Swabha Swayamdipta @swabhz Twitter profile

Last Seen Profiles

@1000toys

@caesar_jg

@chinners1981

@mtdough

@Talon_Warburton

@SnowpiercerTV

@pugcoot

@uwaterlooAI

@fee6x

@SaironThirion

@falsereportband

@daiesie

@KQUIPkids

@yezjih

@stwmaniax

@NurasikinNur

@atz_scans

@out_kasti

@TouchdownBatman

@stw_pdg

@RabbWilder

@Yldz_Rsh

@Happinessi2

@xr9287236149573

@degentral

@jklooz33

@ChristianKise13

@Montecristo_CBA

@HuntmanTen

@Mariete_leon

@TheMBMOracle

@CeresBodegas

@daniijimenez4

@DecryptedCrypt8

@NdueDukagjini

@ElectionWiz

Swabha Swayamdipta

@swabhz

3 years

I'm thrilled to share some personal news - I'll be joining the University of Southern California @CSatUSC as an Assistant Professor of CS and the Gabilan Assistant Professor in Fall 2022. Super excited to be part of the NLP group at USC @nlp_usc and more broadly, SoCal NLP 😃 🏖️

96

25

777

Swabha Swayamdipta

@swabhz

2 years

I'm super honored to receive an outstanding paper award at NeurIPS 2021 for MAUVE🟪 with the awesome team of @KrishnaPillutla @rown @jwthickstun @wellecks @YejinChoinka and Zaid Harchaoui! Learn more about our work 👉

Gautam Kamath

@thegautamkamath

2 years

Outstanding Paper Award 4. MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontiers, by @KrishnaPillutla , @swabhz , @rown , @jwthickstun , @wellecks , @YejinChoinka , Zaid Harchaoui (4/n)

5

21

107

18

34

369

Swabha Swayamdipta

@swabhz

2 years

🎉🎉Super thrilled that our paper on Understanding Dataset Difficulty with V-usable information received an outstanding paper award at #ICML2022 !! 🥳Looking forward to the broader applications of this framework. It was a total delight working with my @allen_ai intern, @ethayarajh

Kawin Ethayarajh

@ethayarajh

2 years

Tomorrow at #ICML2022 , I'll be presenting a long talk on understanding dataset difficulty (Room 301, 11.00 am EST) Would love to see you there!

2

11

111

16

29

349

Swabha Swayamdipta

@swabhz

4 years

Somebody in my group mentoring session today at #acl2020nlp asked me how many papers they need for a PhD, and somebody else was talking about number of citations on their papers. This is on the mind of a *LOT* of PhD students and even undergrads now (it certainly was on mine).

9

45

350

Swabha Swayamdipta

@swabhz

3 years

Excited to announce that I'll be recruiting PhD students 🎓 at @CSatUSC for Fall'22! My research interests include the estimation of dataset quality, interpretability, and data curation towards generalization and bias mitigation in NLP 📚. Apply by Dec 15🕰️

Ph.D. Application Information - USC Viterbi | Thomas Lord Department of Computer Science

PhD Program - Application Information:Application Deadline: Fall Semester - December 15th. *Please note: The Computer Science Department does not accept applications for the Spring semester. How to...

www.cs.usc.edu

12

76

315

Swabha Swayamdipta

@swabhz

4 years

As datasets have grown larger, data exploration has become increasingly challenging. Our new work on Dataset Cartography, at @emnlp2020 with @royschwartz02 , @NickLourie , @yizhongwyz , @HannaHajishirzi , @nlpnoah , @YejinChoinka offers a solution 🗺️ Paper: 1/n

2

60

250

Swabha Swayamdipta

@swabhz

2 years

My heart goes out to my @USCViterbi students from Iran who are still able to show up to class and participate despite everything going on at home. I can only imagine what you're going through. Please let me know how I can help as an instructor and advisor new to the job 🇮🇷

9

22

224

Swabha Swayamdipta

@swabhz

6 months

What are the generative limits of LLMs? Can LLMs, biases and all, still help us understand our society? I'm recruiting PhD students to study these questions and many more at the DILL lab @nlp_usc ! Check out our recent work and your research alignment:

0

48

184

Swabha Swayamdipta

@swabhz

3 years

Experience at #emnlp2020 mentoring session: met some early PhD students who expressed concerns about the relevance of their research when "everyone" seems to be working on BERT. It pays to be creative in research, even though it might not seem so right away. (1/3)

4

21

167

Swabha Swayamdipta

@swabhz

2 years

How common is it nowadays for high school students to seek out research opportunities in AI? I've come across a couple of highly enterprising highschoolers who already do robotics but seem stressed out about getting research internships in academia / industry. Is this normal now?

29

4

147

Swabha Swayamdipta

@swabhz

4 years

@IAugenstein wrote such a beautiful post about this which I return to every now and then to keep myself sane:

Increasing Well-Being in Academia

Academia is far from the best work environment from a mental health perspective. From reports on high suicide rates among PhD students to…

isabelle-augenstein.medium.com

1

35

136

Swabha Swayamdipta

@swabhz

4 years

I believe we can do better as a community by fostering an environment of growth and learning, instead of raw empirical evaluation of ourselves. Too much focus on how to write as many papers as possible is absolute bananas! Not good for science, not good for mental health!

2

8

131

Swabha Swayamdipta

@swabhz

2 years

Thanks to #ACL2022 for this incredible opportunity, and for letting me present remotely! It was an absolute pleasure to see my friends and contemporaries in the session all work on really exciting directions in our relatively new careers ✨

Allen Institute for AI

@allen_ai

2 years

Don't miss Young Rising Star @swabhz at #ACL2022 today! She'll be presenting her Spotlight Talk "The Devil’s in the Data: Mapping and Generating Datasets for Robust Generalization" -

0

4

25

4

6

121

Swabha Swayamdipta

@swabhz

2 years

How can we build better counterfactuals for data augmentation? Check out our #emnlp2022 paper: "NeuroCounterfactuals: Beyond Minimal-Edit Counterfactuals for Richer Data Augmentation"() w/ Phillip Howard, Gadi Singer, @vasudev_lal and @YejinChoinka 🧵1/

2

26

117

Swabha Swayamdipta

@swabhz

4 years

We 👏 are 👏 more 👏than 👏 our 👏 num-papers / 👏num-citations! 👏

1

8

116

Swabha Swayamdipta

@swabhz

2 years

You pour in everything you have into serving the needs of this truly unique country and in an instant USCIS makes you feel so so insignificant. I cannot understand why my parents (almost in their 70s now) need to wait for nearly two years to come visit me for two weeks

Deedy

@deedydas

2 years

1/9 Indians in the US on a work visa can't go home. There’s NO appointments. The now ~200k H-1B Telegram group is growing by 13,000 per month. Even with ~$40m in fees, the 5 VACs process ~10,000 a month. The next availability is ~2024! Here’s 6 stories from hurting families:

164

1K

5K

2

8

114

Swabha Swayamdipta

@swabhz

1 year

I am actively recruiting one PhD student interested in constrained decoding / language generation 🤖 in 🍁 Fall 2023! Deadline for @CSatUSC PhD Applications is Dec 15. Learn more about my lab here: and about @nlp_usc here: ☀️⛱️

USC NLP

USC NLP.

nlp.usc.edu

2

31

103

Swabha Swayamdipta

@swabhz

2 years

Super excited to introduce 🥁🥁🥁 WaNLI, our new human-AI collaborative dataset 👇👇👇

Alisa Liu

@alisawuffles

2 years

We introduce a new paradigm for dataset creation based on human 🧑‍💻 and machine 🤖 collaboration, which brings together the generative strength of LMs and the evaluative strength of humans. And we collect 🎉 WaNLI, a dataset of 108K NLI examples! 🧵 Paper:

14

215

1K

1

13

104

Swabha Swayamdipta

@swabhz

5 months

So honored and excited to step in as an Associate Director of @CAIS_USC - stoked for upcoming collaborations and hopefully impactful work🤞

USC Center for AI in Society

@CAIS_USC

6 months

Check out our new blog post introducing CAIS' new associate directors! Learn about their research, their view on the impact of AI in society, and their plans for the future of CAIS. Link: @USCViterbi @CSatUSC @USCMingHsiehEE @uscsocialwork @USCAnnenberg

0

3

10

14

102

Swabha Swayamdipta

@swabhz

2 years

Evaluating open-ended text generation can be unsatisfactory with existing metrics... Move over 🟦BLEU and 🟥 ROUGE, and try our new metric 🟪MAUVE 👇👇 📢 New paper and oral at NeurIPS 2021 by a fantastic team with @KrishnaPillutla and others ✨

Krishna Pillutla @ ICLR 2024

@KrishnaPillutla

2 years

How can we measure the gap between machine text and human text? We introduce MAUVE, a new comparison measure for open-ended text generation, in our upcoming oral presentation at NeurIPS 2021. Paper: Pip package: 1/n

3

76

301

0

19

100

Swabha Swayamdipta

@swabhz

2 months

#NLProc community: What do y'all do when a reviewer literally copies your "Limitations" text and pastes it under "Weaknesses"? I don't see any reviewing guidelines that disallow this?

How to review for ACL Rolling Review

An initiative of the Association for Computational Linguistics

aclrollingreview.org

13

5

98

Swabha Swayamdipta

@swabhz

6 years

Want to know more about frame-semantics and semantic role labeling? Check out our tutorial at @coling2018 with Collin Baker, Miriam Petruck and Michael Ellsworth. Slides for Section on Frame-SRL: . Full tutorial: Tomorrow at 9 am!

GitHub - swabhs/coling18tutorial: COLING 2018 Tutorial on Multilingual FrameNet: Automatic semantic...

COLING 2018 Tutorial on Multilingual FrameNet: Automatic semantic role labeling for FrameNet - swabhs/coling18tutorial

github.com

1

28

93

Swabha Swayamdipta

@swabhz

2 months

Unbelievably excited about this new work with the amazing @matt and @xiang on finding a model signature, based purely on LLM output logits. Our signature can uniquely identify a model and lead to super cool applications ... 1/3 🤘

Matthew Finlayson

@mattf1n

2 months

Wanna know gpt-3.5-turbo's embed size? We find a way to extract info from LLM APIs and estimate gpt-3.5-turbo’s embed size to be 4096. With the same trick we also develop 25x faster logprob extraction, audits for LLM APIs, and more! 📄 Here’s how 1/🧵

6

79

362

1

8

91

Swabha Swayamdipta

@swabhz

4 years

1/ Last week #acl2020nlp asked: can we move away from just solving datasets, and instead improve on tasks? Our paper at #ICML2020 describes a candidate solution: AFLite, a lightweight adversarial filtering algorithm for data “bias” reduction. 🚨 Paper:

Adversarial Filters of Dataset Biases

Large neural models have demonstrated human-level performance on language and vision benchmarks, while their performance degrades considerably on adversarial or out-of-distribution samples. This...

arxiv.org

1

17

90

Swabha Swayamdipta

@swabhz

4 years

Our new ACL paper on adapting LM pretraining to domains and tasks 👇TL;DR the more you pretrain LMs to unlabeled domain and task data, the better you do on the supervised task.

Suchin Gururangan

@ssgrn

4 years

This model is part of our larger ACL 2020 paper, "Don't Stop Pretraining: Adapt Language Models to Domains and Tasks"; work with @anmarasovic , @swabhz , @kylelostat , @i_beltagy , Doug Downey, and @nlpnoah

3

9

61

4

7

86

Swabha Swayamdipta

@swabhz

7 years

You could go a long way without recurrent neural nets, to answer trivia questions from text...

cs.CL Papers

@arxiv_cs_cl

7 years

Multi-Mention Learning for Reading Comprehension with Neural Cascades. (arXiv:1711.00894v1 [cs.CL]) #NLProc

0

6

16

2

24

77

Swabha Swayamdipta

@swabhz

6 years

I know it'll be the last day of #emnlp2018 ... I know morning sessions are the crunch It's a struggle to find Hall 100 But aren't we all a tough bunch? So wake up early and come to my talk on `Syntactic Scaffolds for Semantic Structures` at 9:36am! Here's a sneak peek:

0

18

74

Swabha Swayamdipta

@swabhz

2 years

This just makes common sense (to quote @complingy ) - even @YejinChoinka needs to admit it now! Can't tell you how happy this news makes me, thanks so much for being my mentor, Yejin! May your genius rub off on all of us 😉

MacArthur Foundation

@macfound

2 years

Computer Scientist @YejinChoinka uses natural language processing to develop AI systems that can understand language and make inferences about the world. Learn more about the 2022 MacArthur Fellow #MacFellow

17

81

581

2

66

Swabha Swayamdipta

@swabhz

6 years

Our #emnlp2018 paper on Syntactic Scaffolds for Semantic Structures is now live - we get improvements for three different semantic tasks using very shallow syntax in a multitask setting (a syntactic scaffold). Paper: Code:

GitHub - swabhs/scaffolding: Frame-Semantic and PropBank Semantic Role Labeling with Syntactic...

Frame-Semantic and PropBank Semantic Role Labeling with Syntactic Scaffolding. - swabhs/scaffolding

github.com

2

18

65

Swabha Swayamdipta

@swabhz

6 years

I will be presenting my work with Ankur Parikh and Tom Kwaitkowski titled Multi-Mention Learning for Reading Comprehension () using Neural Cascades at #ICLR2018 in the PM poster session today. Come check it out! @LTIatCMU @googleresearch #NLProc

2

10

62

Swabha Swayamdipta

@swabhz

2 years

In our new #NAACL2022 paper, we take a closer look at the annotators behind toxic speech identification💀 We study who finds some text toxic, and more importantly, why⁉️ We argue that toxicity in language is way too nuanced to be a simple classification task! Learn more below👇

Maarten Sap (he/him)

@MaartenSap

2 years

Really excited to share our upcoming #NAACL2022 paper where we investigate how annotators' beliefs and identities bias toxic language detection systems, with @swabhz , @Imlaura_vianna , @nlpxuhui , @yejinchoinka , and @nlpnoah 🧵1/n

1

33

173

0

6

63

Swabha Swayamdipta

@swabhz

1 year

Super proud of this work on ambiguity, such a key feature of natural language and how LLMs deal with it (or really do not). We also propose some new ways to evaluate generations. Check it all out 👇

Alisa Liu

@alisawuffles

1 year

Ambiguity is an intrinsic feature of language; as LMs are increasingly deployed to interface with human communication, handling ambiguity is critical🧐. So, we collect 🎉AmbiEnt, a benchmark with diverse ambiguities, challenging even GPT-4 🤖. 🧵

7

105

459

0

9

60

Swabha Swayamdipta

@swabhz

5 months

I'll be at #NeurIPS2023 , please come say hi if you're interested in any of the topics below 👇

Swabha Swayamdipta

@swabhz

6 months

What are the generative limits of LLMs? Can LLMs, biases and all, still help us understand our society? I'm recruiting PhD students to study these questions and many more at the DILL lab @nlp_usc ! Check out our recent work and your research alignment:

0

48

184

2

7

58

Swabha Swayamdipta

@swabhz

6 months

Do come check out what our DILL lab () 🌿 is up to! And yes, we're hiring this fall!

USC NLP

@nlp_usc

6 months

We're excited to attend #SocalNLP today! ICYMI, sunny southern California is a fantastic place to do #NLProc , come check out what USC NLP [] has been working on lately! And did we say we're hiring PhD students this fall? 🌴🏖️☀️

0

2

16

0

7

59

Swabha Swayamdipta

@swabhz

2 years

Never thought I'd be teaching Data-centric NLP to cats! 😹

Data-Centric NLP

🍂 Fall 2022

swabhs.com

Tenghao Huang

@TenghaoHuang45

2 years

9AM, the fire alarm in my apartment was triggered. I woke up and saw my cat curled up in the corner. It was almost class time. I sent an email to Professor @swabhz asking whether I could bring my cat to the classroom. She approved! This is the story behind this picture :)

2

1

40

3

2

59

Swabha Swayamdipta

@swabhz

30 days

Looking forward to welcome @mohitban47 as a distinguished lecturer to @CSatUSC tomorrow and learn about his latest work on Multimodal LLMs: It's going to be a good day at @nlp_usc

0

6

59

Swabha Swayamdipta

@swabhz

3 years

Why this? 🤷🏽‍♀️ vs. Why this vs. that? 🤔 Humans find the second more intuitive to answer. In our latest paper we explore how to get contrastive explanations of model decisions. Work by my intern at @allen_ai @ai2_mosaic : @alon_jacovi 👇👇

Alon Jacovi

@alon_jacovi

3 years

Check out our new paper 😊 (w\ @swabhz @ravfogel @yanaiela @YejinChoinka @yoavgo ) Contrastive Explanations for Model Interpretability This paper is about explaining classifier decisions contrastively against alternative decisions.

5

35

172

0

9

56

Swabha Swayamdipta

@swabhz

3 years

Can the few-shot capabilities of large language models 🤖 change how we do truly low-resource NLP? A new chapter of DeepLo 2022 - with a fantastic lineup of speakers - could provide some answers! Please vote 🗳️ for us: 🙌

deeplo-2022

Workshop Description The workshop will bring together experts in deep learning and natural language processing whose research focuses on learning with scarce data. Specifically, it will provide...

sites.google.com

0

15

50

Swabha Swayamdipta

@swabhz

1 year

At @emnlpmeeting in Abu Dhabi all week #EMNLP2022 🌴 Say hi if you're around and would like to chat!

3

1

48

Swabha Swayamdipta

@swabhz

3 years

Writing an EACL camera-ready, where we are taking a stance on an age-old, problematic norm in linguistics. So proud of my co-authors @nlpxuhui @MaartenSap @nlpnoah @YejinChoinka WAE:White-aligned English(While not perfect, I believe it's more accurate than "standard"/"general")

4

49

Swabha Swayamdipta

@swabhz

4 years

Notes for myself, to keep in mind as I advance in my career. Sharing here to hold myself accountable, as well as help others. Inspired by @timnitGebru @cephaloponderer @emilymbender and many others who have been tirelessly fighting for fairness. (0/4)

1

4

46

Swabha Swayamdipta

@swabhz

7 days

What explains the effectiveness and staying power of truncation based inference algorithms (top-p, top-k) for LLMs? @mattf1n is presenting our poster right now at #iclr24 , go find out!

USC Thomas Lord Department of Computer Science

@CSatUSC

9 days

At #ICLR this week? Explore our faculty and students' latest breakthroughs and contributions in deep learning! @USCViterbi @USCAdvComputing @USCMingHsiehEE @xiangrenNLP @SepantaZeighami @swabhz @mattf1n @gauravsukhatme @SumeetBt @caodefu_dove

2

4

19

0

8

54

Swabha Swayamdipta

@swabhz

3 years

Mental health is more important than that Jan 1st ArXiv deadline y'all!

Dr Ellie Murray, ScD

@EpiEllie

3 years

Grad students, this has been a hell of a year so please take a good long break now. Set aside your research, try new hobbies, binge watch new shows, read fiction or non-fiction from outside your field. Trust me. You will come back to your thesis with new ideas. Rest is work.

19

200

1K

2

3

42

Swabha Swayamdipta

@swabhz

2 years

How acceptable are GPT-3's explanations for common sense questions? Unsurprisingly, the answer is sometimes... Our work tries to show *when* GPT-3 explanations can be acceptable, in a new #NAACL2022 paper 👇👇👇

Sarah Wiegreffe

@sarahwiegreffe

2 years

- How good is GPT-3 at generating human-acceptable free-text explanations? - Can we produce good explanations with few-shot prompting? - Can human preference modeling produce even better explanations from GPT-3? We answer these questions and more in our #NAACL2022 paper. 🧵1/12

5

49

274

0

8

41

Swabha Swayamdipta

@swabhz

4 years

My main takeaway from ACL Day1 so far: Really, really think about what tools you *should* be building, without being tempted simply by the availability of data. From @rctatman 's #WiNLP2020 keynote: focus instead on "minimizing differences in accuracies between subgroups". (1/4)

1

11

40

Swabha Swayamdipta

@swabhz

5 years

How to explain deep learning basics to your grandparents...

Wojciech Zaremba

@woj_zaremba

5 years

15

754

3K

0

9

39

Swabha Swayamdipta

@swabhz

6 years

Want to know what NLI models might actually be learning? Come to our poster at 10:30 am in Elite Hall B to find out! Joint work with @ssgrn @omerlevy_ @royschwartz02 @sleepinyourhat @nlpnoah #NAACL2018

0

10

37

Swabha Swayamdipta

@swabhz

2 years

Hope to see a strong attendance at the 9D: Language Resources & Evaluation 3 session at #NAACL2022 (Elwha A and Underline) at 10:45am PT. We have several exciting papers on creative dataset design, including an outstanding paper presentation on the _NewsEdits_ dataset!

2

3

35

Swabha Swayamdipta

@swabhz

4 years

We're hiring interns for @ai2_mosaic Come join us for some exciting collaborations!

MOSAIC

@ai2_mosaic

4 years

The @ai2_mosaic team @allen_ai is looking for interns for 2021 (Winter, Spring, and Summer). Applications are now open! Apply here: (Deadline: November 30th, 2020) AI2's internship program is open to international students!

1

6

38

1

6

33

Swabha Swayamdipta

@swabhz

3 years

Truly proud of this work with my awesome intern @ethayarajh 👇 While pointwise V-information estimates are orthogonal to those we used to make data maps, I couldn't resist making a plot to study the relationship :)

Kawin Ethayarajh

@ethayarajh

3 years

How can we measure dataset difficulty? And how can we understand *why* a dataset is difficult for a given NLP model? To answer these questions, @YejinChoinka @swabhz and I take an information-theoretic perspective of difficulty. @allen_ai @stanfordnlp 1/

6

84

425

0

6

32

Swabha Swayamdipta

@swabhz

4 years

Updated camera-ready version and code now available! Code: Paper:

GitHub - allenai/cartography: Dataset Cartography: Mapping and Diagnosing Datasets with Training...

Dataset Cartography: Mapping and Diagnosing Datasets with Training Dynamics - allenai/cartography

github.com

Swabha Swayamdipta

@swabhz

4 years

As datasets have grown larger, data exploration has become increasingly challenging. Our new work on Dataset Cartography, at @emnlp2020 with @royschwartz02 , @NickLourie , @yizhongwyz , @HannaHajishirzi , @nlpnoah , @YejinChoinka offers a solution 🗺️ Paper: 1/n

2

60

250

0

6

32

Swabha Swayamdipta

@swabhz

6 years

Our new @NAACLHLT paper - it's remarkably easy to predict entailment classes looking at just the hypothesis.

Omer Levy

@omerlevy_

6 years

TL;DR: A simple text classifier on the hypothesis (that completely ignores the premise) correctly predicts 67% of SNLI because of the heuristics turkers used to create the data. SotA models exploit this, and their actual performance on NLI is much lower.

3

69

193

1

6

32

Swabha Swayamdipta

@swabhz

3 years

Also, remember that people who built the things that seem cool today probably started working on it long back, when they weren't. (3/3)

0

3

31

Swabha Swayamdipta

@swabhz

1 year

Thanks to the Online Asian ML School ( #OAMLS ) organizers at @ACMLConf for an opportunity to teach a class on Data-Centric ML (slides: ). It was my very first 3 hour lecture 😅 and that too in the middle of #EMNLP2022 . What a fantastic week at Abu Dhabi!

oamls-data-centric-machine-learning-ss.pdf

drive.google.com

ACML

@ACMLConf

1 year

Thanks to Swabha Swayamdipta ( @swabhz ) from USC Viterbi School of Engineering for a really cool talk on Data-centric Machine Learning in OAMLS, ACML 2022. For details on ACML 2022, please visit #OAMLS #ACML2022 #MachineLearning

0

1

5

0

5

29

Swabha Swayamdipta

@swabhz

3 months

Looking forward to hosting you at @nlp_usc !

Yanai Elazar

@yanaiela

3 months

This upcoming week, I'll be in the LA area, giving talks at UCLA, UCSB, and USC about recent works involving data, models, and whatever is in between.

4

1

72

0

1

30

Swabha Swayamdipta

@swabhz

3 years

Wait till they tell you to your face in interviews that your chances look good cz you're female

Ana Marasović

@anmarasovic

3 years

Universities be like "submit a diversity statement and when you do that, also specify your gender and you better be one of the two options we like". What a joke

2

3

42

1

0

30

Swabha Swayamdipta

@swabhz

6 years

Totally loved @strubell 's fantastic animations - she made it look so easy. Further evidence that incorporating a syntactic inductive bias is key for NLU. Also check out our super related work on syntactic scaffolds, with improvement on SRL and coreference:

emnlp2020

@emnlp2020

6 years

First #emnlp2018 best long paper talk! Emma Strubell on "Linguistically-Informed Self-Attention for Semantic Role Labeling"

0

24

136

1

4

30

Swabha Swayamdipta

@swabhz

7 years

Yes, the rumors are true!

3

1

29

Swabha Swayamdipta

@swabhz

6 years

True story: At ICLR this year, I went up to a poster by someone I met the previous day, and had to ask him thrice about his poster, before he actually gave me his pitch. He seemed more interested in chitchat. Many more stories, but c'mon - minorities can also talk research!

Maarten Sap (he/him)

@MaartenSap

6 years

Avoid changing the topic to non work-related things when a woman/minority joins the convo! This happens to my female friends way too often, men (probably subconsciously) start talking about the weather instead of technical subjects.

1

3

17

1

4

29

Swabha Swayamdipta

@swabhz

4 years

Come intern with us!

Allen Institute for AI

@allen_ai

4 years

AI2's MOSAIC team is looking for research interns! Come explore the frontiers of #CommonSenseAI with us. The deadline for this round is Feb 5th – Please submit your application (and select MOSAIC as your "Desired Project Team") here: #AiJobs #AIresearch

1

18

45

1

5

29

Swabha Swayamdipta

@swabhz

3 years

You can do better, @emnlpmeeting !!

Shoaib Jameel

@shoaibjameel123

3 years

Wrote to the EMNLP 2021 PC Chairs considering extending the deadline due to COVID-19 "catastrophic" situation in India, but was turned down. I recall in EMNLP 2020 the deadline was extended due to protests in the US, which caused some controversies. Submitting to CIKM 2021 now.

9

78

324

0

1

28

Swabha Swayamdipta

@swabhz

3 years

One of the great privileges of my academic life has been to host extremely talented interns at @ai2_mosaic 💕 A summer ☀️ internship is the 💯 best way to experience the research freedom that @allen_ai offers! Consider applying to our Mosaic Team👇👇👇

MOSAIC

@ai2_mosaic

3 years

Looking for a Summer 2022 research internship? Apply to the Mosaic team @allen_ai !! topics include: commonsense reasoning, generation, vision+language, RL, + more! Applications due Nov 19th! Read about some recent publications:

0

33

115

1

4

28

Swabha Swayamdipta

@swabhz

4 years

Looking forward to this virutal #acl2020nlp . I'll help present a couple of papers and co-lead one of the group mentoring sessions. You can find me at the following events 👇Come say 👋

1

27

Swabha Swayamdipta

@swabhz

4 years

Ahh the days when Twitter pleasantly surprises you, wasn't expecting powerful AI folks to advocate strongly for #responsibleAI Eager to see how Facebook AI research puts its 💰 where its mouth is 🤞

Jerome Pesenti

@an_open_mind

4 years

Last week I raised concerns about using #gpt3 in production because it can easily output toxic language that propagates harmful biases. I thought it was a pretty uncontroversial stance but the responses ranged from complete misunderstanding of AI to total irresponsibility. 1/13

38

314

2K

1

3

28

Swabha Swayamdipta

@swabhz

3 years

As @adveisner once told me (and I have confirmed IRL since), there is no dearth of interesting questions in #nlproc . Just think of all the fantastic problems (e.g. rare languages, low-resource settings) you could work on, powered with BERT-style models! (2/3)

2

1

28

Swabha Swayamdipta

@swabhz

6 years

Why oh why do all these voice assistants have to have female voices???? C'mon @GoogleAI you can do better than reinforce harmful stereotypes. And don't say you have the option to use a male voice too, but never, ever showcase it in worldwide #GoogleIO demos!

WIRED

@WIRED

6 years

Google Duplex—the company's AI-powered virtual assistant—can mimic the chit-chattiness of human speech so well you might not even realize you're talking to a bot

5

63

127

10

8

28

Swabha Swayamdipta

@swabhz

6 years

Our tutorial on frame-semantics and SRL gets a shoutout in The NLP Newsletter, thanks @omarsar0 Updated slides and references can be found here:

GitHub - swabhs/coling18tutorial: COLING 2018 Tutorial on Multilingual FrameNet: Automatic semantic...

COLING 2018 Tutorial on Multilingual FrameNet: Automatic semantic role labeling for FrameNet - swabhs/coling18tutorial

github.com

elvis

@omarsar0

6 years

The NLP Newsletter (Issue 27): Deep INFOMAX, Image to Image Translation, FEVER, Perception Engines, QuAC, Best 150 ML Tutorials,… Feat: @FloydHub_ , @dair_ai , @volkuleshov , @afshinea , @esquire_gh , @swabhz #NLProc

2

5

14

2

5

27

Swabha Swayamdipta

@swabhz

2 years

So honored to have received this opportunity to present my work at the Center for AI in Society at USC @CAIS_USC - please check out our work on social biases in hate speech detection, with my collaborators from UW / AI2 and CMU! Looking forward to more deep dives into this area!

USC Center for AI in Society

@CAIS_USC

2 years

Did you miss yesterday's seminar? Not to worry! Here is the recording of Dr. Swabha Swayamdipta's seminar on inadequacies of current methods for debiasing hate speech detection. ✌️ @USCViterbi @CSatUSC @uscsocialwork @swabhz @BDilkina @phebe_vayanos

0

2

6

0

2

26

Swabha Swayamdipta

@swabhz

11 months

Excited to talk about our work on Contextual Social Bias Frames at #acl2023nlp next week 👇

Xuhui Zhou

@nlpxuhui

11 months

1/ 🧵 How toxic💀 a statement is depends on its social context 👥 Excited to introduce our new #ACL2023 paper: COBRA 🐍 Frames, where we incorporate social contexts for toxic language reasoning. (🚨 Offensive Content Warning) 🔗 🌐

3

32

145

0

2

26

Swabha Swayamdipta

@swabhz

6 years

@NAACLHLT short paper with @ssgrn @royschwartz02 @omerlevy_ @sleepinyourhat @nlpnoah that discovers artifacts in annotations of natural language inference, accepted with strong scores! Transformative, we've been told #naacl2018 #NLProc

2

4

25

Swabha Swayamdipta

@swabhz

3 years

Super proud of my intern @_jennyliang All the schools looking for PhD students, recruit this one 👇

Allen School

@uwcse

3 years

Last but certainly not least, a shout-out to CS & Informatics major @_jennyliang , who earned an honorable mention for her work with @UW_iSchool professor @amyjko on explicit programming strategies to improve collaboration and learning among developers. 5/5

1

24

0

2

24

Swabha Swayamdipta

@swabhz

7 years

Waleed and Matt's new NLP podcast We extended Stack LSTMs to do joint syn-sem parsing!

Greedy, Joint Syntactic-Semantic Parsing with Stack LSTMs

We present a transition-based parser that jointly produces syntactic and semantic dependencies. It learns a representation of the entire algorithm state, using stack long short-term memories. Our...

arxiv.org

1

10

23

Swabha Swayamdipta

@swabhz

5 years

I have the best audience!!

Adina Williams

@adinamwilliams

5 years

Exciting talk by @swabhz on syntactic inductive bias and dataset issues/collection @NYUDataScience #cilvr Nuanced discussion on data augmentation, models detecting their own bias, possibilities for subsampling/downsampling known artifacts. It's a complex issue! #NLProc #NLI

0

2

23

1

0

23

Swabha Swayamdipta

@swabhz

4 years

My friend and colleague @anmarasovic inspires me. It's been a difficult day today to see people in power discredit those they should help uplift. I wish a lot of us could focus only on being scientists, have our voices heard, and not have to worry about who we identify as...

0

23

Swabha Swayamdipta

@swabhz

4 years

@adyantalamadhya @emilymbender 👆we should try to do the best we can. The criteria for evaluating someone should not *just* be how many *CL papers they wrote, rather a more holistic assessment of them as scientists - what kinds of questions they ask, or connections they draw, or what their research vision is..

0

2

23

Swabha Swayamdipta

@swabhz

6 years

Put syntax back where it belongs #sixwordstory

0

23

Swabha Swayamdipta

@swabhz

5 years

Awww, made my day 💕💕💕

0

22

Swabha Swayamdipta

@swabhz

5 years

Topic we didn't get time to address in the #NAACLTransfer tutorial yday in "what is in a representation" is a discussion on bias in embeddings. The keynote by Aravind Narayanan addresses this fascinating topic! #NAACL2019 @seb_ruder @Thom_Wolf @mattthemathman

1

0

22

Swabha Swayamdipta

@swabhz

1 month

Had such a great experience at the OSGAI workshop hosted by @cornell and organized by @srush_nlp and @colinraffel . Please check out our latest work 👇

Sasha Rush

@srush_nlp

1 month

Swabha Swayamdipta - Towards (Closed-Source) LLM Accountability via Logit Signatures (USC) @swabhz Very creative approaches for using the logit signature of LLMs to understand their structure. Lots of excitement when this work came out last month.

1

4

17

1

21

Swabha Swayamdipta

@swabhz

3 years

Highly recommend article for those in the throws of a PhD, or contemplating one 👇 Also those visuals 😍

Maxwell Forbes

@maxforbes

3 years

I wrote a new essay where I try to figure out what "Every PhD Is Different" means:

6

38

153

1

2

21

Swabha Swayamdipta

@swabhz

2 years

THIS!! 👇🔥💯 Unfortunately not restricted to ethics research, any kind of analytical work, or interpretability study (that doesn't look at model weights, for eg) gets quickly brushed aside as "not technical"

Kristian Lum

@KLdivergence

3 years

Declaring something "not technical enough" is a way of writing off research and researchers (disproportionately women) who have been working on the ethical AI questions that aren't purely of the variety "Just change the loss function. Fairness is solved."

3

26

172

1

0

21

Swabha Swayamdipta

@swabhz

2 years

Just realized that Arrested Development is Succession for idiots

3

1

21

Swabha Swayamdipta

@swabhz

1 year

@jmhessel @yanaiela CNN: ChatGPT for NLP Nihilism

0

20

Swabha Swayamdipta

@swabhz

3 years

Learn more about our Dataset Cartography📍work at #emnlp2020 in Q&A session 16, happening in 45 mins!

Swabha Swayamdipta

@swabhz

4 years

As datasets have grown larger, data exploration has become increasingly challenging. Our new work on Dataset Cartography, at @emnlp2020 with @royschwartz02 , @NickLourie , @yizhongwyz , @HannaHajishirzi , @nlpnoah , @YejinChoinka offers a solution 🗺️ Paper: 1/n

2

60

250

0

3

20

Swabha Swayamdipta

@swabhz

3 years

Proud to be a "problem" woman of color with questionable "common sense" #IStandWithTimnit

0

20

Swabha Swayamdipta

@swabhz

4 years

So proud to have Jenny as an intern!

Allen School

@uwcse

4 years

Our latest student spotlight features #UWAllen graduate & @allen_ai scholarship winner Jenny Liang, who combines compassion with technology for social good. #uwinnovates #beboundless

0

4

21

1

0

19

Swabha Swayamdipta

@swabhz

6 years

Tutorial is on now! My section starts after 11. Come to hear what's the biggest deterrent to SRL! #coling2018 #NLProc

Swabha Swayamdipta

@swabhz

6 years

Want to know more about frame-semantics and semantic role labeling? Check out our tutorial at @coling2018 with Collin Baker, Miriam Petruck and Michael Ellsworth. Slides for Section on Frame-SRL: . Full tutorial: Tomorrow at 9 am!

1

28

93

0

1

19

Swabha Swayamdipta

@swabhz

4 years

@zacharylipton Just to clarify, the biggest questions here are about publication counts to get into a PhD program, and judging young researchers by citation counts. Not to mention the pressure of being *first* author in said publications. These are lived experiences of many, not just hyperbole.

2

0

17

Swabha Swayamdipta

@swabhz

4 years

I was first in my family to move to a different continent at 23. I dunno what or if I was thinking, and glad I wasn't

Isabel Cachola

@isabelcachola

4 years

I had cousins go from "if I go to college" to "when I go to UT" when they saw me walk the stage 👩🏽‍🎓

4

3

53

1

0

18

Swabha Swayamdipta

@swabhz

7 years

#WiML2016 So many women interested in machine learning!

1

2

18

Swabha Swayamdipta

@swabhz

4 years

Hear, hear!

Jade Abbott

@alienelf

4 years

"Sorry about my English" is an apology no one should feel necessary to give. We should normalize that English is not everyone's first language. We should downplay the importance of good grammar. It serves as a colonial gate-keeping mechanism

46

2K

3K

0

3

18

Swabha Swayamdipta

@swabhz

4 years

Channel whatever power/influence you have to uplift others, especially those who are not seen/heard otherwise. This is the hardest one, and (theoretically) comes easier the higher up you are, but you gotta try with whatever leverage you got. (4/4)

0

1

16

Swabha Swayamdipta

@swabhz

6 years

Ughh so true

0

1

17

Swabha Swayamdipta

@swabhz

5 years

Thanks for hosting me, I had many wonderful discussions!

StanfordCSLI

@StanfordCSLI

5 years

The Stanford Natural Language Processing Group meets tomorrow. @swabhz speaks on Learning Challenges in Natural Language Processing #NLProc #AI

0

5

26

2

0

17

Swabha Swayamdipta

@swabhz

2 years

Pretty amazing that the organization where my colleagues and I work on NLP also does super cool projects such as these 👇Will miss being part of @allen_ai after next month 🥲

Allen Institute for AI

@allen_ai

2 years

Locusts are putting South African pastures used to feed livestock at risk. @EarthRangerTech is helping Agri Eastern Cape fight back. Read more in the @business article by @AntonySguazzin + @vollgraaffR :

0

2

6

0

1

17

Swabha Swayamdipta

@swabhz

5 years

@gregd_nlp We regret to inform you that your costume has been rejected. The reviewers unanimously agreed that by not being sota, your paper did not meet the scary standards of Halloween, which was very competitive this year.

0

17

Swabha Swayamdipta

@swabhz

2 years

The 3rd edition of ⬇️DeepLo⬆️ is here! Consider submitting your regural workshop papers, extended abstracts, or cross submissions, on a variety of exciting topics including few shot learning and low resource learning, by April 8th ⏰ See below for more details 👇

DeepLo Workshop

@DeepLoNLP

2 years

Send us your work on low-resource NLP (from prompting and transfer learning, to representation learning and self-/semi-/un-supervised learning paradigms)! We accept long/short/non-archival papers. Deadline: Apr 8 Check out CFP for more details!

0

5

6

0

3

16

Swabha Swayamdipta

@swabhz

6 years

"Not 100% of your identity is based on your achievement or work" --- so important to keep in mind, and not said enough. Keep inspiring us, @timnitGebru !

1

2

16

Swabha Swayamdipta

@swabhz

17 days

Great job by @jaspreetranjit_ , paper out on arxiv soon!

USC Center for AI in Society

@CAIS_USC

23 days

Congratulations to the ShowCAIS Best Student Poster Award winners Jaspreet Ranjit, David Chu, Priyanka Dey, and Caroline Johnston!!! We are so proud of you 🥳✌️ Check out their abstracts: @Carol_Marge_J @jaspreetranjit_ @USCViterbi @uscsocialwork

1

2

27

0

15

Swabha Swayamdipta

@swabhz

3 years

Paper coming soon! Hope the #NLProc community stops using SAE / GAE terminology in the coming years. Excuses about comparison to prior work not welcome

2

0

15