This document summarizes the career journey of Won Ik Cho from his undergraduate studies in electrical engineering and mathematics to his current PhD research focusing on computational linguistics. It describes his early interests in source separation and automatic music composition that led him to study speech processing. It then outlines his transition to researching computational linguistics through developing speech recognition technologies for robot systems, including annotating a Korean utterance corpus to identify speech intentions. The document discusses Cho's struggles to publish his work and different approaches he has taken to categorize utterances and incorporate linguistic and acoustic features for his native Korean language.
This document provides an overview and preparation for the TOEFL iBT Speaking section. It discusses changes from the old TOEFL to the new TOEFL iBT, including the addition of speaking tasks. It outlines the 6 speaking tasks, including their timing, content, and goals. It also analyzes sample responses at different score levels and common pronunciation problems to address beforehand like vowels, consonants, and challenging word combinations.
This document provides an overview and examples of the TOEFL iBT Speaking section. It discusses the changes from the old TOEFL to the new TOEFL iBT, including the addition of speaking tasks. It describes the 6 speaking tasks which integrate skills like reading, listening and speaking. It also provides sample responses for each scoring level and discusses common pronunciation problems for English learners.
This document summarizes a workshop on teaching reading using a workshop model. It discusses the goals of implementing a reading workshop, including using a balanced approach with both overt instruction and situated practice. Key elements of the reading workshop model are explored, such as modeling, coaching, scaffolding, articulation, reflection and exploration. Structures to support reading development, such as read alouds, guided reading, conferring and strategy groups are also outlined.
The document discusses reading assessment. It notes that reading overlaps with and is similar to the other language skills of listening, speaking and writing. Assessments also overlap these skills. Reading performance is classified into selective, extensive, and interactive types, mirroring classifications for listening and speaking. The document outlines numerous assessment tasks used to evaluate reading and the variations that exist. It concludes that isolating reading from the other skills is difficult as assessments overlap skills.
Year 9 assignment.imaginative literary transformationjennifer_lawrence
The document outlines an assignment for Year 9 students that requires them to choose a global issue, collect related news articles with different perspectives, and write a poetry anthology expressing the topic and views. Students will present their anthology and discuss their process. The assignment aims to improve students' understanding of global issues and expressing information through poetry.
This document discusses a personalized statistical writing analysis service for research articles at the Japan Advanced Institute of Science and Technology. It provides an overview of the analysis process and focuses on five key aspects: vocabulary fit, readability, word type balance, style and usage, and lexicogrammatical errors. Feedback is tailored for individual authors based on comparisons to target corpora. Interviews with users found that explanatory notes, ratios, and some readability measures could be simplified. Future plans include integrating metrics into an online portal and comparing drafts to published versions.
Language Assessment - Assessing Reading Full Description with Picture and Cha...EFL Learning
There are four main elements of the reading process: the reader, the text, the reader's goal, and the result of the reading. Reading tasks can be categorized as perceptive, selective, interactive, or extensive based on whether they focus on bottom-up or top-down processing and form versus meaning. Common reading tasks include cloze, comprehension questions, ordering, and information transfer. Test design should consider the type of reading, length, focus, and processing involved.
1. The document discusses the development of reading from a socio-historical perspective, outlining five stages of language development and the evolution of writing from pictures to alphabetic systems.
2. It describes key skills and subskills involved in reading, including word recognition strategies, comprehension abilities, and study skills.
3. Theories of reading are examined, including viewing it as a set of divisible skills or as a holistic process, and the need for an integrated approach that teaches skills within the context of authentic reading is emphasized.
This document provides guidance for students taking an English language exam that consists of reading and writing sections. It outlines the timing and structure for answering questions, with 5 minutes allotted to read questions and passages, and around 10 minutes for each question. It describes different types of questions, such as "what" questions requiring evidence from the text and "how" questions involving language analysis. Trickier questions involving inference, comparison, or multiple focuses are also discussed. Strategies are presented for achieving high marks on whole text literature questions by linking responses to historical context. Requirements for different grade levels are defined in terms of textual understanding, language analysis, and contextual awareness.
The document provides examples of rubrics from various colleges and universities. It includes rubrics for writing, oral presentations, critical thinking, science, social science, fine arts, and other subject areas. The rubrics generally rate student work on a scale from 1 to 4 or 1 to 6, with higher scores indicating more proficient or accomplished work. The rubrics provide descriptors for what student work and skills would demonstrate at each rating level.
This document discusses how to assess and test reading. It begins by introducing the members of the reading group and includes an index of topics to be discussed. Some of the key topics covered include:
The importance of teaching reading, different types of reading like intensive and extensive reading, principles of teaching reading with examples of pre, during and post reading activities, recommendations for assessing reading with consideration for level and age, and different ways to test reading comprehension including cloze tests, ordering exercises, matching questions and multiple choice questions.
The document provides guidance on developing effective reading assessments and ensuring they are appropriate for the reading level and age of students. It offers examples of classroom activities and testing methods that can be used to evaluate reading skills
The document discusses different levels of excellence in analyzing a poem. Candidates who demonstrate low excellence allude to the main idea but provide explanations that lack perceptive analysis. Candidates who demonstrate high excellence set up and expand on the important ideas, identify specific techniques through examples, and tie it all together with perceptive comments that show original insight.
This syllabus outlines the semester plan for an English class at SMP N 2 Demak in semester 1. It includes 6 competence standards covering listening, speaking, reading and writing skills. For each standard there are basic competences, materials, activities, indicators and assessment details. The document provides an overview of what will be taught each week across the semester including topics, learning resources, time allocation and assessment methods.
This document outlines achievement levels and descriptors for oral communication, visual interpretation, reading comprehension, and writing based on a scale of 0 to 8. It provides descriptors for each criterion that define the skills demonstrated by students at different achievement levels. For example, a student scoring between 1-2 in oral communication will have difficulty understanding most information in familiar situations and making limited attempts to respond appropriately. A student scoring between 7-8 will show understanding of information in familiar situations and respond appropriately with clear pronunciation.
The document discusses various aspects of reading assessment, including different genres, skills, strategies, and types of reading. It describes two primary hurdles for becoming efficient readers as mastering bottom-up and top-down processing strategies. Several microskills and macroskills for reading are identified. Common strategies for reading comprehension include using prior knowledge, predicting, identifying main ideas, questioning, and making inferences. The types of reading discussed are perceptive, selective, interactive, and extensive reading.
This document provides an overview of a course on trends and research applications in natural language processing (NLP). It begins with introducing the goals of the course, which are to understand interesting NLP tasks and novel projects through a research-oriented webinar. The document then covers various NLP topics like question answering, machine translation, sentiment analysis, natural language generation applications, and challenges in NLP like grounded language and embodied language. It also provides tips for aspiring NLP researchers.
1) Listening is the most important of the four language skills and is used most often in everyday communication.
2) Traditional listening lessons involved short dialogues, dictation, and recognizing words and sounds, but now listening is recognized as its own skill to be developed.
3) Effective listening instruction includes teaching strategies like planning, monitoring comprehension, and clarifying meaning, and integrating both top-down and bottom-up processing.
This document summarizes key points about teaching speaking skills from the book "How To Teach Speaking" by Scott Thornbury. It discusses problems that second language learners face, similarities and differences between first and second language acquisition, communication strategies, and the knowledge needed for speaking. It concludes by outlining three theories - behaviorist, cognitivist, and socio-cultural - that are relevant to teaching speaking and their implications, such as modeling, practice, awareness raising, and teacher support within group activities.
Won Ik Cho presented on his research related to intention understanding in Korean natural language processing. He discussed developing annotation guidelines and corpora to classify Korean utterances by speech act, considering factors like intonation, context, and rhetoricalness. He proposed a method using text-based analysis combined with speech-aided disambiguation. Future work includes developing structured paraphrasing for argument extraction and an improved dialog manager.
Marriage of speech, vision and natural language processingYaman Kumar
Speech generally is considered to have three parts to it: vision, aural, and the social construct. In recent years, although the field has been moving at a dramatic pace, progress is being made in silos. The primary reason for this being that speech is considered "spoken text" by practitioners and researchers alike. Most open-source datasets due to their distance from real-world conditions help in spreading this false impression. In this condition, it is not surprising that common and important features of speech like intonation and disfluency do not get captured by this intent. This tutorial aims to provide an appreciation of the "full-stack" of speech - aural, vision and the textual (or social construct) parts with a special emphasis on aspects that may have significance for current and future research.
Week 1 class Introduction ntut pragmatics.pdfripipurba11
This document introduces an English pragmatics course. It outlines various subfields of linguistics including pragmatics, which is defined as the study of how context and situation affect meaning in actual conversations. The document provides examples to illustrate the difference between semantics and pragmatics. It also lists important concepts in pragmatics like context, implicature, speech acts, and politeness. Finally, it outlines activities for the first class, including a survey of pragmatics theses from the department and assigned reading.
Dyslexia: notes from the field is a presentation by Lisa Matthews about her experiences living with dyslexia. It is presented in a slide format but also as a durational artwork. The presentation includes 6 sections: a definition of dyslexia, how dyslexia affects Matthews day-to-day, describing dyslexia as not always a superpower, things not to say to someone with dyslexia, things that help those with dyslexia, and discussing both good and not-so-good design experiences. Matthews hopes sharing her experiences can help increase awareness of dyslexia on Global Accessibility Awareness Day.
The document discusses defining the LSP (Language for Specific Purposes) approach. It provides a central definition of LSP from Dudley Evans and St. John (1998) that focuses on designing courses to meet the specific needs of learners using the methodology and activities of the relevant discipline. It also discusses some relative characteristics that may or may not be present, such as courses being designed for specific disciplines. The document then examines how well this definition encompasses a new area within LSP called English for Socio-Cultural Purposes (ESCP), which focuses on learners in prison, those with disabilities, or immigrants seeking citizenship. A key motivation of the LSP approach is to help learners who are at a disadvantage due
welcome to the unit 6 Hortatory Exposition with the main topic "Means of Transportation"
The course consist of two activities that you will finish (online) in two weeks for every unit lesson, so that you will work on the materials one oral cycle for one week. another one written cycle also for one week
The first two weeks, you will learn and do activity about "ORAL CYCLE'.
The second two weeks, you will learn and do activity about "WRITTEN CYCLE"
So, if you have any questions please do not hesitate to sent me and email or post in the pin-board to discuss with other participants
Discussion Board Guidelines Students must respond individuallyLyndonPelletier761
Discussion Board Guidelines
Students must respond individually to the Discussion Board question and forge a discussion with at least two class members.
Discussion Board postings should be "original thought" posting. Unless requested to do so, students should not use the internet to create discussion board responses.
Discussion Board postings should be 300 words minimum in length, provide evidence of reading comprehension through illustrations from the reading assignments, critical thinking, and consideration of others' views.
When Discussion Board work is assigned on the syllabus schedule, students must post their responses by the appointed time and respond to at least two class members' posts within 48 hours/at the appointed time.
Discussion Board work is time-sensitive. Late submissions will not be accepted/will receive a zero.
It is STRONGLY suggested that you visit the discussion board frequently after your work is completed to see where the discussion is going and to further respond to your group members (or your instructor) if need be.
Please see the Discussion Board Rubric for grad
Assignment
Grading Rubric
Meets/Exceeds
Expectations
(10 Pts)
Approaches
Expectations
(8 Pts)
Needs Improvement
(6 Pts)
In-complete
and/or not
done correctly
(1 Pt)
No
Marks/Missing
(0 Pts)
Purpose
Purpose is clear.
Shows awareness of
purpose.
Shows limited
awareness of purpose.
No awareness.
No
Marks/Missing
Main idea Clearly presents a
main idea and
supports it throughout
the paper.
There is a main idea
supported
throughout most of
the paper.
Vague sense of a main
idea, weakly supported
throughout the paper.
No main idea.
Student didn't
turn in
assignment.
Overall and
Style
Well-planned and well-
thought out. Includes
title, introduction, and
statement of main
idea, transitions,
conclusion, using
APA standard: correct
font, font size and no
more than 1000
words.
Good overall
organization includes
the main
organizational tools
but font is incorrect
or word count is
more than 1000
and/or line spacing
is not doubled, or no
APA style.
There is a sense of
organization, although
some of the
organizational tools are
missing and/or late.
No sense of
organization.
Student didn't
turn in
assignment.
Content Exceptionally well-
presented and
discussed; ideas are
detailed, well-
developed with
evidence & facts
based on text
information.
Well-presented and
discussed; ideas are
detailed, developed
and supported with
less evidence and
details.
Content not as clear or
solid; some ideas are
present but not
particularly developed
or supported with
minimal evidence of
clear knowledge of
subject.
Content is not
complete.
Student didn't
turn in
assignment.
Structure Sentences are clear
and varied in pattern,
from simple to
complex, with
excellent use of
punctuation.
Sentences are clear
but may ...
The document provides information about Internal Assessments (IA) and External Assessments (EA) for the IB Theory of Knowledge course. An IA is a presentation by 1-5 students on a Problem of Knowledge taught to an audience. It is assessed based on identification of a knowledge issue, treatment of issues, knower's perspective, and connections. An EA is a 1600 word essay on a prescribed topic, graded externally and based on understanding issues, knower's perspective, analysis quality, and organization. The document outlines calendars and expectations for developing IA and EA projects over the course of a semester, including research methods, outlining, peer reviews, and presentations.
Research writing tips to prevent journal rejection: 5 Ground truths for clear...Nigel Daly
This talk discusses academic publishing from the perspective of non-native English speakers writing. Based on over 20 years of teaching, researching, publishing, reviewing and copy editing in a non-English speaking country, I share 5 tips to improve research writing quality. These tips are demonstrated with examples.
Discourse analysis examines how sentences form coherent units like paragraphs and conversations. Cohesion refers to grammatical and lexical relationships between discourse elements, created through devices like reference, substitution, ellipsis, and conjunction. However, coherence depends on how people interpret a discourse based on their background knowledge and expectations. While cohesive devices link utterances, coherence involves constructing a meaningful interpretation in line with one's experience of the world.
This document describes a course on Natural Language Processing. The course is divided into 6 units covering basics of NLP, language modeling techniques, syntactic parsing, semantic parsing, information extraction, and machine translation. Students are required to have a basic knowledge of English grammar, theoretical computer science foundations, and Python. The course aims to help students differentiate various NLP tasks, model language, perform syntactic and semantic parsing, and apply NLP techniques like information extraction and machine translation. Students will be evaluated through tests, assignments, and class participation.
This document provides information about an optional webinar for students in LIST 4373 on literacy instruction. The webinar will take place on Monday, March 7 from 6:00-7:15 PM or Tuesday, March 8 from 2:00-3:15 PM, or students can watch a recorded session. The link and login information for the live webinar is provided. Technical support contact information is also included if students have trouble logging in.
The documents discuss exams and language tasks. The first document lists reasons why learners hate exams, such as exams trying to catch learners not knowing something and being unpredictable. The second document discusses characteristics of good exams, including interesting preparation, enjoyable tasks reflecting real-world needs, and educating teachers. It asks what tasks represent these characteristics and how performance can be marked. The following documents discuss language tasks, such as note-making, writing memos, and translation. They provide assessment criteria for speaking skills based on the Common European Framework, including interaction fluency, coherence, and range/accuracy. Sample speaking tasks are provided to assess these skills.
Evo research topics to r qs (judith hanks), january 2016 (1)ClassResearchEVO
Dr. Judith Hanks discusses ways to narrow the focus of research from broad topics to specific research questions. She outlines two approaches: starting with a topic and title then developing questions and data collection, or beginning with data collection and allowing questions to emerge during analysis. Refining questions involves considering agency, feasibility, and what researchers truly want to understand. Classrooms provide rich research contexts if questions are carefully focused. Developing good questions involves curiosity, imagination, and flexibility to changing understandings.
Natural Language Processing: L01 introductionananth
This presentation introduces the course Natural Language Processing (NLP) by enumerating a number of applications, course positioning, challenges presented by Natural Language text and emerging approaches to topics like word representation.
This document summarizes the development of Korean natural language processing (NLP) and corpora studies. It discusses how Korean NLP has followed classical computational linguistics approaches but faces challenges from Korean's complex syntax and morphology and writing system. It then outlines some of the major past and ongoing projects in Korean NLP including corpora, treebanks, and datasets created by organizations like KAIST, NIKL, ETRI, and NIA. It notes trends in Korean corpus studies from large annotated corpora and treebanks in earlier decades to current work on downstream tasks, pretraining, and benchmarks for large language models.
This document proposes a framework called PaperCard for transparently documenting machine assistance in academic writing. With advances in generative AI, machines can now play a larger role in the writing process. However, this raises legal and ethical issues regarding fairness, accountability, and transparency that have yet to be addressed. PaperCard would require authors to declare any machine assistance, including details about the specific AI system, prompts provided, and edits made to generated content. This framework aims to govern the use of machines in academic writing by promoting transparency around their involvement.
This document presents a method for cross-modal knowledge distillation from text to speech models using dropout-based teacher confidence. It aims to leverage large pre-trained language models trained on text for speech tasks by distilling knowledge from a text-based BERT model to an end-to-end speech understanding model. The method uses dropout perturbations on the teacher's logits to calculate a confidence score for each sample, which is then used to weight the distillation loss and guide the training of the student speech model, especially when combined with scheduling strategies like decaying and triangular schedules. The method is evaluated on a public spoken language understanding dataset and shows improvements over baseline distillation, demonstrating the effectiveness of using teacher confidence to manage knowledge transfer across modal
Assessing How Users Display Self-Disclosure and Authenticity in Conversation with Human-Like Agents: A Case Study of Luda Lee (presented at AACL-IJCNLP 2022)
The document summarizes the construction and analysis of a Korean hate speech corpus. It discusses how hate speech was defined and annotated, including the guidelines developed for identifying social bias and measuring toxicity. Over 10,000 online comments were annotated. Analysis found that toxicity is usually present with biased comments, and gender-related bias tended to coincide with more toxic expressions than other biases. The corpus was created to help analyze real-world hate speech detection in Korean and address gaps in previous work.
This document summarizes a research paper presented at DIS 2021 about developing an assistive word segmentation tool for digital minorities typing in Korean. It describes how existing tools require word segmentation that can be difficult. The presented system uses a data-driven approach with a corpus of conversational text and character-level neural models to provide segmentation suggestions without modifying the original text. Evaluation shows it achieves natural segmentation and is robust to errors. A web interface is proposed to provide segmentation options to support digital minorities and alleviate annoyance for other users when typing user-generated text. Future work includes demonstration of the system and applications.
This document summarizes a presentation on measuring and addressing gender bias in machine translation between different language pairs. It discusses how gender bias can occur when translating between languages with gender-neutral pronouns to gendered languages, or languages without grammatical gender to those with grammatical gender. The presentation proposes a template-based approach to simultaneously evaluate translation gender bias across multiple language pairs, and measures fluency and biasedness in the translations. Results suggest that while more resources can improve fluency, it does not guarantee reducing gender bias in translations, and social stereotypes remain a major cause of biased translations.
This document summarizes an academic presentation on measuring and addressing gender bias in machine translation. It discusses two main types of translation gender bias related to gender-neutral pronouns and grammatical gender. The authors propose a combined cross-lingual evaluation approach using template translations between language pairs that vary in their use of gender-neutral pronouns and grammatical constructs. They measure translation quality and gender bias across the language pairs using automatic metrics and human evaluations. The results suggest translation gender bias is more common when translating to languages with grammatical gender constructs and that increasing training data may not fully address social stereotypes that can amplify bias.
This document summarizes an effort to curate open Korean natural language datasets for global users. It provides an overview of 32 open Korean datasets across various criteria like documentation status, license for use and distribution. The curation is intended to be updated and maintained on arXiv and GitHub as a living document. It acknowledges original dataset creators and the Ko-NLP project for hosting the work. The overview aims to address the need for more openly available Korean datasets and resources to support non-Korean NLP researchers.
1) The document describes a method for cross-modal knowledge distillation from pretrained language models to improve end-to-end spoken language understanding systems.
2) A pretrained BERT model is fine-tuned on text transcripts then used as a teacher to distill knowledge into a student end-to-end speech SLU system using mean absolute error loss.
3) Experimental results found this simple distillation approach helped the student learn uncertainty from the teacher model and improved performance over a speech-only baseline, demonstrating cross-modal knowledge sharing is effective for spoken language tasks.
2010 PACLIC - pay attention to categoriesWarNik Chow
This document summarizes a research paper on a proposed method called Metadata Projection Matrix (MPM) for sentence modeling that allows controlling attention to certain syntactic categories. The method uses a projection matrix to incorporate syntactic category information when calculating attention weights. Experimental results on several datasets show MPM outperforms baselines on tasks where attention to specific categories is important, like detecting terms or irony, but is weaker on more context-dependent tasks. The method is best suited to applications where syntactic structure significantly informs predictions.
The membership Module in the Odoo 17 ERPCeline George
Some business organizations give membership to their customers to ensure the long term relationship with those customers. If the customer is a member of the business then they get special offers and other benefits. The membership module in odoo 17 is helpful to manage everything related to the membership of multiple customers.
Principles of Roods Approach!!!!!!!.pptxibtesaam huma
Principles of Rood’s Approach
Treatment technique used in physiotherapy for neurological patients which aids them to recover and improve quality of life
Facilitatory techniques
Inhibitory techniques
How to Configure Time Off Types in Odoo 17Celine George
Now we can take look into how to configure time off types in odoo 17 through this slide. Time-off types are used to grant or request different types of leave. Only then the authorities will have a clear view or a clear understanding of what kind of leave the employee is taking.
How to Store Data on the Odoo 17 WebsiteCeline George
Here we are going to discuss how to store data in Odoo 17 Website.
It includes defining a model with few fields in it. Add demo data into the model using data directory. Also using a controller, pass the values into the template while rendering it and display the values in the website.
Credit limit improvement system in odoo 17Celine George
In Odoo 17, confirmed and uninvoiced sales orders are now factored into a partner's total receivables. As a result, the credit limit warning system now considers this updated calculation, leading to more accurate and effective credit management.
No, it's not a robot: prompt writing for investigative journalismPaul Bradshaw
How to use generative AI tools like ChatGPT and Gemini to generate story ideas for investigations, identify potential sources, and help with coding and writing.
A talk from the Centre for Investigative Journalism Summer School, July 2024
Join educators from the US and worldwide at this year’s conference, themed “Strategies for Proficiency & Acquisition,” to learn from top experts in world language teaching.
Beginner's Guide to Bypassing Falco Container Runtime Security in Kubernetes ...anjaliinfosec
This presentation, crafted for the Kubernetes Village at BSides Bangalore 2024, delves into the essentials of bypassing Falco, a leading container runtime security solution in Kubernetes. Tailored for beginners, it covers fundamental concepts, practical techniques, and real-world examples to help you understand and navigate Falco's security mechanisms effectively. Ideal for developers, security professionals, and tech enthusiasts eager to enhance their expertise in Kubernetes security and container runtime defenses.
How to Show Sample Data in Tree and Kanban View in Odoo 17Celine George
In Odoo 17, sample data serves as a valuable resource for users seeking to familiarize themselves with the functionalities and capabilities of the software prior to integrating their own information. In this slide we are going to discuss about how to show sample data to a tree view and a kanban view.
Split Shifts From Gantt View in the Odoo 17Celine George
Odoo allows users to split long shifts into multiple segments directly from the Gantt view.Each segment retains details of the original shift, such as employee assignment, start time, end time, and specific tasks or descriptions.
Tales of Two States: A Comparative Study of Language and Literature in Kerala...
1908 working memory
1. Human Interface Laboratory
How a Young Speech Researcher
Dived into Computational Linguistics and
Where He Heads Now
2019. 8. 26, @Working Memory
Won Ik Cho
2. About Me
• 조원익
B.S. in EE/Mathematics (SNU, ’10~’14)
Ph.D. student (SNU INMC, ‘14~)
• Academic background
Interested in mathematics > EE!
Double major?
• ...
Early years in Speech processing lab
• Source separation
• Voice activity & endpoint detection
• Automatic music composition
Currently studying on computational linguistics
1
5. Early years
• Why started mathematics?
Long dreamed romance
4
From 2008!
but...
6. Early years
• Undergraduate design project: Music source separation
Why? – To automatically extract and transcribe the score of polyphonic
music for guitar orchestration
• And result?
5
Image: http://kimgooni.blog.me/221427899920
7. Early years
• Could not give up mathematics... (although it dismissed me)
First, aimed at cryptography laboratory
6
Image top: https://blog.goodaudience.com/cryptography-for-dummies-part-1-a811d4852daa
bottom: https://www.coindesk.com/bitcoin-hits-new-2019-high-above-8900
or...?
10. Early years
• Signal processing domain revisited
9
EE!
Circuit
Power
Semiconductor
ControlBio
Communication
?
?
System
11. Early years
• One step toward grand dream of polyphony music transcription?
Paper survey on ...
• Multiple pitch estimation
• Music grammar
• Automatic music composition
10
Image: https://www.youtube.com/watch?v=TwQybAwL7NY
12. Early years
• And reality...
11
Image: https://www.lumenvox.com/resources/caseStudies/redmond/redmond-software-3.aspx
13. Early years
• First paper technical report on rule-based 4-part chorus
composition!
Done while attending to a lecture on music theory (harmony)
EXTREMELY HEURISTIC AND NOT NATURAL!
12
14. Early years
• Started a new government-funded project from April 2017
Development of free-running speech
recognition technologies for
embedded robot system
13
Image: https://www.musicaslanguage.com/, https://imgur.com/gallery/iWKad22
15. Dive into computational linguistics
• New task?
Development of free-running speech recognition technologies for
embedded robot system
로봇용 free-running 임베디드 자연어 대화음성인식을 위한 원천 기술 개발
• In other words:
Non wake-up-word based speech understanding system
...?
14
오늘 또
떨어졌네
이게 대체
며칠째
파란불이냐
지금 손실이
얼마지
16. Dive into computational linguistics
• How?
Related to many aspects of (speaker-dependent) speech recognition
• Speaker-dependency (in terms of a personal assistant)
• Noisy far-talk recognition and beamforming
• Speech intention understanding
– To which utterances should AI react?
15
오늘 또
떨어졌네
이게 대체
며칠째
파란불이냐
지금 손실이
얼마지
17. Dive into computational linguistics
• It’s about finding an internal intention of a human speech
And in Korean?
16
Image: Top https://onepageinfo.tistory.com/52
Bottom: https://m.blog.naver.com/barim12/220831241685
데이터도 없는데 어떻게 해요!
일단 만들어!
18. Dive into computational linguistics
• Need to find what `questions’ and `commands’ are
• Sentence type?
밥 먹었다
밥 먹었니
밥 먹어라
• But...
너랑 오랜만에 밥 먹고 싶네
부른지가 언젠데 밥 안 먹냐
밥 좀 작작 먹어라
• ...?
17
Image: Top https://www.pinterest.co.kr/pin/367887863281568581/?lp=true
Bottom https://www.hkn24.com/news/articleView.html?idxno=69187
19. Dive into computational linguistics
• WHY??
The utterance intention should be identified among colloquial
conversations (non-wake-up-word-based!)
In pragmatics and sentence-level semantics, this is called `speech act’
(Searle, 1976)
• And in some cases, `dialog act’ (Stolcke et al., 2000)
• It is best to use dialog history and prosodic information, if possible
But only text data (about 800K, unlabeled single utterances) for us
• and need to show the potential for the left years!
Let’s first choose 20K randomly and make a labeled corpus on:
18
20. Dive into computational linguistics
• Intonation-dependent utterances
How to figure out if the utterances is intonation-dependent?
19
천천히 가고 있어! (utterance)
천천 히 가 고 있 어 (transcript)
question
statement
command
?
21. Dive into computational linguistics
• Intonation-dependent utterances
Underspecified sentence enders
• -어, -지, -대, -해, -라고, -다며, etc. (differs from –다, -니, -라)
• Sentence type is determined based upon the sentence-final intonations that are
assigned considering the speech act
Conversation maxim (Levinson, 2000)
• 정보성-원리 Informativeness-principle (단순화 버전)
– 화자: 필요한 것 이상으로 말하지 말라.
» Do not say more than is required (bearing the Q-principle in mind)
– 청자: 화자가 일반적으로 말한 것은 전형적으로 그리고 특칭적으로 해석하라.
» What is generally said is stereotypically and specifically exemplified.
– e.g., 내일 학생회관에서 두시 반에 만나서 얘기해 (질문? x)
20
22. Dive into computational linguistics
• Introducing phonetic features: Intonation-dependency
Annotating proper intention for possible cases of intonation
• 기본적으로 문말 억양을 고려함
• 한 가지 intonation에서 여러 intention이 가능한 경우는 ambiguous한 것으로 봄
• 부사, 수일치 등과 관련하여, 서술이 아닌 것으로 해석하기 어색한 것들은 제외함.
• 너무 많은 정보를 담고 있는 문장을 질문으로 판단하는 것을 피함
• Wh-particle들이 의문사의 기능을 하지 않는 경우들에 주의함
• 많은 한국어 문장이 그렇듯, 주어가 생략되어 1,2,3-인칭 등으로 해석할 수 있을 경
우에는, 각각을 대입해 보고, 어색하지 않은 것들로 판단함
• 호격의 유무에 주의함
21
23. Dive into computational linguistics
• Intention understanding – how?
Our approach (for Korean)
22
단일 문장인가?
Intonation 정보로
결정 가능한가?
Question set이 있고
청자의 답을 필요로 하는가?
Effective한 To-do list가
청자에게 부여되는가?
No
Yes
No
Yes
요구 (Commands)
수사명령문 (RC)
Full clause를
포함하는가?
No
No
Compound sentence: 힘이 강한 화행에 중점
(서로 다른 문장도 같은 토픽일 때 한 문장으로 간주)
Fragments (FR)
질문 (Questions)
No
Context-dependent (CD)
Yes
Yes
Yes
Intonation 정보가
필요한가?
Yes
Intonation-dependent (ID)
No Questions /
Embedded form
Requirements /
Prohibitions
수사의문문 (RQ)
Target: single sentence
without context
nor punctuation
Otherwise
서술 (Statements)
24. Dive into computational linguistics
• System overview: Text-based sieving + Speech-aided analysis
Compatible to text-speech alignment
23
25. Dive into computational linguistics
• 2017.04~2017.11
Setup phase
• Study syntax, semantics, and pragmatics for making up an annotation
guideline for colloquial utterances
• 2017.12~2018.08
Corpus annotation
• Along with undergraduate students (design project)
• 2018.09~2018.11
Implement classifiers and make a documentation (paper on arXiv)
• Repository for an open sourcing
• 2018.12~
Modifying & maintaining the repository
• https://github.com/warnikchow/3i4k
Article (in Korean) on the annotation guideline (DisCog 26:3)
24
26. The Struggle within
• Project is successful (so far) - But does it bring me a publication?
Approach #1: First, let’s make a similar guideline for English! (2017.11)
• Manual tagging on Cornell Movie dialog dataset (binary: only obligatory/non-
obligatory)
25
and REJECT! (for a signal processing conference)
... My sense here is that the two positive reviewers come from a linguistics
background, and are happy to see linguistic insights being applied to a new real-
world problem. The two negative reviewers seem more aware of the methods
currently in use for this type of problem. The field today is hugely dominated by
data-driven methods with very few linguistics insights, so I think it's important to
make room for papers like this. But at the same time, I think this is a bridge too
far for people working in this area to really engage with. There were lots of missed
opportunities – in addition to the reviewers' comments, this paper would have
been saved by experiments on a common dataset, like DSTC2 or ATIS. I therefore
hesitantly recommend reject.
27. The Struggle within
• How about finer categorization and bigger dataset?
Approach #2: `question’, `command’, and `statement’
26
and REJECT! (for a computational linguistics conference)
... One weakness or point of criticism is that it did not become clear to me whether
the annotated corpus is being made available open source as a corpus for further
study ... I am not entirely convinced that it is the best idea to use abbreviations /
names for features that are so similar to established "academic" terminology for
sentence types. While the intention is obvious to point out the relationship, it might
be a good idea to make the difference more explicit in the names (int, imp, dec) ...
We learn very little about the annotation guidelines, their granularity, publication
status etc. ... We learn little about the prospective application to spoken language
corpora and the expected impact of an application to spoken / phonetic data incl.
phonological features.
28. The Struggle within
• And the issues solved?
Approach #3: More on justification
27
and REJECT! (for an AI/ML conference)
... I agree with the main motivation of categorize the utterance in a conversation
based on the expected response of that one who receives the information. This
position could make procedures more dynamic and direct. However, the authors do
not make an argument that the proposed categories are sufficient for a dialogue,
they might be the main but I will suggest to pay attention to clarifications and
continuations which not necessary correspond to an answer or action. ... It is not
clear what shows the results of the classification regarding the proposal of category.
A good classification means that the proposal categories are good? I missing a
guide on how to interpret this relation. ...
29. The Struggle within
• Should I do it for my own language?
Approach #4: Similar categorization in Korean, incorporating a new label
regarding acoustic cues for a head-final language
28
and REJECT! (for an NLP conference)
Summary: The paper presents an approach for intention recognition in Korean,
leveraging both text and acoustic information.
Strengths: The approach is relevant for SLU or dialog systems, and addresses issues
of recognition for head-final languages. It also considers acoustic information
specifically in those cases where acoustic cues are the best discriminators.
Weaknesses: The paper would benefit from specific comparison between the
proposed, somewhat complex architecture and other possible alternative models to
justify the system. It was unclear whether the approach was limited to head-final
languages, and Korean in particular, which would yield a narrow result, and if so,why.
31. The Struggle within
• And some unexpected invitations from linguistic venues/journals
30
32. And to speech again
• Conference to be presented: ICPhS
International Congress of Phonetic Science
• 8/5-9, Melbourne, Australia
31
(although I couldn’t attend...)
33. And to speech again
• Prosody-ambiguous statements?
Problems in: Wh-intervention?
• Needs disambiguation
32
몇 개 가져오래
Should I bring some?
How many should I bring?
They told you to bring some?
34. And to speech again
• Prosody-ambiguous statements? How about constructing a corpus that contains ONLY the
utterances whose syntactic ambiguity can be resolved by introducing prosody?
# Wh- particles
누구 (nwukwu, who), 뭐 (mwe, what),
어디 (eti, where), 언제 (encey, when),
어떻게 (ettehkey, how), 몇 (myech, how much)
왜 (way, why) was not utilized because it is not used as an existential quantifier
# Predicates
Depend on the wh- particle being adopted
Chosen among 5,800 frequently used lexicons
Pronouns and polarity items were added in some cases
# Reportive particles
Added to form an evidential mood
Induces rhetoricalness for some questions
# Sentence enders (SEs)
SEs with an unfixed role (underspecified SE)
e.g. -래 (ray), -어 (e), -지 (ci)
SEs with a fixed role
e.g. -까 (kka: interrogative)
# Politeness suffix
Attached at the end of a sentence to assign politeness
Restricts rhetoricalness under some circumstances
33
35. And to speech again
• Prosody-ambiguous statements?
Created 1,292 sentences
Constructed 3,552 utterances (with speech intention, under consensus)
34
36. And to speech again
• Prosody-ambiguous statements?
Recorded 7,104 utterances (female/male)
35
37. And to speech again
• Prosody-ambiguous statements?
36
AdvisorPresenter
Editing
Wrote paper
Checked dataset & Recording
Co-author / Equal contribution
Checked dataset
Proofreading
38. And to speech again
• Prosody-ambiguous statements?
37
39. And to speech again
• Prosody-ambiguous statements?
And in future?
• Disambiguation inspired by neuro-scientific phenomenon
38
41. Interdisciplinary
• AI ethics? (ACL workshop topic)
Measuring gender bias in machine translation
Originally claimed to deal with the proposal of government project ... but
40
42. Done and afterward
• Done
억양 의존성 및 rhetoricalness를 고려한, 음성인식 output 분석에 적합한 일
반언어학적 speech act 분류 방법론 제시
한국어를 위한 annotation guideline 정립, corpus 구성 및 모델 학습
한국어 Speech intention의 disambiguation을 위한 corpus 구성
질문/요구 paraphrasing 위한 parallel corpus 제작 (under progress)
• Afterward?
Speech disambiguation을 위한 co-attention framework 개량
대화체/비정형 질문/요구의 structured paraphrasing
Task-oriented와 non-oriented 간 code switching이 자유로운 dialog
manager 시스템의 개발
41
43. Done and afterward
• Where do I head now?
42
Image: Top https://phdcomics.com/comics/archive_print.php?comicid=1733
Bottom: https://slideplayer.com/slide/15366786/
44. Reference
• Searle, John R. A classification of illocutionary acts. Language in society 5.1 (1976): 1-23.
• Stolcke, Andreas, et al. Dialogue act modeling for automatic tagging and recognition of
conversational speech. Computational linguistics 26.3 (2000): 339-373.
• Levinson, Stephen C. Presumptive meanings: The theory of generalized conversational
implicature. MIT press (2000).
• Cho, Won Ik, Hyeon Seung Lee, Ji Won Yoon, Seok Min Kim, and Nam Soo Kim. Speech
intention understanding in a head-final language: A disambiguation utilizing intonation-
dependency. arXiv preprint arXiv:1811.04231 (2018).
• Cho, Won Ik, and Nam Soo Kim. Discourse component-based Korean speech act categorization
to resolve the vagueness in understanding text intention: A computational linguistics
perspective. Discourse and Cognition 26.3 (2019): 227-247. [Korean]
• Cho, Won Ik*, Jeonghwa Cho*, Jeemin Kang, and Nam Soo Kim. Prosody-semantics interface in
Seoul Korean: Corpus for a disambiguation of wh- intervention. Proc. ICPhS (2019): 3902-3906.
43