UE 903 NLG class: center, middle background-image:url(images/data-background-light.jpg) # UE903 EC1: Application to Text ## Natural Language Generation (NLG) ## Introduction ### Claire Gardent <br><br> <img src="images/lecture1/logos-synalp-transparent.png" width="60%"/> <br><br><br> .footnote[.bold[[Claire Gardent](mailto:claire.gardent@loria.fr) CNRS / LORIA]] --- .left-column[ ## Today ] .right-column[ ### Course Logistics ### What is Text Production ? ### Pre-Neural Approaches to Text Production ] --- class: middle, center, large # Logistics --- .left-column[ ## UE903 ### Course Logistics ] .right-column[ ### Setting Up the Scene * Two Lectures ### Studying the various Approaches * Presentations and Quizzes ### Final Exam * 1 hour: NLG (C. Gardent) - 1 hour: Sentiment Analysis, etc. (D. Langlois) ### Contact and Documentation * Email: claire.gardent@loria.fr * Website: https://members.loria.fr/CGardent/teaching/ue903/README.html ] --- .left-column[ ## UE903 ### Course Logistics ] .right-column[ ### Schedule and syllabus <br> | **Event type** | **Date**|**Time** | **Description** | **Course material** | |:-----------------------|:--------|:--------|:------------------------------------------------------|-----------------------------------------------------------:| | Session 1 | 10/09 |10-12pm |**Introduction to Text Production** | [[slides]](https://members.loria.fr/CGardent/teaching/ue903/l0-introduction.pdf)| | | | | **Pre-Neural Approaches to Text Production** | [[slides]](https://members.loria.fr/CGardent/teaching/ue903/l1-preneural.pdf) | | Session 2 | 17/09 |10-12am |**Neural Approaches to Text Production** | [[slides]](https://members.loria.fr/CGardent/teaching/ue903/) | | Session 3 | 24/09 |2-5pm |**Presentations + Quiz** | | | Session 4 | 26/09 |10-12pm |**Presentations + Quiz** | | | Session 5 | 01/10 |10-12pm |**Presentations + Quiz** | | | Session 6 | 03/10 |10-12pm |**Presentations + Quiz** | | | Exam | 06/02 |2-4pm | | | | ] --- .left-column[ ## UE903 ### Course Logistics ] .right-column[ ### Presentations + Quiz <br> * Presentations - [pre-assigned](https://homepages.loria.fr/CGardent/teaching/ue903/README.html) - 10 minutes + 5 mn Q&A - Based on scientific paper - Slides sent to the class at the latest at 12pm on the day before the presentation - graded * Quiz on the presented scientific papers - 15 minutes per paper (usually half an hour for 2 papers) - slides and paper are available - graded ] --- .left-column[ ## UE903 ### Course Logistics ] .right-column[ ### Presentations + Quiz <br> * [Reading List](https://homepages.loria.fr/CGardent/teaching/ue903/README.html) * Each student presents one paper (see [List](https://homepages.loria.fr/CGardent/teaching/ue903/README.html)) * All students read all papers * The slides for the presentation must be sent to the lecturer AND to the students at the latest at 12pm before the presentation day * After the presentations, a written quiz must be answered by each student . Usually the quiz will bear on two (related) presentations. So roughly, there will be half an hour of talks and Q/A followed by half an hour of written quiz. * Presentation: 10', Q&A: 5', Quiz: 15' ] --- .left-column[ ## UE903 ### Course Logistics ] .right-column[ ### Grading * 50% Written Exam * 10% Presentation * 40% Quiz scores ] --- class: middle, center, large # Questions ? --- .left-column[ ## UE903 ### Course Logistics ## Introduction ] .right-column[ ### What is Text Production ? Input and goals define different types of text production tasks <img src="images/lecture1/inputsandgoals.png" width="90%"/> ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ] .right-column[ ### Three Main Types of Input #### Meaning Representations * Abstract Meaning Representations * Dependency Trees (Deep and Shallow) * Discourse Representation Structures * Logical Formulae (First Order Logic, Description Logic ) #### Data * Knowldege Bases * Data Bases * Numerical data from signal processing #### Text * Short, Long * Multiple or single document * Dialog or Discourse ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ] .right-column[ ### Various Types of Communicative Goals #### Describing, Verbalising * a KB fragment * an entity in a DB * an image, a video #### Summarising * A text * Several texts * The content of a KB #### Simplifying * For children, foreigners, disabled people #### Paraphrasing, Reformulating * Expert/non expert * Varied chatbot output ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ### Example Applications ] .right-column[ ### Example 1: Verbalising Knowledge Base Fragments  ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ### Example Applications ] .right-column[ ### Example 2: Generating from Data Bases  <small>[Angeli, Liang and Klein 2010](http://www.aclweb.org/anthology/D10-1049) ; [Konstas and Lapata 2012](https://aclanthology.info/pdf/P/P12/P12-1039.pdf)</small><!-- .element: style="text-align: left; float: right; width: 5m0%; font-size: small;" --> ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ### Example Applications ] .right-column[ ### Example 3: Captioning Images and Videos [Vinyals et al., 2015](https://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Vinyals_Show_and_Tell_2015_CVPR_paper.pdf) .center[<img src="images/lecture1/img2txt.png" width="60%"/>] ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ### Example Applications ] .right-column[ ### Example 4: Summarising .center[<img src="images/lecture1/summarising.png" width="70%">] [Grusky et al., 2018](http://aclweb.org/anthology/N18-1065)<!-- .element: style="text-align: right; float: right; width: 50%; font-size: small;" --> ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ### Example Applications ] .right-column[ ### Example 5: Generating Headlines .center[<img src="images/lecture1/headline.png" width="60%">] [Chopra et al., 2016](http://www.aclweb.org/anthology/N16-1012)<!-- .element: style="text-align: right; float: right; width: 50%; font-size: small;" --> ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ### Example Applications ] .right-column[ ### Example 6: Summarising Multiple Documents .center[<img src="images/lecture1/multidoct.png" width="80%">] ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ### Example Applications ] .right-column[ ### Example 7: Conversational Agents .pull-left[<img src="images/lecture1/chatbot.png" width="50%">] .pull-right[ A: Where are you going ?<br> B: I'm going to the police station.<br> A: I'll come with you<br> B: No, no, non, you're not going anywhere<br> A: Why?<br> B: I need you to stay here<br> A: I don't know what you are talking about. ] <small>[Li et al., 2016a](http://www.aclweb.org/anthology/N16-1014); [Li et al., 2016b](http://www.aclweb.org/anthology/P16-1094) </small><!-- .element: style="text-align: right; float: right; width: 50%; font-size: small;" --> ] --- .left-column[ ## UE903 ### Course Logistics ### Introduction ### Example Applications ### Summary ] .right-column[ ### NLG * Many different inputs *Data, Meaning Representations, Text* * Many different communicative goals *Verbalise, summarise, compress, simplify, respond, compare* …. ### Some Terminology * D2T Generation: Generating from Data * MR2T Generation: Generating from Meaning Representations * T2T Generation: Generating from Text Summarisation, Compression, Paraphrasing, Simplification ] <section data-background-iframe="https://nbviewer.jupyter.org/urls/mastertal.gitlab.io/UE803/notebooks/Exercise_sheet_7.ipynb" data-background-interactive> </section>