Citation Context

...9 13.60 flexible 60.34 21.87 58.05 23.86 53.13 28.57 51.98 31.30 71.17 13.84 Table 1: Case insensitive TER and BLEU scores on newssyscomb2009 (dev) and newstest2009 (test) for five source languages. (=-=Rosti et al., 2008-=-) are marked as incr-wmt08. The sentence specific alignment order yields about a half BLEU point gain on the tuning set and a one BLEU point gain on the test set. All system combination experiments yi...

Citation Context

...eaknesses. By offsetting weaknesses with strengths of other systems, combination can produce higher quality than does any component system. One approach to system combination uses confusion networks (=-=Rosti et al., 2008-=-; Karakos et al., 2008). In the most common form, a skeleton sentence is chosen from among the one-best system outputs. This skeleton determines the ordering of the final combined sentence. The remain...

Citation Context

...he combination of machine translation systems there have been two main approaches described in recent publications. One uses confusion network decoding to combine translation systems as described in (=-=Rosti et al., 2008-=-) and (Karakos et al., 2008). The other approach selects whole hypotheses from a combined n-best list (Hildebrand and Vogel, 2008). Our setup follows the approach described in (Hildebrand and Vogel, 2...

Citation Context

...nents of the system with reference to code while Section 4 shows how to run the system. Section 5 summarizes results in recent evaluations and Section 6 concludes. 2. Related Work Confusion networks (=-=Rosti et al., 2008-=-; Karakos et al., 2008) are a popular form of system combination. This approach combines k-best output from multiple systems. A single k-best list entry is selected as the backbone, which determines w...

Citation Context

... several years, the system combination methods based on confusion networks developed rapidly (Bangalore et al., 2001; Matusov et al., 2006; Sim et al., 2007; Rosti et al., 2007a; Rosti et al., 2007b; =-=Rosti et al., 2008-=-; He et al., 2008), which show state-of-the-art performance in benchmarks. A confusion network consists of a sequence of sets of candidate words. Each candidate word is associated with a score. The op...

Citation Context

...r distributions for each system (Peterson et al., 2009b). For example, some systems are weak at word order while others have more trouble with nouns and verbs. Existing system combination techniques (=-=Rosti et al., 2008-=-; Karakos et al., 2008; Leusch et al., 2009a) ignore these distinctions by learning a single weight for each system. This weight is used in word-level decisions and therefore captures only lexical cho...

Citation Context

...ent has to incorporate computationally expensive movements of word blocks. In this paper, we show how one can combine translation outputs efficiently, extending the incremental alignment procedure of =-=[1]-=-. A comparison between different system combination design choices is performed on an Arabic speech translation task. Index Terms— Machine Translation, System Combination, Confusion Networks, Alignmen...

Citation Context

...ently, confusion networkbased networks, either single (Bangalore et al., 2001; Matusov et al., 2006; Rosti et al., 2007a; Sim et al., 2007; He et al., 2008) or multiple networks (Rosti et al., 2007b; =-=Rosti et al., 2008-=-), have become the state-of-the-art methodology to implement the combination strategy. A CN is essentially a directed acyclic graph which is built by aligning a set of translation hypotheses against a...