29.1553, Rising Stars: Meet Jennifer Hu

Tue Apr 10 19:27:19 UTC 2018

LINGUIST List: Vol-29-1553. Tue Apr 10 2018. ISSN: 1069 - 4875.

Subject: 29.1553, Rising Stars: Meet Jennifer Hu

Moderators: linguist at linguistlist.org (Damir Cavar, Malgorzata E. Cavar)
Reviews: reviews at linguistlist.org (Helen Aristar-Dry, Robert Coté,
                                   Michael Czerniakowski)
Homepage: http://linguistlist.org

Please support the LL editors and operation with a donation at:
           http://funddrive.linguistlist.org/donate/

Editor for this issue: Clare Harshey <clare at linguistlist.org>
================================================================

Date: Tue, 10 Apr 2018 15:27:02
From: LINGUIST List [linguist at linguistlist.org]
Subject: Rising Stars: Meet Jennifer Hu

Dear Readers,

For several years, we have featured linguists with established careers and
interesting stories to tell. This year, we will also be highlighting “Rising
Stars” throughout our Fund Drive, undergraduates who were nominated by their
mentors for their exceptional interest in linguistics and eager participation
in the global community of language researchers.

Selected nominees were asked to share their view of the field of linguistics:
what topics they see emerging as important or especially interesting, what
role they see the field filling in the coming decades, and how they plan to
contribute. We hope you will enjoy the perspectives of these students, who
represent the bright future of our field.

Today we are excited to share with you the perspective of Jennifer Hu, a
senior at Harvard University. Jennifer studies Linguistics and Mathematics,
and is highly involved in several research projects. Her own honors thesis
focuses on cross-linguistic investigation of Bayesian models of pragmatics.

******************************************************************

With the recent revolution in robotics and machine learning, linguistics is
playing an increasingly important role as we develop and interact with systems
of artificial intelligence. Just as we communicate with other humans through
language, it is most natural for us to communicate with robots and other
automated systems through speech, text, and sign. These new types of
interactions will demand a robust understanding of linguistics, as language
processing poses many unique challenges for machines.

We have already made significant progress in developing systems for speech
recognition, question answering, and other language processing tasks. If one
analyzes the errors produced by state-of-the-art systems, however, one finds
that many of these models – while obtaining high performance on the tasks for
which they are designed – are not fully capable of language understanding. For
example, the Story Cloze Test requires a system to choose the correct ending
to a simple four-sentence story as a way of approximating understanding of
causal relationships between events. The best model achieves an impressive 75%
accuracy on the Story Cloze Test, but is able to achieve 72% accuracy without
even being exposed to the stories! These results suggest that the success of
the model might not reflect genuine understanding of the events in the
stories, but other confounds latent in the task. This should lead us to
inquire whether other models have truly learned the linguistic abilities that
their tasks were designed to measure. Similarly, the type of training data
that these models require to achieve reasonable performance is cognitively
implausible, given what we know about the input to which human learners are
exposed. With very little exposure to negative data, children produce
linguistic errors in a systematic, predictable way. These two issues in the
design of current models suggest that knowledge of the theoretical
underpinnings of language can help bring us closer to building systems that
truly approximate human intelligence.

There is no better time for linguists to take advantage of and contribute to
concurrent advances in the computer and cognitive sciences. With increasing
large-scale datasets, computing power, and understanding of the human brain,
linguists have more tools than ever to pursue the scientific study of
language. In the coming years, I expect and hope to see growth in the
subfields of computational linguistics and psycholinguistics. I am excited by
the prospect of being able to reverse engineer our capacity for language, and
through collaboration with computer science and cognitive science, I believe
we can achieve this goal in the coming decades.

By studying linguistics, we can not only develop new insights into the
structure of language, but also shape the way humans will interact with
systems of artificial intelligence in the years to come. I plan to continue
contributing to this exciting field by obtaining a PhD and ultimately pursuing
a career focused on research, education, and outreach.

******************************************************************

If you have a student who you believe is a “Rising Star” in linguistics, we
would love to hear about them! We are still accepting nominations for
exceptional young linguists. Please see the call for nominations for more
information: https://linguistlist.org/issues/29/29-831.html

If you have not yet–please visit our Fund Drive page to learn more about us
and why we need your help! The LINGUIST List relies on your generous donations
to continue its support of linguists around the world.

https://funddrive.linguistlist.org/

Gratefully,
The LINGUIST List Team

------------------------------------------------------------------------------

*****************    LINGUIST List Support    *****************
Please support the LL editors and operation with a donation at:
            http://funddrive.linguistlist.org/donate/

----------------------------------------------------------
LINGUIST List: Vol-29-1553	
----------------------------------------------------------