<div dir="ltr"><div><br></div><div>here's this week's reading for cacl:</div><br><div class="gmail_quote">---------- Forwarded message ----------<br>From: <b class="gmail_sendername">Marten van Schijndel</b> <span dir="ltr"><<a href="mailto:van-schijndel.1@buckeyemail.osu.edu">van-schijndel.1@buckeyemail.osu.edu</a>></span><br>Date: Wed, Oct 15, 2014 at 10:52 AM<br>Subject: [CaCL] glove paper<br>To: CaCL <<a href="mailto:cacl@mail.ling.ohio-state.edu">cacl@mail.ling.ohio-state.edu</a>><br><br><br>-----BEGIN PGP SIGNED MESSAGE-----<br>
Hash: SHA1<br>
<br>
GloVe<br>
<a href="http://www-nlp.stanford.edu/projects/glove/glove.pdf" target="_blank">http://www-nlp.stanford.edu/projects/glove/glove.pdf</a><br>
<br>
Abstract: Recent methods for learning vector space representations of<br>
words have succeeded in capturing fine-grained semantic and syntactic<br>
regularities using vector arithmetic, but the origin of these<br>
regularities has remained opaque. We analyze and make explicit the<br>
model properties needed for such regularities to emerge in word<br>
vectors. The result is a new global log-bilinear regression model that<br>
combines the advantages of the two major model families in the<br>
literature: global matrix factorization and local context window<br>
methods. Our model efficiently leverages statistical information by<br>
training only on the nonzero elements in a word-word co-occurrence<br>
matrix, rather than on the entire sparse matrix or on individual<br>
context windows in a large corpus. The model produces a vector space<br>
with meaningful substructure, as evidenced by its performance of 75%<br>
on a recent word analogy task. It also outperforms related models on<br>
similarity tasks and named entity recognition.<br>
-----BEGIN PGP SIGNATURE-----<br>
Version: GnuPG v1<br>
<br>
iQIcBAEBAgAGBQJUPoomAAoJEGlPBEtlZwvAsckQAK9RlUXU2F7JIjcsCMWrexHp<br>
4qRqtM/3IF5cxmwy/oQKCqrhs98k2TUtsk5QhYSK1cFi22Uc3CsuwPzxiLxX+mNi<br>
/4aM4ETCLvp/qNjqx/WgG3qZ4zJsyOCSVIC7jh2fptG/Oc2i/n/SJKbcsQ3jG5ap<br>
3qdRzlaAA1cUrN6f1+v7flTMvS7DG4N1rbDpqbehy7MP2rvKowMT4fwR0jSSbFWO<br>
Xz0FX6egyT+Yklt8jdtPW74Odewy1WSAC455lkvzi5SDffBcneH0vx9V+mO2ct2a<br>
dRx4jdLJoGQ2Q903QkdpXjR8+okbjYHYALEKWRsjhRzfCtOXTKlUo7segYpp267u<br>
mxVUO4XacW9oo2yO8yQBHYrOl5M8wlvVBOhwy24gBc/EZrw7LtjnyL3GVNHtY+aB<br>
1WrKUC96qqLHIhuYFPy05/5K3+Cr/7AXE0ENP+JphbQr35iDVGBmOLEXNlbBx+6b<br>
zSSs3+/4xR88vhZVgQfTQopDJLJqYK1xRqNdRu12/Pd2prpHCv5TQ0Yyt7vELigi<br>
NZOLWVNtm7BmibTgFJ1UHUcN0nx+UntLNNgWv8RsRXwUMXSZX30nNPYHPKpfAdHW<br>
2888b2rp21Ciu8aWjs9dgq+k5FrsPQcSlYzGAmgML6Z/U9Y4mTeCjTCUZPeIoFHH<br>
YyeydxDbbOhos0dlFXaj<br>
=FnSx<br>
-----END PGP SIGNATURE-----<br>
<br>
_______________________________________________<br>
CaCL mailing list<br>
<a href="mailto:CaCL@mail.ling.ohio-state.edu">CaCL@mail.ling.ohio-state.edu</a><br>
<a href="http://mail.ling.ohio-state.edu/cgi-bin/mailman/listinfo/cacl" target="_blank">http://mail.ling.ohio-state.edu/cgi-bin/mailman/listinfo/cacl</a><br>
</div><br></div>