In the research paper Paraphrasing for Style, the authors Wei Xu, Alan Ritter, William B. Dolan, Ralph Grisham and Colin Cherry investigate into the task of paraphrasing language while targeting a particular writing style. For this they use the plays of William Shakespeare and their modern translations as a testbed for evaluating paraphrase systems targeting a particular writing style. The authors further present three new automatic evaluation metrics whose primary goal is to measure the degree to which automatic paraphrases match the target style. Some of the key observations of these experiments are :
1) In the context of paraphrasing language targeting only a particular writing style, ...view middle of the document...
Thus, depending excessively on the availability of parallel corpora diminishes the scope of the problem significantly and makes it a domain-specific application. Hence, a more generic model is needed to address the problem of paraphrasing in a better way.
The paper says that the automatic metrics proposed correspond to a specific writing style and can be generalized across various writing styles. What if these metrics are also domain dependent ? There are two ways to handle such a case -
1) Create new metrics for each and every domain which is not very practical.
2) Fallback to the existing metrics BLEU and PINC.
Since the paper is only considering certain plays written by Shakespeare, it might come across unseen data while translating it to modern English. The paper doesn't explicitly mention the way it is handling this previously unseen data. Appropriate smoothing algorithms must be employed.
Phrase based machine translation is generally used in cases where there is a fundamental semantic alignment between a pair of sentences. In our case, on a comprehensive level, this is not the case. Exploring other algorithms can prove beneficial.
The paper is majorly concerned with writing style correlation and leaves...