127 Matching Annotations
  1. May 2024
    1. Create an account using the sidebar on the right of the screen. Pin the Hypothesis extension in Chrome (1 and 2), then activate the sidebar by clicking the button in the location bar (3).

      какие-то шаги от создателей Hypothesis

  2. Jan 2024
    1. A task is used further down in the workflow. When I think about planning features, they occur before development takes off. The task is used by the developer as they are breaking down the issue into smaller components.

      task context of creating "task": developer as they break down a larger issue/epic into smaller pieces

    2. You can see how the constant jumping between these two tools in the first scenario is super annoying, and also very risky as none of the changes you make in Figma are also automatically being updated in the same GitLab designs.
    3. As a positive example of where this works well: Our VS Code GitLab Workflow extension allows users to not only see comments that were written inside the GitLab UI, but also allows these users to respond to these comments right from the IDE, the tool where they actually have to make these changes.
  3. May 2023
    1. Practice (Dash): Undone Action Item — Individual items (action items and ideas) are marked with a dash preceding them. All items, no matter what they are, are therefore treated as items to be processed. (Plus): Done Action Item — If the item is an action item (todo), when the item is complete, a vertical line is drawn through the “dash” thus making it resemble a “plus”. This makes the dashed items stand out quite well despite the fact that the same color pen or pencil may be used. (Right Arrow): Waiting – (i.e. for another action) — Drawing an arrow pointing to the the item denotes that it is something that is waiting on another action to happen or deliverable. (Left Arrow): Delegated — Drawing an arrow pointing to the left of the item denotes that it has been delegated (with a note to whom and the date) . (Triangle): Data Point — Turning the dash into a triangle denotes a data point (a fact or figure you wish to remember for instance). (Circle) — A circle around any of the above means that it has been carried forward, moved to another list or otherwise changed status — i.e. a “Waiting” item has now become an Action Item elsewhere (with a note about where that item has gone). The beauty of this system is that it is all built upon, and extensions of, the original dash. Therefore, it is easy to change items from one state to another (an undone action item to a done one, an undone action item to waiting or delegated) and in the case of an non-dashed item changing completely the item is circled to denote that.
    1. Throughout the day, mark each task as completed, in-progress, or delegated. Feel free to create your own symbols.

      Similar to the sorts of to do list task key in many bullet journals, the Analog system has "task signals" : - black filled circle means "complete task" - half filled circle means task is in progress - a right arrow in the circle means the task was delegated - a cross in the circle means that the task is an appointment, potentially with the appointment time added to the to do item

      The system suggests that you can "create your own" task signals, though in true minimalist fashion, it doesn't give other suggestions. Presumably one could do other pattern fills of the circle or symbols within it to mean other things (example: bullet journal key symbols).

      Interestingly, the to do circles start out not blank, but with a single thin line splitting the circle in half vertically. This is apparently a design choice, perhaps to make it easier to fill in half of the circle?

  4. Dec 2022
    1. 如果你已经在文件列表中创建了文件夹的话,你可以右键点击文件夹,选择 新建笔记 选项来创建笔记。此时,新建笔记会存放在该文件夹下。当然,你也可以使用文件列表顶部的按钮来创建笔记或者文件夹。当你停止输入两秒,或关闭面板后,Obsidian 都会自动为你进行保存。你也可以通过快捷键 Ctrl/Cmd+S 来手动保存。相关条目 重命名笔记 Links to this page索引由此开始

      2 #task

    1. So she was considering in her own mind (as well as she could, for the hot day made her feel very sleepy and stupid), whether the pleasure of making a daisy-chain would be worth the trouble of getting up and picking the daisies, when suddenly a White Rabbit with pink eyes ran close by her.


    2. Alice was beginning to get very tired of sitting by her sister on the bank, and of having nothing to do: once or twice she had peeped into the book her sister was reading, but it had no pictures or conversations in it, “and what is the use of a book,” thought Alice “without pictures or conversations?”


  5. Nov 2022
    1. e argue that mutual learningwould benefit sentiment classification since it enriches theinformation required for the training of the sentiment clas-sifier (e.g., when the word “incredible” is used to describe“acting” or “movie”, the polarity should be positive)

      By training a topic model that has "similar" weights to the word vector model the sentiment task can also be improved (as per the example "incredible" should be positive when used to describe "acting" or "movie" in this context

    2. . However, such a framework is not applicablehere since the learned latent topic representations in topicmodels can not be shared directly with word or sentencerepresentations learned in classifiers, due to their differentinherent meanings

      Latent word vectors and topic models learn different and entirely unrelated representations

    1. Obsidian Kanban for Someday Maybe

      Examples include a separate Kanban board for - camping tasks - home-based tasks - Obsidian management - such as organize old files in database

  6. Oct 2022
    1. It was Ranke, too, who set the historian’s task: to find out wie eseigentlich gewesen ist, what really happened, or, literally, how itreally was. His goal is one that will remain forever just beyond ourgrasp for reasons I explained in a “Note on Sources” in The Guns ofAugust (a paragraph that no one ever reads but I think is the bestthing in the book). Summarized, the reasons are that we who writeabout the past were not there. We can never be certain that we haverecaptured it as it really was. But the least we can do is to stay withinthe evidence.
  7. Aug 2022
    1. We further expand on this and try to identify communities amongst a plethora of tweets using clustering techniques.

      Task for self: try to re-create

  8. May 2022
    1. Abstract

      Your task:

      What is the topic and the main idea of this article, based on the abstract? TAG the sentences with "topic" and "main idea" - make your annotations public, please.

  9. Mar 2022
  10. Dec 2021
    1. In her book Meaning in Life and Why it Matters (2010), the US philosopher Susan Wolf has argued, against Taylor, that even if Sisyphus is feeling fulfilled, owing to the nature of the activity in which he is engaged, his feeling fulfilled is not reasonable. She argues that the object of an agent’s feeling of fulfilment is represented in the mind of the agent as being something that is objectively good. In the case of Sisyphus, even if he is fulfilled, he misrepresents the source of his fulfilment as good. So Wolf argues that the meaningfulness of an activity or goal is a function not only of one’s being fulfilled in engaging in it/pursuing it, but the feeling of fulfilment also needs to be a fitting response to the activity or goal. And it is only a fitting response if the activity or goal is objectively pursuit-worthy. What would render an activity or goal objectively pursuit-worthy? Objective pursuit-worthiness is in part owing to the value of the end or activity coming from outside of oneself. Wolf admits that determining objective pursuit-worthiness is not easy, and involves a process of discovery. A possibility not explicitly endorsed by Wolf (but not rejected by her, either) is that an activity or goal is objectively pursuit-worthy if engaging in that activity is conducive to the cultivation of virtues. My suggestion is that, so long as an activity is generally conducive to developing character traits and intellectual traits that will enhance and not impede the flourishing of both an agent and those with whom they interact, it is objectively pursuit-worthy.

      Intellectual and character growth holistically, in a positive manner. That is the objective worthiness of any task. These goals, must be outlined in any activity undertaken, and their worth must be scaled according to our own standards. Because time is limited, it serves us well to maximise the objective worthiness that we extract from any task/activity/job/career. Thus time frames must also be defined. A multifaceted growth mindset requires a constant reexamination of all these aspect- i.e.- objective worthiness and time frame. An unexamined life is not worth living, and this shall be applied ona daily basis. One must recognise where they lose time, fill those holes, and proceed to extract some meaning from it.

  11. Nov 2021
  12. Sep 2021
    1. A potentially interesting task management plugin for obsidian. I'm a little worried about long-term support. I'm going to wait and see what happens.

  13. Aug 2021
    1. Task 2 Briefly not the main topics throughout each section of the text. This section explains the differences in philosophy from other types of science. It shows how they collect knowledge and that there is not a definite answer to philosophical questions. Restate the main point of th opening paragraph in your own words. The opening paragraphs main point is that philosophy is the study of value. Many time it is undermined by other types of science that claim it is pointless. These people do not allow any kind of growth outside of said science. Estate the main point of the closing paragraph in your own words Philosophy’s goal is not to come to a conclusion but instead expand you thoughts. This is the knowledge that makes it possible to grow how we think. Everyone find and define two vocabulary words Unalloyed-complete and unreserved Dogmatism-the tendency to lay down principles as incontrovertibly true.

    2. Task 1. What does the title tell you about the text? The author of this text is going to explain why and how philosophy is so valuable to people. What do the section titles tell you about the text? The section titles show how the author felt they needed to break up his statement to prove their point. What do the footnotes tell you about the text? The footnotes could possibly contain examples or references that has been cited.

    3. The Value of Philosophy

      Task 1 WHAT DOES THE TITLE TELL YOU ABOUT THE TEXT? The title tells me that the reading will be probably about what philosophy has to value and what it is.

  14. Jul 2021
    1. How Task management software is useful for small as well as big organizations? Task management software is used to handle tasks performed by employees.. Essentially, this means that processes must be smooth and productivity must be improved and useful for any sized organizations.

      Visit: https://anuchandola009.blogspot.com/2021/07/how-task-management-software-is-useful.html Call us :+919315540497

    1. Task Management Softwares Task Management Software is used to manage tasks by monitoring employees . You can track employees in real time, which will allow employees to concentrate more on their tasks.

      Visit: https://issuu.com/shivaniss1/docs/task_management.pptx Call us:+919315540497

  15. Mar 2021
    1. Source maps are on my radar, but i'm playing catchup. See: #124 (comment) for a rundown of where we are right now. Source maps are a pretty big project, it's not li
  16. Feb 2021
    1. Maryanne Garry 🐑🇳🇿. (2020, December 12). A person with the virus who, say, has lunch with friends is a witness to an event in which the virus was possibly transmitted, and a suspect who might have transmitted it to others. Our new paper in PoPS @lorraine_hope @rachelz @drayeshaverrall and Jamie Robertson https://t.co/FoOlx78HB2 [Tweet]. @drlambchop. https://twitter.com/drlambchop/status/1337676716936896512

  17. Oct 2020
    1. The purpose of this toolkit is to familiarize the user with intellectual and developmental disabilities (I/DD) and communication techniques to assist in successfully facilitating a healthy living program for individuals with I/DD. These are suggested techniques particularly helpful for those who don’t have any experience working with individuals who have I/DD.

      This toolkit can be used as a starting point for working with or developing programs that people with a developmental or intellectual disability might use. It does not go into detail but offers strategies for communication, behavior and setting learners up for success.

      10/10 It is short and highly informative with free resources within it.

    1. One of the primary tasks of engineers is to minimize complexity. JSX changes such a fundamental part (syntax and semantics of the language) that the complexity bubbles up to everything it touches. Pretty much every pipeline tool I've had to work with has become far more complex than necessary because of JSX. It affects AST parsers, it affects linters, it affects code coverage, it affects build systems. That tons and tons of additional code that I now need to wade through and mentally parse and ignore whenever I need to debug or want to contribute to a library that adds JSX support.
  18. Sep 2020
    1. Henrich defends this sweeping thesis with several studies, including a test known as the Triad Task. Subjects are shown three images—say, a rabbit, a carrot, and a cat. The goal is to match a “target object”—the rabbit—with a second object. A person who matches the rabbit with the cat classifies: The rabbit and the cat are animals. A person who matches the rabbit with the carrot looks for relationships between the objects: The rabbit eats the carrot.
  19. Aug 2020
  20. Jul 2020
  21. Jun 2020
  22. May 2020
    1. If something is important enough to write down and keep track of, it’s important enough to schedule.
    2. One of the main goals is to avoid some giant, never-ending task backlog, either within the current week’s “To do later” section, constantly being carried over week after week and growing in size, or in a separate backlog.md, or similar, file, also just growing in size and causing most tasks to be completely burried & lost. Either a task should be scheduled as “I plan to work on this” or it should be completely discarded (it can always come back later if it turns out to be important).
    1. Things that arose throughout the week that captured my interest or demanded my attention and which I felt the need or capacity to do now.
    2. Tasks which I am giving myself permission to avoid doing at all this week. The checkboxes here are for indicating that these tasks have been scheduled elsewhere.
    3. Tasks that I may have been forwarding along or which I may have initially added to the “To do later” list which I no longer plan to do – if they happen, great, but I’m not actively planning to complete them.
  23. Dec 2019
  24. Sep 2019
    1. Recently I was thrilled to learn that the web platform offers such an affordance in the form of the <details> tag.
  25. Aug 2019
    1. Montaigne called “On Vanity” one of those essays which, being quite long and not at all confined by the titles he gave them, “require a decision to read them and time set aside.”
    1. What would it take, I wonder, to post-process the kinds of PDF files that governments typically produce, in order to add Purple Numbers?


  26. Mar 2019
  27. arxiv.org arxiv.org

    1. The goal here is explicitly not to improve the state of the art in the narrow domain of restaurantbooking, but to take a narrow domain where traditional handcrafted dialog systems are known toperform well, and use that to gauge the strengths and weaknesses of current end-to-end systemswith no domain knowledge



    2. Unsurprisingly, perfectly coded rule-based systems can solve the simulated tasks T1-T5 perfectly,whereas our machine learning methods cannot. However, it is not easy to build an effective rule-based




    4. We implemented a rule-based system for this task in the followingway. We initialized a dialog state using the 3 relevant slots for this task: cuisine type, location andprice range. Then we analyzed the training data and wrote a series of rules that fire for triggers likeword matches, positions in the dialog, entity detections or dialog state, to output particular responses,API calls and/or update a dialog state. Responses are created by combining patterns extracted fromthe training set with entities detected in the previous turns or stored in the dialog state. Overall webuilt 28 rules and extracted 21 patterns. We optimized the choice of rules and their application priority(when needed) using the validation set, reaching a validation per-response accuracy of 40.7%. Wedid not build a rule-based system forConciergedata as it is even less constrained.



    1. A Network-based End-to-End Trainable Task-oriented Dialogue System

      这个end-to-end的系统,在意图识别的阶段用的是cnn+LSTM 在状态管理(belief state tracking)也用的LSTM,在policy的时候自定义了一套算法,将前面的几个输出向量做了个线性模型,输出。

    2. Finally, the policy network output is generated bya three-way matrix transformation,


    3. a distributed representationgenerated by an intent network and a probabilitydistribution over slot-value pairs called the beliefstate

      造出来的一个belief state的概念:

      由intent网络生成的分布式表示和对slot-value组的概率表示叫做belief stat。

    1. An End-to-End Trainable Neural Network Model withBelief Tracking for Task-Oriented Dialog

    1. In learning such neural network based dialogmodel, we propose hybrid offline training and on-line interactive learning methods. We first let theagent to learn from human-human conversationswith offline supervised training. We then improvethe agent further by letting it to interact with usersand learn from user demonstrations and feedbackwith imitation and reinforcement learning.


      • 1 首先离线有监督学习 人和人的对话数据
      • 2 然后让模型和人交互,基于反馈和模仿用强化学习来学习

      为了解决样本效率问题,提出了learning-from-user and learning-from-simulationl两个方案。

    2. We design neural net-work based dialog system that is able to ro-bustly track dialog state, interface with knowl-edge bases, and incorporate structured queryresults into system responses to successfullycomplete task-oriented dialog.


    3. End-to-End Learning of Task-Oriented Dialogs


    1. A potential draw-back with such pre-training approach is that themodel may suffer from the mismatch of dialoguestate distributions between supervised training andinteractive learning stages. While interacting withusers, the agent’s response at each turn has a di-rect influence on the distribution of dialogue statethat the agent will operate on in the upcoming di-alogue turns.

      策略学习也是对话过程很重要的一环。 最近的策略学习过程有用基于有监督的预训练然后线上强化学习再训练的来提高学习的方案。但是这种方案有个潜在的毛病,在离线的数据中受限于数据量,线上一旦碰到了不常见的情况,容易直接恢复不来。(这个问题应该只是推断吧?有什么实证么?)


    2. These system components areusually trained independently, and their optimiza-tion targets may not fully align with the overallsystem evaluation criteria (e.g. task success rateand user satisfaction). Moreover, errors made inthe upper stream modules of the pipeline propa-gate to downstream components and get amplified,making it hard to track the source of errors

      传统pipeline方案的问题点: 1 流程比较复杂,每步骤独立训练,但是流程输入和输出有依赖,错误放大,难以跟进。

    3. Dialogue Learning with Human Teaching and Feedback in End-to-End Trainable Task-Oriented Dialogue Systems


    1. Attention-Based Recurrent Neural Network Models for Joint Intent Detection and Slot Filling

      用一个模型来解决两个不同类型的问题,intent detect是分类,填槽是序列标注。都用基于attention机制的RNN来搞定了

    2. The attentionmechanism later introduced in [12] enables the encoder-decodermodel to learn a soft alignment and to decode at the same time.


      D. Bahdanau, K. Cho, and Y. Bengio, “Neural machine trans-lation by jointly learning to align and translate,”arXiv preprintarXiv:1409.0473, 2014

    1. We present a general solution towards building task-orienteddialogue systems for online shopping, aiming to assist on-line customers in completing various purchase-related tasks,such as searching products and answering questions, in a nat-ural language conversation manner. As a pioneering work, weshow what & how existing natural language processing tech-niques, data resources, and crowdsourcing can be leveragedto build such task-oriented dialogue systems for E-commerceusage. To demonstrate its effectiveness, we integrate our sys-tem into a mobile online shopping application. To the bestof our knowledge, this is the first time that an dialogue sys-tem in Chinese is practically used in online shopping scenariowith millions of real consumers. Interesting and insightful ob-servations are shown in the experimental part, based on theanalysis of human-bot conversation log. Several current chal-lenges are also pointed out as our future directions



      M = (I, C, A)

      I是intent,C是product category, A是商品attribute。 M是根据用户Query得到的信息的表示。

      意图分类:PhraseLDA 1000个topic

      产品分类: a CNN-based approach that resembles (Huang et al. 2013)and (Shen et al. 2014

    2. System Formalization


    1. Retrieval-based MethodsRetrieval-based methods choose a response from candidateresponses. The key to retrieval-based methods is message-response matching. Matching algorithms have to overcomesemantic gaps between messages and responses [28].


      B. Hu, Z. Lu, H. Li, and Q. Chen. Convolutional neu-ral network architectures for matching natural lan-guage sentences. InAdvances in neural informationprocessing systems, pages 2042–2050, 2014.

      单轮的匹配 match(X,Y) = X^TAy

      X:message的向量表示, y:回复的向量表示。

      H. Wang, Z. Lu, H. Li, and E. Chen. A dataset for re-search on short-text conversations. InProceedings ofthe 2013 Conference on Empirical Methods in NaturalLanguage Processing, pages 935–945, Seattle, Wash-ington, USA, October 2013. Association for Compu-tational Linguistics

      Z. Lu and H. Li. A deep architecture for matchingshort texts. InInternational Conference on Neural In-formation Processing Systems, pages 1367–1375, 2013.

      B. Hu, Z. Lu, H. Li, and Q. Chen. Convolutional neu-ral network architectures for matching natural lan-guage sentences. InAdvances in neural informationprocessing systems, pages 2042–2050, 2014

      M. Wang, Z. Lu, H. Li, and Q. Liu. Syntax-based deepmatching of short texts.InIJCAI, 03 2015

      Y. Wu, W. Wu, Z. Li, and M. Zhou. Topic augmentedneural network for short text conversation.CoRR,2016


    2. TASK-ORIENTED DIALOGUESYSTEMSTask-oriented dialogue systems have been an important branchof spoken dialogue systems. In this section, we will reviewpipeline and end-to-end methods for task-oriented dialoguesystems.


      • 1 pipeline,也就是包含SLU+DST+PL+NLG
      • 2 end-to-end
    3. 2.2 End-to-End Methods


      • 一个是信用分配问题,一个用户的反馈很难传播到上游每个组件中。
      • 另一个是问题流程的相互依赖。一个组件的输入依赖上一个组件的输出。一部分变动其他都得动。(这个真的是问题么?)


      • A network-based end-to-end trainable task-oriented di-alogue system
      • Learningend-to-end goal-oriented dialog.

      下文中,首次提出了一个联合训练dialogue state tracking和policy learning来优化得到更鲁棒的系统行为。

      • Towards end-to-end learn-ing for dialog state tracking and management us-ing deep reinforcement learning




      • Language understanding.NLU/SLU,目标是解析理解用户输入为intent,slot

      • Dialogue state tracker. 根据当前对话输入信息结合历史信息给出当前会话状态。

      • Dialogue policy learning.基于当前对话状态给出接下来要采取的行动

      • Natural language generation(NLG). 将映射的选择的动作行为转换生成对应的输出回复。

    5. 2.1.3 Policy learning

      策略学习 基于前面state tracker的状态表示,策略学习(policy learning)是来生成下一个可用的系统行动。无论是监督学习或者强化学习都可以用来优化策略学习。 H. Cuayhuitl, S. Keizer, and O. Lemon. Strategic di-alogue management via deep reinforcement learning.arxiv.org, 2015.

      通常都用一个基于规则的agent来初始化系统。 Z. Yan, N. Duan, P. Chen, M. Zhou, J. Zhou, andZ. Li. Building task-oriented dialogue systems for on-line shopping. InAAAI Conference on Artificial Intel-ligence, 2017

      然后用监督学习来基于规则生成的规则来学习。Building task-oriented dialogue systems for on-line shopping. 强化学习,Strategic di-alogue management via deep reinforcement learning.结果据说比很多系统,rule based,superviesed都好

    6. A statistical dialog system



      • S. Young, M. Gai, S. Keizer, F. Mairesse, J. Schatz-mann, B. Thomson, and K. Yu. The hidden informa-tion state model: A practical framework for pomdp-based spoken dialogue management. 在DSTC比赛中结果形式是每轮对话中每个slot的一个概率分布。各种统计学方法如下:
      • 规则集, Z. Wang and O. Lemon. A simple and generic belieftracking mechanism for the dialog state tracking chal-lenge: On the believability of observed information. InSIGDIAL Conference, pages 423–432, 2013
      • CRF S. Lee and M. Eskenazi. Recipe for building robustspoken dialog state trackers: Dialog state trackingchallenge system description. InSIGDIAL Conference,pages 414–422, 2013

        S. Lee. Structured discriminative model for dialogstate tracking. InSIGDIAL Conference, pages 442–451, 2013

      H. Ren, W. Xu, Y. Zhang, and Y. Yan. Dialog statetracking using conditional random fields. InSIGDIALConference, pages 457–461, 2013.

      • maximum entropy model J. Williams. Multi-domain learning and generaliza-tion in dialog state tracking. InSIGDIAL Conference,pages 433–441, 2013.

      • web-style ranking J. D. Williams. Web-style ranking and slu combina-tion for dialog state tracking

      深度学习的状态管理。用一个滑动窗口来在任意数量可能值上输出一个概率序列。 M. Henderson, B. Thomson, and S. Young. Deep neu-ral network approach for the dialog state tracking chal-lenge. InProceedings of the SIGDIAL 2013 Confer-ence, pages 467–471, 2013

      多领域的RNN状态跟进模型: B. Thomson, M. Gasic, P.-H. Su, D. Vandyke, T.-H. Wen, and S. Young. Multi-domain dialog state tracking using recurrent neuralnetworks.

      基于neural belief tracker(NBT)来检测slot-value对。 Neural belief tracker: Data-driven dia-logue state tracking.

    7. Dialogue State Tracking

      跟进对话状态是保障dialog system的robust的核心。主要目标是预测每轮对话的用户目标。经典的状态结构通常叫做slot-filling 或者 sematic frame.

      传统用手工规则的方法: D. Goddeau, H. Meng, J. Polifroni, S. Seneff, andS. Busayapongchai. A form-based dialogue managerfor spoken language applications. InSpoken Language,1996. ICSLP 96. Proceedings., Fourth InternationalConference on, volume 2, pages 701–704. IEEE, 1996

      基于规则的方法倾向于常见的错误,然后很多结果并不是想要的。 J. D. Williams. Web-style ranking and slu combina-tion for dialog state tracking. InSIGDIAL Conference,pages 282–291, 2014

    8. Slot filling

      填槽这个问题更多的是看成一个序列标注的问题。句子中的每个词都打上一个语义标签。输入是由词组成的句子,输出是每个词对应的slot/concept IDs.

      DBN 类的处理:

      • A Deoras and R. Sarikaya. Deep belief network basedsemantic taggers for spoken language understanding.

        L. Deng, G. Tur, X. He, and D. Hakkani-Tur. Use ofkernel deep convex networks and end-to-end learningfor spoken language understanding


      • G. Mesnil, X. He, L. Deng, and Y. Bengio. Investi-gation of recurrent-neural-network architectures andlearning methods for spoken language understanding.Interspeech, 2013.
      • K. Yao, G. Zweig, M. Y. Hwang, Y. Shi, and D. Yu.Recurrent neural networks for language understand-ing. InInterspeech, 2013
      • R. Sarikaya, G. E. Hinton, and B. Ramabhadran.Deep belief nets for natural language call-routing
      • K. Yao, B. Peng, Y. Zhang, D. Yu, G. Zweig, andY. Shi. Spoken language understanding using longshort-term memory neural networks. InIEEE Insti-tute of Electrical & Electronics Engineers, pages 189 –194, 2014
    9. Language Understanding

      目标是根据一个用户utterance/query 得到其对应的语义slot。slots是预先根据场景定于的。通常来说有两种类型的表示,一个是句子级别的类别,例如用户的意图和utterance的类别。另外一个是单词级别的信息抽取,例如命名实体和槽位填充。

      意图识别是根据一句话来检测用户的意图。 基于深度学习的意图识别: L. Deng, G. Tur, X. He, and D. Hakkani-Tur. Use ofkernel deep convex networks and end-to-end learningfor spoken language understanding. InSpoken Lan-guage Technology Workshop (SLT), 2012 IEEE, pages210–215. IEEE, 2012

      G. Tur, L. Deng, D. Hakkani-T ̈ur, and X. He. Towardsdeeper understanding: Deep convex networks for se-mantic utterance classification. InAcoustics, Speechand Signal Processing (ICASSP), 2012 IEEE Interna-tional Conference on, pages 5045–5048. IEEE, 2012.

      D. Yann, G. Tur, D. Hakkani-Tur, and L. Heck. Zero-shot learning and clustering for semantic utteranceclassification using deep learning. 2014.

      尤其是这个用CNN来抽取query vector进行query分类。 H. B. Hashemi, A. Asiaee, and R. Kraft. Query intentdetection using convolutional neural networks. InIn-ternational Conference on Web Search and Data Min-ing, Workshop on Query Understanding, 2016

      P.-S. Huang, X. He, J. Gao, L. Deng, A. Acero, andL. Heck. Learning deep structured semantic modelsfor web search using clickthrough data. InProceedingsof the 22nd ACM international conference on Confer-ence on information & knowledge management, pages2333–2338. ACM, 2013

      Y. Shen, X. He, J. Gao, L. Deng, and G. Mesnil.Learning semantic representations using convolutionalneural networks for web search. InProceedings of the23rd International Conference on World Wide Web,pages 373–374. ACM, 2014.

    1. The first step is lexical analysis, i.e. word segmentation and part-of-speech (POS)tagging. The words and POS labels are used as features in the subsequent models. Forthe shared task we used HanLP [1] as our Chinese lexical analyzer.

      SLU 模型做法:

      • 1 第一步是词汇分析,也就是分词,然后词性标注。本文用的是HanLP做词性分析。

      • 2 第二步是槽位边界检测。这个任务看成一个用BILOU进行序列标注的。我们用了基于字和词的序列标注。基于字的 版本是用一个window为7的CRF,用此法特征和词典特征,另外基于词的的CRF模型是window size为5的词法特征,词性特征和词典特征。词典特征是指“当前字词是否 prefix/infix/suffix 在实体词典中某个条目关系。”每个CRF输出n(3)个输出,这整个2n个输出用到下一步。用基于字的序列标注是为了弥补分词效果差带来的可能影响。

      • 3 第三部是槽位类型识别。用的是LR+L正则分类器,预测出的slot,上下文的字词,上下文的词性标注作为特征。

      • 4 第四步是槽位纠正。这个是为了解决因为ASR导致的错误识别造成的结果。用的是一个基于搜索的方法。鉴于已经有各种槽位类型的词典,如果一个预测出来的槽位s类型T没有在对应的槽位词典中,那么就用s作为查询词来在根据最小编辑距离来查询槽位词典中的记录。这个操作会进行两次,一个是s作为中文字符,另一个是s作为拼音来查询。最好的结果是从这两个查处的结果中重新排序后得到的。

      • 5 最后一步是意图分类。用的是XGBoost及其默认参数。用到的特征是单词token,query length,以及前面步骤预测出来的槽位。

    2. Each rule is of the form “if thequery q is listed in a particular lexicon L, and the preceding queries and their predicteddomain labels satisfy certain conditions, then q is assigned a certain intent label and,with the exception of short commands, the entire q is regarded as a slot of the typecorresponding to L.” The rules are arranged in sequential order in accordance with theirpriorities

      规则的具体形式是,"如果query q被列在了一个特定的词汇表L,并且其前面的queries和它们预测出来的领域标签满足特定条件,那么q就可以被打上一个特定的意图标签,并且对于短的命令来说,整个query q是当作对应于L的一个槽位类型".所有规则按照优先级顺序组织的。

    3. Figure 1 shows the framework of our SLU system, which consists of the context-dependent rules for entity-only queries and the context-independent model for querieswith IISPs. The entire system feeds the query to the rules first. If the rule-based compo‐nent returns null result, that means the query is judged to contain IISPs and the model-based component will continue to process it. Otherwise, it means the query is regardedas entity-only and the result of the rules is returned as the final output


    4. s in real use cases of dialog systems, the queries in the shared task can be roughlydivided into two kinds, viz. queries with intent-indicating salient phrases and querieswithout. By intent-indicating salient phrase (IISP) it is meant a phrase in the query thatshows the intent of the query. E.g. the phrase “” in the query “” andthe phrases “” in the query “” are IISPs.


  28. Feb 2019
    1. Spoken language understanding (SLU) comprises two tasks, intent identification andslot filling. That is, given the current query along with the previous queries in the samesession, an SLU system predicts the intent of the current query and also all slots (entitiesor labels) associated with the predicted intent. The significance of SLU lies in that eachtype of intent corresponds to a particular service API and the slots correspond to theparameters required by this API. SLU helps the dialog system to decide how to satisfythe user’s need by calling the right service with the right information



      • 1 意图分类的复杂性
      • 2 世界知识
      • 3 用户状态
    1. 对话管理也可以看成是一个分类任务,即每个对话状态和一个合适的对话动作相对应.和其它有监督的学习任务一样,分类器可以从标注的语料库中训练得到.但是,在某状态下系统应该选择的动作不能仅仅是模仿在训练数据中同一状态对应的动作,而应该是选择合适的动作能够导致一个成功的对话.因此,把对话过程看成是一个决策过程更为合适,从而根据对话的整体成功来优化动作的选择过程[32].因而这是一个规划问题,并且可以用强化学习[33]方法学习获得最优的结果
    2. 对话系统从本体构成和业务逻辑角度,可分为领域任务型和开放型的信息交互.领域任务型系统针对具体应用领域,具有比较清晰的业务语义单元的定义、本体结构以及用户目标范畴,例如航班查询、视频搜索、设备控制等等,这类交互往往是以完成特定的操作任务作为交互目标;而开放型信息交互则不针对特定领域,或说面向非常广泛的领域,交互目标并非业务任务,而是满足用户其它方面的需求,例如开放的百科问答、聊天等.它虽然能一定程度上显示人工智能的力量,但因其并不专注于帮助人解决现实任务问题,其实际使用范围较为狭窄.近年来,随着移动终端的高速发展,面向任务的自然人机对话系统和相关的认知控制理论得到了越来越多的学术和产业界重视,这也是本文讨论的重点
    1. We com-plement recent work by showing the effec-tiveness of simple sequence-to-sequenceneural architectures with a copy mecha-nism. Our model outperforms more com-plex memory-augmented models by 7% inper-response generation and is on par withthe current state-of-the-art on DSTC2, areal-world task-oriented dialogue dataset

      用一个带有copy机制的简单seq2seq框架超过现有最好的真实DSTC2 7个点。

    1. Both NLU and NLG are implementedwith template-based models


    2. Symptom ExtractionWe follow the BIO(begin-in-out) schema for symptom identification(Figure 1). Each Chinese character is assigned alabel of ”B”, ”I” or ”O”. Also, each extractedsymptom expression is tagged withTrueorFalseindicating whether the patient suffers from thissymptom or not. In order to improve the anno-tation agreement between annotators, we createtwo guidelines for the self-report and the conver-sational data respectively. Each record is anno-tated by at least two annotators. Any inconsis-tency would be further judged by the third one.The Cohen’s kappa coefficient between two anno-tators are71%and67%for self-reports and con-versations respectively

      症状数据抽取,BIO格式。每个中文字符标注为“B","I","O".每个抽取出的症状根据病人真实情况打标为“True","False"。3人2个都标过的才有效,第三人评判。Cohhen kappa 相关性来作为标注标准。

    3. In this paper, we make a move to builda dialogue system for automatic diagno-sis. We first build a dataset collected froman online medical forum by extractingsymptoms from both patients’ self-reportsand conversational data between patientsand doctors. Then we propose a task-oriented dialogue system framework tomake the diagnosis for patients automat-ically, which can converse with patients tocollect additional symptoms beyond theirself-reports. Experimental results on ourdataset show that additional symptoms ex-tracted from conversation can greatly im-prove the accuracy for disease identifica-tion and our dialogue system is able tocollect these symptoms automatically andmake a better diagnosis

      In this paper, we make a move to builda dialogue system for automatic diagno-sis. We first build a dataset collected froman online medical forum by extractingsymptoms from both patients’ self-reportsand conversational data between patientsand doctors. Then we propose a task-oriented dialogue system framework tomake the diagnosis for patients automat-ically, which can converse with patients tocollect additional symptoms beyond theirself-reports. Experimental results on ourdataset show that additional symptoms ex-tracted from conversation can greatly im-prove the accuracy for disease identifica-tion and our dialogue system is able tocollect these symptoms automatically andmake a better diagnosis


    1. PyDial: A Multi-domain Statistical Dialogue System Toolkit


      其总的架构包含Sematic Decode,Belief Tracker,Policy Reply System,Language generator. 整体来说整个系统都支持了基于规则的判断过程,也融合了模型的支持。源码值得一看的。

    1. the dataset used in our experiment hasonly the tags of filled information slots extracted by patternmatching between dialogue log and final order information

      用到的数据集是一个coffee ordering的对话过程的数据,31567通对话,142412对会话。数据只有用正则匹配出来的填充的标签信息。

    2. the agent model swaps the in-put and output sequences, and it also takes the tag of filledinformation slots as an input which is extracted from dia-logue in previous turns by pattern matching with the orderinformation in ground truth

      agent model 构建前先预训练。网络结构和user model一样,但是输入和输出反转,同时也把之前对话中已经填充的槽位信息作为输入。但是这俩部分信息并不是简单的直接拼接在一起,而是来学习适合的attention 权重来更好的利用注意力机制。此外任何其他额外的语义意图标签都不必用。

    3. By directly learningfrom the raw dialogue logs, the network takes the agent ut-Figure 2: The network structure: encoder-decoder structurewith the attention mechanismteranceXa:xa1;xa2;:::;xanas the input sequence and takescorresponding user utteranceYu:yu1;yu2;:::;yumas the tar-get sequence.

      User model.直接用双向的LSTM,以agent的utterance作为X,对应的用户的utterance作为Y。

    4. In the task-oriented dialogues, a user usually firstly showsthe intention to the agent and then answers the agent’s ques-tions one by one to specify the demand.


      用户通常是被动的,偶尔的有一轮问题。换句话说用户基本上都是在一轮中回答由agent提出的问题。所以可以基于一个用户只需要考虑一轮回答来给出回复这样的假设来构建user model,,让agent model来处理多轮对话。

    5. we propose a uSer andAgent Model IntegrAtion (SAMIA) framework inspired byan observation that the roles of the user and agent models areasymmetric. Firstly, this SAMIA framework model the usermodel as a Seq2Seq learning problem instead of ranking ordesigning rules. Then the built user model is used as a lever-age to train the agent model by deep reinforcement learning.In the test phase, the output of the agent model is filtered bythe user model to enhance the stability and robustness. Ex-periments on a real-world coffee ordering dataset verify theeffectiveness of the proposed SAMIA framework.


    1. Simply put, your frog is the most important task. It is the thing that moves the needle the most towards your ideal future. It is probably very important, but not very urgent. So our advice (and Brian Tracy’s as well) is to make sure you tackle this MIT at the beginning of your day when you have the most energy and willpower available. Tackling your MIT first thing is extremely important!

      MIT in morning

      max willpower (not decreased cause of decision fatigue during day)

      • max impact on long term goal.. -important but not urgent -procrastinate because (a) big, overwhelm where to start (b) fear of failure - perfection , sense of worth linked
    2. If I were to summarize the book in a single statement, it would be that you should do your most important task (MIT) first thing in the morning.

      eat that frog summary

  29. Oct 2018
    1. Instead of continuing the millennia old search for the universal ontology, different types of ontologies have been proposed in computer science. [.] [C]lassification of ontologies [is] based on their granularity and [their] thematic scope [differs] into top-level, domain, task, and application ontologies ... (Hitzler, Janowicz, Berg-Cross, Obrst, Sheth, Finin, Cruz 2012)
  30. Oct 2017
    1. Or, as a soldier of a desert war wrote in last autumn’s New York Times, is our central task the task of learning how to die—not (as he put it) to die ‘as individuals, but as a civilization’ (Scranton, 2013), in the Anthropocene?

      I found this statement incredibly depressing yet profound. Depressing in the idea that our central task is learning how to die (really who wants to be that morbid and think like that) (potentially digital humanists?), yet profound, because the soldier is not talking about us as individuals, but as a human civilization, as a whole, as a group, as a collective.

  31. Apr 2017
    1. was extremely saddened to read of his children’s recent exposure to measles due the Disneyland outbreak. To read about my journey leaving the anti-vaccination movement, click here.

      Italicized text framing the rest of the article. Editor comments about this being an anti vaccination article. Notes their emotional response (sad) to the phenomena in the article.

  32. Mar 2017
    1. Learningto complete a whole task involves four levels ofinstruction: (a) the problem, (b) the tasks re-quired to solve the problem, (c) the operationsthat comprise the tasks, and (d) the actions thatcomprise the operations. Effective instructionshould engage students in all four levels of per-formance: the problem level, the task-level, theoperation-level, and the action-level.

      Steps is learning to complete a whole task. This could be an extension of our matrix.

  33. Sep 2016
    1. Activities such as time spent on task and discussion board interactions are at the forefront of research.

      Really? These aren’t uncontroversial, to say the least. For instance, discussion board interactions often call for careful, mixed-method work with an eye to preventing instructor effect and confirmation bias. “Time on task” is almost a codeword for distinctions between models of learning. Research in cognitive science gives very nuanced value to “time spent on task” while the Malcolm Gladwells of the world usurp some research results. A major insight behind Competency-Based Education is that it can allow for some variance in terms of “time on task”. So it’s kind of surprising that this summary puts those two things to the fore.

  34. May 2016
  35. Jun 2015
    1. Email address
      1. Email address and name fields need to be longer.
      2. Is it clear that we need one or more of email and phone?

      Do we actually ever need both? If not, we could just ask "How would you like us to contact you?" And then show the relevant field with progressive disclosure.

  36. Feb 2015
    1. Week 3 (Jan 27) - Tasks - Ch 3 Discuss the strengths and limitations of vis tools that are for a specific purpose and those that are general. Distinguish among the three levels of actions in the task abstraction framework. Given a visualization, identify the actions the vis allows and the targets of those actions. Transform domain-specific tasks into the task abstraction framework.

      Important chapter for why a vis tool is being used.