language model and cyk

考虑划分短语的问题 是否可以通过使用语言模型的信息，然后通过进行cyk parse 来寻找短语呢？ xiangyu 做过这方面的工作，不过他使用的是另外的统计信息。今天本来是想给 自己的模型加语言模型方面的特征来，忽然想到了这个问题。 如果能够parse 的比较好的话 双语同时做parse，那么是否可以获取对齐信息呢 这样做的一个很大优点就是速度快。parse的速度是O(n^3) 而ITG这种同步句法分析则需要O(n^6) 速度真是个问题啊。

interested researchers

Michel Galley Microsoft Research Redmond & ttic
ttic
IvorTsang
nus nlp
scohen

Favourite Researchers

mcollins
jason
nasmith
gildea

不错的一段话

大哉问。学习需要明师。但明师可遇不可求，所以退而求其次你需要好书，并尽早建立自修的基础。迷时师渡，悟了自渡，寻好书看好书，就是你的自渡法门。切记，徒学不足以自行，计算机是实作性很强的一门科技，你一定要动手做，最忌讳眼高手低。学而不思则罔，思而不学则殆，一定要思考、沉淀、整理。--侯捷

我心目中的研究层次

提出一个好的问题
解决一个未曾解决过的好问题
解决一个未曾解决过的一般问题
改善一个别人解决过的问题
重现别人的工作

这些都有意义 ，意义的大小自上而下。对于我等新手最好是从下而上开始，不过，要记住，这不是目的，只是为了做上层的研究。

CCG grammar

I have never heard this grammar before.
In today's discussion meeting , I know this.
I will take a look at it in the future.

phrase alignment using posterior regularization

EM algorithm for phrase alignment training usually leads to degeneracy. How to design a posterior constraints to solve this problem ? If I can design efficient constraints , I can use the PR framework to solve this problem.

Posterior vs. Parameter Sparsity in Latent Variable Models

The author study the POS problem using the posterior regularization framework. what's the result for the grammar
induction problem ?  I will study this problem if I have time .

interested in semantic parsing

One of the ultimate goals in the field of natural language processing (NLP), as
an important subfield of artificial intelligence, is to enable computers to converse
with humans through the media of human natural language. To achieve this goal,
two important issues need to be studied. First, it is important for computers to
accurately capture the complete underlying semantics of natural language sen-
tences. Second, the computers should be able to accurately produce meaningful
human-understandable natural language sentences from certain given semantics.
These two tasks are usually referred to as semantic parsing and natural language
generation respectively in the NLP field.

加油！

加油！！