Shani
Gershtein
Shani
Gershtein
Bio
Shani is a NLP Data Scientist at GE Healthcare. Prior to GE Healthcare, she worked as a Data Scientist at Ebay.
She holds an M.Sc. in Industrial Engineering – Business Analytics from Tel Aviv University and a B.Sc. in Industrial Engineering & Management from Ben-Gurion University of the Negev.
Additionally, Shani volunteered as a “Job Search” Program Manager at Baot – helping data scientists find their next dream job.
Bio
Shani is a NLP Data Scientist at GE Healthcare. Prior to GE Healthcare, she worked as a Data Scientist at Ebay.
She holds an M.Sc. in Industrial Engineering – Business Analytics from Tel Aviv University and a B.Sc. in Industrial Engineering & Management from Ben-Gurion University of the Negev.
Additionally, Shani volunteered as a “Job Search” Program Manager at Baot – helping data scientists find their next dream job.
Abstract
Working on a chat-bot presents three major challenges: (1) understanding the user (2) fulfilling their requests, and (3) generating an adequate response. Let’s say you have a well speaking English bot. You already have a successful solution for English, and you start thinking about your next users: those coming from Spain, Japan, perhaps Israel. You’ll need to translate utterances from each language to English, run your current English flow, and – finally – translate your bot’s English response to the user’s native language. Supervised machine translation is the obvious starting point, but getting a paired corpora is very expensive, and even more so in a specific field, such as Healthcare or Finance.
In this talk, I’ll go over the steps of creating an unsupervised MT solution model, discuss the various challenges and opportunities, and review the latest industry solutions for understanding and generating multilingual text in specific domains, enhancing your bot’s language capabilities. I will also discuss the evaluation metrics and quality estimation methods you’ll need to master it in production.
Abstract
Working on a chat-bot presents three major challenges: (1) understanding the user (2) fulfilling their requests, and (3) generating an adequate response. Let’s say you have a well speaking English bot. You already have a successful solution for English, and you start thinking about your next users: those coming from Spain, Japan, perhaps Israel. You’ll need to translate utterances from each language to English, run your current English flow, and – finally – translate your bot’s English response to the user’s native language. Supervised machine translation is the obvious starting point, but getting a paired corpora is very expensive, and even more so in a specific field, such as Healthcare or Finance.
In this talk, I’ll go over the steps of creating an unsupervised MT solution model, discuss the various challenges and opportunities, and review the latest industry solutions for understanding and generating multilingual text in specific domains, enhancing your bot’s language capabilities. I will also discuss the evaluation metrics and quality estimation methods you’ll need to master it in production.
Planned Agenda
8:45 | Reception |
---|---|
9:30 | Opening words by Shir Meir Lador, Data Science leader at Intuit |
9:45 | Yael Karov - AI For Assisting in Task Completion |
10:15 | Ofra Amir - Agent Strategy Summarization: Describing Agent Behavior to People |
10:45 | Break |
11:00 | Lightning talks |
12:30 | Lunch & Poster session |
---|---|
13:30 | Roundtable session & Poster session |
14:30 | Roundtable closure |
14:45 | Gal Yona - How Fair Can We Be |
15:15 | Daphna Weissglas - Turning Data Science Into Precision Medicine Empowering Millions |
15:45 | Closing remarks |
Planned Agenda
8:45 | Reception |
---|---|
9:30 | Opening words by Shir Meir Lador, Data Science leader at Intuit |
9:45 | Yael Karov - AI For Assisting in Task Completion |
10:15 | Ofra Amir - Agent Strategy Summarization: Describing Agent Behavior to People |
10:45 | Break |
11:00 | Lightning talks |
12:30 | Lunch & Poster session |
13:30 | Roundtable session & Poster session |
14:30 | Roundtable closure |
14:45 | Gal Yona - How Fair Can We Be |
15:15 | Daphna Weissglas - Turning Data Science Into Precision Medicine Empowering Millions |
15:45 | Closing remarks |