We study the impact of source length and verbosity of the tuning dataset on the performance of parameter optimizers such as MERT and PRO for statistical machine translation. In particular, we test whether the verbosity of the resulting translations can be modified by varying the length or the verbosity of the tuning sentences. We find that MERT learns the tuning set verbosity very well, while PRO is sensitive to both the verbosity and the length of the source sentences in the tuning set; yet, overall PRO learns best from high-verbosity tuning datasets. Given these dependencies, and potentially some other such as amount of reordering, number of unknown words, syntactic complexity, and evaluation measure, to mention just a few, we argue for the need of controlled evaluation scenarios, so that the selection of tuning set and optimization strategy does not overshadow scientific advances in modeling or decoding. In the mean time, until we develop such controlled scenarios, we recommend using PRO with a large verbosity tuning set, which, in our experiments, yields highest BLEU across datasets and language pairs.

@InProceedings{guzman-nakov-vogel:2015:CoNLL, author = {Guzm\'{a}n, Francisco and Nakov, Preslav and Vogel, Stephan}, title = {Analyzing Optimization for Statistical Machine Translation: MERT Learns Verbosity, PRO Learns Length}, booktitle = {Proceedings of the Nineteenth Conference on Computational Natural Language Learning}, month = {July}, year = {2015}, address = {Beijing, China}, publisher = {Association for Computational Linguistics}, pages = {62--72}, url = {http://www.aclweb.org/anthology/K15-1007}, abstract = {We study the impact of source length and verbosity of the tuning dataset on the performance of parameter optimizers such as MERT and PRO for statistical machine translation. In particular, we test whether the verbosity of the resulting translations can be modified by varying the length or the verbosity of the tuning sentences. We find that MERT learns the tuning set verbosity very well, while PRO is sensitive to both the verbosity and the length of the source sentences in the tuning set; yet, overall PRO learns best from high-verbosity tuning datasets. Given these dependencies, and potentially some other such as amount of reordering, number of unknown words, syntactic complexity, and evaluation measure, to mention just a few, we argue for the need of controlled evaluation scenarios, so that the selection of tuning set and optimization strategy does not overshadow scientific advances in modeling or decoding. In the mean time, until we develop such controlled scenarios, we recommend using PRO with a large verbosity tuning set, which, in our experiments, yields highest BLEU across datasets and language pairs.} }