later

In this episode, Robin Jia talks about how to build robust NLP systems. We discuss the different senses in which a system can be robust, reasons to care about system robustness, and the challenges involved in evaluating robustness of NLP models. We talk about how to build certifiably robust models through interval bound propagation and discrete encoding functions, as well as how to modify data collection procedures through active learning for more robust model development.

Robin Jia is currently a visiting researcher at Facebook AI Research, and will be an assistant professor in the Department of Computer Science at the University of Southern California starting Fall 2021.

123 - Robust NLP, with Robin Jia

We invited Jayant Krishnamurthy and Hao Fang, researchers at Microsoft Semantic Machines to discuss their platform for building task-oriented dialog systems, and their recent TACL paper on the topic. The paper introduces a new formalism for task-oriented dialog to effectively handle references and revisions in complex dialog, and a large realistic dataset that uses this formalism.

Leaderboard associated with the dataset: https://microsoft.github.io/task_oriented_dialogue_as_dataflow_synthesis/
Jayant's Twitter handle: https://twitter.com/jayantkrish
Hao's Twitter handle: https://twitter.com/hfang90

124 - Semantic Machines and Task-Oriented Dialog, with Jayant Krishnamurthy and Hao Fang

How can we build Visual Question Answering systems for real users? For this episode, we chatted with Danna Gurari, about her work in building datasets and models towards VQA for people who are blind. We talked about the differences between the existing datasets, and Vizwiz, a dataset built by Gurari et al., and the resulting algorithmic changes. We also discussed the unsolved challenges in this field, and the new tasks they result in.

Danna Gurari is an Assistant Professor as well as Founding Director of the Image and Video Computing group in the School of Information at University of Texas at Austin (UT-Austin).

Vizwiz project page: https://vizwiz.org/

The hosts for this episode are Ana Marasović and Pradeep Dasigi.

125 - VQA for Real Users, with Danna Gurari

We invited Lisa Li to talk about her recent work, Prefix-Tuning: Optimizing Continuous Prompts for Generation. Prefix tuning is a lightweight alternative to finetuning, and the idea is to tune only a fixed-length task-specific continuous vector, and to keep the pretrained transformer parameters frozen. We discussed how prefix tuning compares with finetuning and other efficient alternatives on two tasks in various experimental settings, and in what scenarios prefix tuning is preferable.

Lisa is a Phd student at Stanford University. Lisa's webpage: https://xiangli1999.github.io/

The hosts for this episode are Pradeep Dasigi and Ana Marasović.

126 - Optimizing Continuous Prompts for Generation, with Lisa Li

Humanitarian AI Today’s guest host Mark Dalton from the United Nations speaks with Monica Lam, Professor and Faculty Director of Stanford University's Open Virtual Assistant Lab (OVAL) about her team's work and the importance of voice technology to the humanitarian community.

Monica Lam from Stanford University's Open Virtual Assistant Lab

In this episode, we talk to Shunyu Yao about recent insights into how transformers can represent hierarchical structure in language. Bounded-depth hierarchical structure is thought to be a key feature of natural languages, motivating Shunyu and his coauthors to show that transformers can efficiently represent bounded-depth Dyck languages, which can be thought of as a formal model of the structure of natural languages. We went on to discuss some of the intuitive ideas that emerge from the proofs, connections to RNNs, and insights about positional encodings that may have practical implications. More broadly, we also touched on the role of formal languages and other theoretical tools in modern NLP.

Papers discussed in this episode:

- Self-Attention Networks Can Process Bounded Hierarchical Languages (https://arxiv.org/abs/2105.11115)
- Theoretical Limitations of Self-Attention in Neural Sequence Models (https://arxiv.org/abs/1906.06755)
- RNNs can generate bounded hierarchical languages with optimal memory (https://arxiv.org/abs/2010.07515)
- On the Practical Computational Power of Finite Precision RNNs for Language Recognition (https://arxiv.org/abs/1805.04908)

Shunyu Yao's webpage: https://ysymyth.github.io/

The hosts for this episode are William Merrill and Matt Gardner.

129 - Transformers and Hierarchical Structure, with Shunyu Yao

In this episode, we invite John Hewitt to discuss his take on how to probe word embeddings for syntactic information. The basic idea is to project word embeddings to a vector space where the L2 distance between a pair of words in a sentence approximates the number of hops between them in the dependency tree. The proposed method shows that ELMo and BERT representations, trained with no syntactic supervision, embed many of the unlabeled, undirected dependency attachments between words in the same sentence.

Paper: https://nlp.stanford.edu/pubs/hewitt2019structural.pdf
GitHub repository: https://github.com/john-hewitt/structural-probes
Blog post: https://nlp.stanford.edu/~johnhew/structural-probe.html
Twitter thread: https://twitter.com/johnhewtt/status/1114252302141886464
John's homepage: https://nlp.stanford.edu/~johnhew/

88 - A Structural Probe for Finding Syntax in Word Representations, with John Hewitt

In this episode, we invite Zhou Yu to give an overview of dialogue systems. We discuss different types of dialogue systems (task-oriented vs. non-task-oriented), the main building blocks and how they relate to other research areas in NLP, how to transfer models across domains, and the different ways used to evaluate these systems. Zhou also shares her thoughts on exciting future directions such as developing dialogue methods for non-cooperative environments (e.g., to negotiate prices) and multimodal dialogue systems (e.g., using video as well as audio/text).

Zhou Yu's homepage: http://zhouyu.cs.ucdavis.edu/

89 - Dialog Systems, with Zhou Yu

What function do the different attention heads serve in multi-headed attention models?  In this episode, Lena describes how to use attribution methods to assess the importance and contribution of different heads in several tasks, and describes a gating mechanism to prune the number of effective heads used when combined with an auxiliary loss. Then, we discuss Lena’s work on studying the evolution of representations of individual tokens in transformers model.

Lena’s homepage:
https://lena-voita.github.io/

Blog posts:
https://lena-voita.github.io/posts/acl19_heads.html
https://lena-voita.github.io/posts/emnlp19_evolution.html

Papers:
https://arxiv.org/abs/1905.09418
https://arxiv.org/abs/1909.01380

98 - Analyzing Information Flow In Transformers, With Elena Voita

In this episode we talked with Victor Sanh and Thomas Wolf from HuggingFace about model distillation, and DistilBERT as one example of distillation. The idea behind model distillation is compressing a large model by building a smaller model, with much fewer parameters, that approximates the output distribution of the original model, typically for increased efficiency. We discussed how model distillation was typically done previously, and then focused on the specifics of DistilBERT, including training objective, empirical results, ablations etc. We finally discussed what kinds of information you might lose when doing model distillation.

104 - Model Distillation, with Victor Sanh and Thomas Wolf

In this episode, we invite Hao Tan and Mohit Bansal to talk about multi-modal training of transformers, focusing in particular on their EMNLP 2019 paper that introduced LXMERT, a vision+language transformer.  We spend the first third of the episode talking about why you might want to have multi-modal representations.  We then move to the specifics of LXMERT, including the model structure, the losses that are used to encourage cross-modal representations, and the data that is used.  Along the way, we mention latent alignments between images and captions, the granularity of captions, and machine translation even comes up a few times.  We conclude with some speculation on the future of multi-modal representations.

Hao's website: http://www.cs.unc.edu/~airsplay/
Mohit's website: http://www.cs.unc.edu/~mbansal/
LXMERT paper: https://www.aclweb.org/anthology/D19-1514/

107 - Multi-Modal Transformers, with Hao Tan and Mohit Bansal

How do we know, in a concrete quantitative sense, what a deep learning model knows about language? In this episode, Ellie Pavlick talks about two broad directions to address this question: structural and behavioral analysis of models. In structural analysis, we often train a linear classifier for some linguistic phenomenon we'd like to probe (e.g., syntactic dependencies) while using the (frozen) weights of a model pre-trained on some tasks (e.g., masked language models). What can we conclude from the results of probing experiments? What does probing tell us about the linguistic abstractions encoded in each layer of an end-to-end pre-trained model? How well does it match classical NLP pipelines? How important is it to freeze the pre-trained weights in probing experiments? In contrast, behavioral analysis evaluates a model's ability to distinguish between inputs which respect vs. violate a linguistic phenomenon using acceptability or entailment tasks, e.g., can the model predict which is more likely: "dog bites man" vs. "man bites dog"? We discuss the significance of which format to use for behavioral tasks, and how easy it is for humans to perform such tasks.

Ellie Pavlick's homepage: 
https://cs.brown.edu/people/epavlick/

BERT rediscovers the classical nlp pipeline , by Ian Tenney, Dipanjan Das, Ellie Pavlick
https://arxiv.org/pdf/1905.05950.pdf?fbclid=IwAR3gzFibSBoDGdjqVu9Gq0mh1lDdRZa7dm42JuXXUfjG6rKZ44iHIOdV6jg

Inherent Disagreements in Human Textual Inferences by Ellie Pavlick and Tom Kwiatkowski 
https://www.mitpressjournals.org/doi/full/10.1162/tacl_a_00293

109 - What Does Your Model Know About Language, with Ellie Pavlick

In this episode, Tom Kwiatkowski and Michael Collins talk about Natural Questions, a benchmark for question answering research. We discuss how the dataset was collected to reflect naturally-occurring questions, the criteria used for identifying short and long answers, how this dataset differs from other QA datasets, and how easy it might be to game the benchmark with superficial processing of the text. We also contrast the holistic design in Natural Questions to deliberately targeting specific linguistic phenomena of interest when building a QA dataset.

Dataset: https://ai.google.com/research/NaturalQuestions
Paper: https://www.mitpressjournals.org/doi/full/10.1162/tacl_a_00276

110 - Natural Questions, with Tom Kwiatkowski and Michael Collins

We invited Jon Clark from Google to talk about TyDi QA, a new question answering dataset, for this episode. The dataset contains information seeking questions in 11 languages that are typologically diverse, i.e., they differ from each other in terms of key structural and functional features. The questions in TyDiQA are information-seeking, like those in Natural Questions, which we discussed in the previous episode. In addition, TyDiQA also has questions collected in multiple languages using independent crowdsourcing pipelines, as opposed to some other multilingual QA datasets like XQuAD and MLQA where English data is translated into other languages. The dataset and the leaderboard can be accessed at https://ai.google.com/research/tydiqa.

111 - Typologically diverse, multi-lingual, information-seeking questions, with Jon Clark

We invited Steven Cao to talk about his paper on multilingual alignment of contextual word embeddings. We started by discussing how multilingual transformers work in general, and then focus on Steven’s work on aligning word representations. The core idea is to start from a list of words automatically aligned from parallel corpora and to ensure the representations of the aligned words are similar to each other while not moving too far away from their original representations. We discussed the experiments on the XNLI dataset in the paper, analysis, and the decision to do the alignment at word level and compare it to other possibilities such as aligning word pieces or higher level encoded representations in transformers.

Paper: https://openreview.net/forum?id=r1xCMyBtPS
Steven Cao’s webpage: https://stevenxcao.github.io/

112 - Alignment of Multilingual Contextual Representations, with Steven Cao

We invited Marco Tulio Ribeiro, a Senior Researcher at Microsoft, to talk about evaluating NLP models using behavioral testing, a framework borrowed from Software Engineering. Marco describes three kinds of black-box tests the check whether NLP models satisfy certain necessary conditions. While breaking the standard IID assumption, this framework presents a way to evaluate whether NLP systems are ready for real-world use. We also discuss what capabilities can be tested using this framework, how one can come up with good tests, and the need for an evolving set of behavioral tests for NLP systems.

Marco’s homepage: https://homes.cs.washington.edu/~marcotcr/

114 - Behavioral Testing of NLP Models, with Marco Tulio Ribeiro

In this special episode, Carissa Schoenick, a program manager and communications director at AI2 interviewed Matt Gardner about AllenNLP. We chatted about the origins of AllenNLP, the early challenges in building it, and the design decisions behind the library. Given the release of AllenNLP 1.0 this week, we asked Matt what users can expect from the new release, what improvements the AllenNLP team is working on for the future versions.

115 - AllenNLP, interviewing Matt Gardner

polnalyubvi - Кометы (тот кто погас будет ярче светить)

We interviewed Sameer Singh for this episode, and discussed an overview of recent work in interpreting NLP model predictions, particularly instance-level interpretations. We started out by talking about why it is important to interpret model outputs and why it is a hard problem. We then dove into the details of three kinds of interpretation techniques: attribution based methods, interpretation using influence functions, and generating explanations. Towards the end, we spent some time discussing how explanations of model behavior can be evaluated, and some limitations and potential concerns in evaluation methods.

Sameer Singh is an Assistant Professor of Computer Science at the University of California, Irvine.
Some of the techniques discussed in this episode have been implemented in the AllenNLP Interpret framework (details and demo here: https://allennlp.org/interpret).

117 - Interpreting NLP Model Predictions, with Sameer Singh

In this episode, we talked about Coreference Resolution with Marta Recasens, a Research Scientist at Google. We discussed the complexity involved in resolving references in language, the simplification of the problem that the NLP community has focused on by talking about specific datasets, and the complex coreference phenomena that are not yet captured in those datasets. We also briefly talked about how coreference is handled in languages other than English, and how some of the notions we have about modeling coreference phenomena in English do not necessarily transfer to other languages. We ended the discussion by talking about large language models, and to what extent they might be good at handling coreference.

later

Ksenia