. Rep-
resentative speech acts may involve diagnoses and hypotheses regarding diseases
and symptoms. Sim- ilarly, Commissive speech acts may reveal a doc- tor’s plan
or intention regarding the administration of drugs or tests. Thus, it may be
beneficial for a classifier to know whether a sentence contains cer- tain
semantic entities. We experimented with two different sources of semantic
information. Semantic Lexicon: Basilisk (Thelen and Riloff, 2002) is a
bootstrapping algorithm that has been used to induce semantic lexicons for
terrorist events (Thelen and Riloff, 2002), biomedical concepts (McIntosh,
2010), and subjective/objective nouns for opinion analysis (Riloff et al.,
2003). We ran Basilisk over our collection of 15,383 veteri- nary message board
posts to create a semantic lex- icon for veterinary medicine. As input,
Basilisk requires seed words for each semantic category. To obtain seeds, we
parsed the corpus using a noun phrase chunker, sorted the head nouns by fre-
quency, and manually identified the 20 most fre- quent nouns belonging to four
semantic categories: DISEASE/SYMPTOM, DRUG, TEST, and TREAT- MENT. However, the
induced TREATMENT lexicon was of relatively poor quality so we did not use it.
The DISEASE/SYMPTOM lexicon appeared to be of good quality, but it did not
improve the performance of our speech act classifiers. We suspect that this is
due to the fact that diseases were not distinguised from symptoms in our
lexicon.3 Representative speech acts are typically associated with disease
diagnoses
2openlibrary.org/b/OL2413134M/English_
speech_act_verbs 3We induced a single lexicon for diseases and symptoms be-
cause it is difficult to draw a clear line between them seman- tically. A
veterinary consultant explained to us that the same term (e.g., diabetes) may be
considered a symptom in one con- text if it is secondary to another condition
(e.g., pancreatitis) but a disease in a different context if it is the primary
diagnosis.
and
hypotheses, rather than individual symptoms. In the end, we only used the DRUG
and TEST se- mantic lexicon in our classifiers. We used all 1000 terms in the
DRUG lexicon, but only used the top 200 TEST words because the quality of the
lexicon seemed questionable after that point. Semantic Tags: We also used
bootstrapped con- textual semantic taggers (Huang and Riloff, 2010) that had
been previously trained for the domain of veterinary medicine. These taggers
assign seman- tic class labels to noun phrase instances based on the
surrounding context in a sentence. The tag- gers were trained on 4,629
veterinary message board posts using 10 seed words for each semantic cate- gory
(see (Huang and Riloff, 2010) for details). To ensure good precision, only tags
that have a confi- dence value ≥ 1.0 were used. Our speech act classi- fiers used
the tags associated with two semantic cat- egories: DRUG and TEST.
3.3
Classification
To
create our classifiers, we used the Weka (Hall et al., 2009) machine learning
toolkit. We used Sup- port Vector Machines (SVMs) with a polynomial kernel and
the default settings supplied by Weka. Because a sentence can include multiple
speech acts, we created a set of binary classifiers, one for each of the four
speech act classes. All four classifiers were applied to each sentence, so a
sentence could be as- signed multiple speech act classes.
4
Evaluation
4.1
Data Set
Our
data set consists of message board posts from the Veterinary Information
Network (VIN), which is a web site (www.vin.com) for professionals in vet-
erinary medicine. Among other things, VIN hosts message board forums where
veterinarians and other veterinary professionals can discuss issues and pose
questions to each other. Over half of the small an- imal veterinarians in the
U.S. and Canada use the VIN web site. We obtained 15,383 VIN message board
threads representing three topics: cardiology, endocrinol- ogy, and feline
internal medicine. We did basic cleaning, removing html tags and tokenizing
num- bers. We then applied the Stanford part-of-speech
tagger
(Toutanova et al., 2003) to each sentence to obtain part-of-speech tags for the
words. For our ex- periments, we randomly selected 150 message board threads
from this collection. Since the goal of our work was to study speech acts in
sentences, and not the conversational dialogue between different writ- ers, we used
only the initial post of each thread. These 150 message board posts contained a
total of 1,956 sentences, with an average of 13.04 sentences per post. In the
next section, we explain how we manually annotated each sentence in our data
set to create gold standard speech act labels.
4.2
Gold Standard Annotations
To
create training and evaluation data for our re- search, we asked two human
annotators to manually label sentences in our message board posts. Iden-
tifying speech acts is not always obvious, even to people, so we gave them
detailed annotation guide- lines describing the four speech act classes
discussed in Section 3.1. Then we gave them the same set of 50 message board
posts from our collection to an- notate independently. Each annotator was told
to assign one or more speech act classes to each sen- tence (COM, DIR, EXP,
REP), or to label the sen- tence as having no speech acts (NONE). The vast
majority of sentences had either no speech acts or at most one speech act, but
a small number of sen- tences contained multiple types of speech acts. We
measured the inter-annotator agreement of the two human judges using the kappa
(κ) score (Car- letta, 1996). However, kappa agreement scores are only
applicable to labelling schemes where each in- stance receives a single label.
Therefore we com- puted kappa agreement in two different ways to look at the
results from two different perspectives. In the first scheme, we discarded the
small number of sen- tences that had multiple speech act labels and com- puted kappa
on the rest.4 This produced a kappa score of .95, suggesting extremely high
agreement. However, over 70% of the sentences in our data set have no speech
act at all, so NONE was by far the most common label. Consequently, this
agreement score does not necessarily reflect how consistently the judges agreed
on the four speech act classes.
4Of
the 594 sentences in these 50 posts, only 22 sentences contained multiple
speech act classes.
In
the second scheme, we computed kappa for each speech act category independently.
For each category C, the judges were considered to be in agreement if both of
them assigned category C to the sentence or if neither of the judges assigned
cat- egory C to the sentence. Table 1 shows the κ agree- ment scores using this
approach.
Speech
Act Kappa (κ) score Expressive .97 Directive .94 Commissive .81 Representative
.77
Table
1: Inter-annotator (κ) agreement
Inter-annotator
agreement was very high for both the Expressive and Directive classes.
Agreement was lower for the Commissive and Representative classes, but still
relatively good so we felt comfort- able that we had high-quality annotations.
To create our final data set, the two judges adjudi- cated their disagreements
on this set of 50 posts. We then asked each annotator to label an additional
(dif- ferent) set of 50 posts each. All together, this gave us a gold standard
data set consisting of 150 anno- tated message board posts. Table 2 shows the
distri- bution of speech act labels in our data set. 71% of the sentences did not
include any speech acts. These were usually expository sentences containing
factual information. 29% of the sentences included one or more speech acts, so
nearly 1 3 of the sentences were conversational in nature. Directive and
Expressive speech acts are by far the most common, with nearly 26% of all
sentences containing one of these speech acts. Commissive and Representative
speech acts are less common, each occurring in less than 3% of the sentences.5
4.3
Experimental Results
4.3.1
Speech Act Filtering
For
our first experiment, we created a speech act filtering classifier to distinguish
sentences that con- tain one or more speech acts from sentences that do not
contain any speech acts. Sentences labelled as
5These
numbers do not add up to 100% because some sen- tences contain multiple speech
acts.
Speech
Act # sentences distribution None 1397 71.42% Directive 311 15.90% Expressive
194 9.92% Representative 57 2.91% Commissive 51 2.61%
Table
2: Speech act class distribution in our data set.
having
one or more speech acts were positive in- stances, and sentences labelled as
NONE were neg- ative instances. Speech act filtering could be useful for many
applications, such as information extrac- tion systems that only seek to
extract facts. For ex- ample, information may be posed as a question (in a
Directive) rather than a fact, information may be mentioned as part of a future
plan (in a Commis- sive) that has not actually happened yet, or informa- tion
may be stated as a hypothesis or suspicion (in a Representative) rather than as
a fact. We performed 10-fold cross validation on our set of 150 annotated
message board posts. Initially, we used all of the features defined in Section
3.2. How- ever, during the course of our research we discov- ered that only a
small subset of the lexical and syn- tactic features seemed to be useful, and
that remov- ing the unnecessary features improved performance. So we created a
subset of minimal lexsyn features, which will be described in Section 4.3.2.
For speech act filtering, we used the minimal lexsyn features plus the speech
act clues and semantic features.6
Class
P R F Speech Act .86 .83 .84 No Speech Act .93 .95 .94
Table
3: Precision, Recall, F-measure for speech act fil- tering.
Table
3 shows the performance for speech act filtering with respect to Precision (P),
Recall (R), and F-measure score (F).7 The classifier performed well, recognizing
83% of the speech act sentences with 86% precision, and 95% of the expository
(no
6This
is the same feature set used to produce the results for row E of Table 4. 7We
computed an F1 score with equal weighting of preci- sion and recall.
Commissives
Directives Expressives Representatives Features P R F P R F P R F P R F
Baselines Com baseline .45 .08 .14 - - - - - - - - - Dir baseline - - - .97 .73
.83 - - - - - - Exp baseline 1 - - - - - - .58 .18 .28 - - - Exp baseline 2 - -
- - - - .97 .86 .91 - - - Rep baseline - - - - - - - - - 1.0 .05 .10 Classifiers
U Unigram .45 .20 .27 .87 .84 .85 .97 .88 .92 .32 .12 .18 A U+all lexsyn .52
.33 .40 .87 .84 .86 .98 .88 .92 .30 .14 .19 B U+minimal lexsyn .59 .33 .42 .87
.85 .86 .98 .88 .92 .32 .14 .20 C B+speechActClues .57 .31 .41 .86 .84 .85 .97
.91 .94 .33 .16 .21 D C+semTest .64 .35 .46 .87 .84 .85 .97 .91 .94 .33 .16 .21
E D+semDrug .63 .39 .48 .86 .84 .85 .97 .91 .94 .32 .16 .21
Table
4: Precision, Recall, F-measure for four speech act classes. The highest F
score for each category appears in boldface.
speech
act) sentences with 93% precision.
4.3.2
Speech Act Categorization
BASELINES
Our
next set of experiments focused on labelling sentences with the four specific
speech act classes: Commissive, Directive. Expressive, and Represen- tative. To
assess the difficulty of identifying each speech act category, we created
several simple base- lines using our intuitions about each category. For
Commissives, we created a heuristic to cap- ture the most obvious cases of
future tense (because Commissive speech acts represent a writer’s com- mitment
toward a future course of action). For ex- ample, the presence of the phrases
‘I will’ and ‘I shall’ were hypothesized by Cohen et al. (2004) to be useful
bigram clues for Commissives. This base- line looks for future tense verb
phrases with a 1st person pronoun within one or two words preceding the verb phrase.
The Com baseline row of Table 4 shows the results for this heuristic, which
obtained 8% recall with 45% precision. The heuristic applied to only 9
sentences in our test set, 4 of which con- tained a Commissive speech act.
Directive speech acts are often questions, so we created a baseline system that
labels all sentences containing a question mark as a Directive. The Dir
baseline row of Table 4 shows that 97% of sentences
with
a question mark were indeed Directives.8 But only 73% of the Directive sentences
contained a question mark. The remaining 27% of Directives did not contain a
question mark and generally fell into two categories. Some sentences asked a
ques- tion but the writer ended the sentence with a period (e.g., “Has anyone
seen this before.”). And many di- rectives were expressed as requests rather
than ques- tions (e.g., “Let me know if anyone has a sugges- tion.”). For
Expressives, we implemented two baselines. Exp baseline 1 simply looks for an
exclamation mark, but this heuristic did not work well (18% re- call with 58%
precision) because exclamation marks were often used for general emphasis
(e.g., “The owner is frustrated with cleaning up urine!”). Exp baseline 2 looks
for the presence of four common expressive words (appreciate, hi, hello,
thank), in- cluding morphological variations of appreciate and thank. This
baseline produced very good results, 86% recall with 97% precision. Obviously a
small set of common expressions account for most of the Expressive speech acts
in our corpus. However, the word “hi” did produce some false hits because it
was used as a shorthand for “high”, usually when report- ing test results
(e.g., “hi calcium”).
8235
sentences contained a question mark, and 227 of them were Directives.
Finally,
as a baseline for the Representative class we simply looked for the words
diagnose(d) and sus- pect(ed). The Rep baseline row of Table 4 shows that this
heuristic was 100% accurate, but only pro- duced 5% recall (matching 3 of the
57 Representa- tive sentences in our test set).
CLASSIFIER
RESULTS
The
bottom portion of Table 4 shows the results for our classifiers. As we explained
in Section 3.3, we created one classifier for each speech act cate- gory, and
all four classifiers were applied to each sentence. So a sentence could receive
anywhere from 0-4 speech act labels indicating how many dif- ferent types of
speech acts appeared in the sentence. We trained and evaluated each classifier
using 10- fold cross-validation on our gold standard data set. The Unigram (U)
row shows the performance of classifiers that use only unigram features. For Di-
rectives, we see a 2% F-score improvement over the baseline, which reflects a
recall gain of 11% but a corresponding precision loss of 10%. The uni- grams
are clearly helpful in identifying many Direc- tive sentences that do not end
in a question mark, but at some cost to accuracy. For Expressives, the unigram
classifier achieves an F score of 92%, iden- tifying slightly more Expressive
sentences than the baseline with the same level of precision. For Com- missives
and Representatives, the unigram classi- fiers performed susbtantially better
than their corre- sponding baseline systems, but performance is still
relatively weak. Row A (U+ all lexsyn) in Table 4 shows the re- sults using
unigram features plus all of the lexical and syntactic features described in
Section 3.2.1. The lexical and syntactic features dramatically im- prove
performance on Commissives, increasing F score from 27% to 40%, and they
produce a 2% re- call gain for Representatives but with a correspond- ing loss
of precision. However, we observed that only a few of the lex- ical and
syntactic features had much impact on per- formance. We experimented with
different subsets of the features and obtained even better performance when
using just 10 of them, which we will refer to as the minimal lexsyn features.
The minimal lexsyn fea- ture set consists of the 4 Tense+Person features, the
Early Punctuation feature, the Sentence begins with
Modal/Verb
feature, and the 4 Sentence Position fea- tures. Row B shows the results using
unigram fea- tures plus only these minimal lexsyn features. Preci- sion
improves for Commissives by an additional 7% and Representatives by 2% when
using only these lexical and syntactic features. Consequently, we use the
minimal lexsyn features for the rest of our exper- iments. Row C shows the
results of adding the speech act clue words (see Section 3.2.2) to the feature
set used in Row B. The speech act clue words produced an additional recall gain
of 3% for Expressives and 2% for Representatives, although performance on Com-
missives dropped 2% in both recall and precision. Rows D and E show the results
of adding the se- mantic features. We added one semantic category at a time to
measure the impact of them separately. Row D adds two semantic features for the
TEST cat- egory, one from the Basilisk lexicon and one from the semantic
tagger. The TEST semantic features produced an F-score gain of 5% for
Commissives, improving recall by 4% and precision by 7%. Row E adds two
semantic features for the DRUG category. The DRUG features produced an
additional F-score gain of 2% for Commissives, improving recall by 4% with a
slight drop in precision.
4.4
Analysis
Together,
the TEST and DRUG semantic features dra- matically improved the classifier’s
ability to recog- nize Commissive speech acts, increasing its F score from 41%
→ 48%. This result demonstrates that in the domain of veterinary medicine, some
types of semantic entities are associated with speech acts. Our intuition
behind this result is that commitments are usually related to future actions.
In veterinary medicine, TESTS and DRUGS are associated with ac- tions performed
by doctors. Doctors help their pa- tients by prescribing or administering drugs
and by conducting tests. So these semantic entities may serve as a proxy to
implicitly represent actions that the doctor has done or may do. In future
work, ex- plicitly recognizing actions and events many be a worthwhile avenue
to further improve results. We achieved good success at identifying both Di-
rectives and Expressives, although simple heuristics also perform well on these
categories. We showed that training a Directive classifier can help to iden-
tify
Directive sentences that do not end with a ques- tion mark, although at the
cost of some precision. The Commissive speech act class benefitted the most from
the rich feature set. Unigrams are clearly not sufficient to identify Commissive
sentences. Many different types of clues seem to be important for recognizing these
sentences. The improvements obtained from adding semantic features also sug-
gests that domain-specific semantics can be useful for recognizing some speech
acts. However, there is still ample room for improvement, illustrating that
speech act classification is a challenging problem. Representative speech acts
were by far the most difficult to recognize. We believe that there are several
reasons for their low performance. First, Representatives were sparse in the
data set, occur- ring in only 2.91% of the sentences. Consequently, the
classifier had relatively few positive training instances. Second,
Representatives had the low- est inter-annotator agreement, indicating that
human judges had difficulty recognizing these speech acts too. The judges often
disagreed about whether a hypothesis or suspicion was the writer’s own belief
or whether it was stated as a fact reflecting general medical knowledge. The
message board text genre is especially challenging in this regard because the
writer is often presumed to be expressing his/her be- liefs even when the
writer does not explicitly say so. Finally, our semantic features could not
distinguish between diseases and symptoms. Access to a re- source that can
reliably identify disease terms could potentially improve performance in this
domain.
5
Conclusions
Our
goal was to identify speech act sentences in message board posts and to
classify the sentences with respect to four categories in Searle’s (1976)
speech act taxonomy. We achieved good results for speech act filtering and the
identification of Direc- tive and Expressive speech act sentences. We found that
Representative and Commissive speech acts are much more difficult to identify,
although the per- formance of our Commissive classifier substantially improved
with the addition of lexical, syntactic, and semantic features. Except for the
semantic class information, our feature set is domain-independent and could be
used to recognize speech act sentences
in
message boards for any domain. Furthermore, our features only rely on
part-of-speech tags and do not require parsing, which is of practical
importance for text genres such as message boards that are littered with
ungrammatical text, typos, and shorthand nota- tions. In future work, we
believe that segmenting sen- tences into clauses may help to train classifiers
more precisely. Ultimately, we would like to identify the speech act
expressions themselves because some sentences contain speech acts as well as
factual in- formation. Extracting the speech act expressions and clauses from
message boards and similar text genres could provide better tracking of
questions and answers in web forums and be used for sum- marization.
6
Acknowledgments
We
gratefully acknowledge that this research was supported in part by the National
Science Founda- tion under grant IIS-1018314. Any opinions, find- ings, and
conclusion or recommendations expressed in this material are those of the
authors and do not necessarily reflect the view of the U.S. government.
References
James Allen. 1995. Natural language understanding (2nd ed.).
Benjamin-Cummings Publishing Co., Inc., Redwood City, CA, USA.
Jean Carletta. 1996. Assessing agreement on classifi- cation tasks:
the kappa statistic. Comput. Linguist., 22:249–254,
June. Vitor R. Carvalho and William W. Cohen. 2005. On the
collective classification of email ”speech acts”. In SI- GIR ’05: Proceedings of
the 28th annualinternational ACM SIGIR conference on Research and development
in information retrieval, pages 345–352, New York, NY, USA. ACM Press.
Vitor R. Carvalho and William W. Cohen. 2006. Improv- ing ”email
speech acts” analysis via n-gram selection. In Proceedings of the HLT-NAACL
2006 Workshop on Analyzing Conversations in Text and Speech, ACTS ’09, pages
35–41, Stroudsburg, PA, USA. Association for Computational Linguistics.
William W. Cohen, Vitor R. Carvalho, and Tom M. Mitchell. 2004.
Learning to classify email into “speech acts”. In EMNLP, pages 309–316. ACL.
Jade Goldstein and Roberta Evans Sabin. 2006. Using speech acts to categorize
email and identify email gen-res. In Proceedings of the 39th Annual Hawaii
Inter- national Conference on System Sciences - Volume 03, pages 50.2–,
Washington, DC, USA. IEEE Computer Society.
Mark Hall, Eibe Frank, Geoffrey Holmes, Bernhard Pfahringer, Peter
Reutemann, and Ian H. Witten. 2009. The weka data mining software: an update.
SIGKDD Explor. Newsl., 11:10–18, November.
Ruihong Huang and Ellen
Riloff. 2010. Inducing domain-specific semantic class taggers from (almost)
nothing. In Proceedings of the 48th Annual Meeting of the Association for
ComputationalLinguistics, ACL ’10, pages 275–285, Stroudsburg, PA, USA.
Associa- tion for Computational Linguistics.
Minwoo Jeong, Chin-Yew Lin, and Gary Geunbae Lee. 2009.
Semi-supervised speech act recognition in emails and forums. In Proceedings of
the 2009 Con- ference on Empirical Methods in Natural Language Processing:
Volume 3 - Volume 3, EMNLP ’09, pages 1250–1259, Stroudsburg, PA, USA.
Association for Computational Linguistics.
Andrew Lampert, Robert Dale, and Cecile Paris. 2006. Classifying
speech acts using verbal response modes. In Proceedings of the 2006
Australasian Language Technology Workshop (ALTW2006), pages 34–41. Sydney
Australia : ALTA. Tara McIntosh. 2010. Unsupervised discovery of neg- ative
categories in lexicon bootstrapping. In Pro- ceedings of the 2010 Conference on
Empirical Meth- ods in Natural Language Processing, EMNLP ’10, pages 356–365,
Stroudsburg, PA, USA. Association for Computational Linguistics.
John Mildinhall and Jan Noyes. 2008. Toward a stochas- tic speech
act model of email behavior. In CEAS. Jacqueline Nastri, Jorge Pena, and
Jeffrey T. Hancock. 2006. The construction of away messages: A speech act
analysis. J. Computer-Mediated Communication, pages 1025–1045.
Sujith Ravi and Jihie Kim. 2007. Profiling student inter- actions in
threaded discussions with speech act classi- fiers. In Proceeding of the 2007
conference on Arti- ficial Intelligence in Education: Building Technology Rich
Learning Contexts That Work, pages 357–364, Amsterdam, The Netherlands, The
Netherlands. IOS Press.
Ellen Riloff, Janyce Wiebe, and Theresa Wilson. 2003. Learning
subjective nouns using extraction pattern bootstrapping. In Proceedings of the
seventh confer- ence on Natural language learning at HLT-NAACL 2003 - Volume 4,
CONLL ’03, pages 25–32, Strouds- burg, PA, USA. Association for Computational
Lin- guistics. John R. Searle. 1976. A classification of illocutionary acts.
Language in Society, 5(1):pp. 1–23.
Michael Thelen and Ellen Riloff. 2002. A bootstrapping method for
learning semantic lexicons using extrac- tion pattern contexts. In Proceedings
of the ACL-02 conference on Empirical methods in natural language processing -
Volume 10, EMNLP ’02, pages 214–221, Stroudsburg,PA, USA. Association for
Computational Linguistics.
Kristina Toutanova, Dan Klein, Christopher D. Manning, and Yoram
Singer. 2003. Feature-rich part-of-speech tagging with a cyclic dependency
network. In Pro- ceedings of the 2003 Conference of the North Ameri- can
Chapter of the Association for ComputationalLin- guistics on Human Language
Technology - Volume 1, NAACL ’03, pages 173–180, Stroudsburg, PA, USA.
Association for Computational Linguistics.
Douglas P. Twitchell and Jay F. Nunamaker Jr. 2004. Speech act
profiling: a probabilistic method for ana- lyzing persistent conversations and
their participants. In System Sciences, 2004. Proceedings of the 37th An- nual
Hawaii International Conference on, pages 1–10, January.
Douglas P. Twitchell, Mark Adkins, Jay F. Nunamaker Jr., and Judee
K. Burgoon. 2004. Using speech act theory to model conversations for automated
classi- fication and retrieval. In Proceedings of the Inter- national Working
Conference Language Action Per- spective CommunicationModelling(LAP 2004),pages
121–130.
A. Wierzbicka. 1987. English speech act verbs: a se- mantic
dictionary. Academic Press, Sydney, Orlando.







