Alignment by Agreement

By Percy Liang, B. Taskar, Daniel Klein. 2006

511 citations recorded by the source corpus.

We present an unsupervised approach to symmetric word alignment in which two simple asymmetric models are trained jointly to maximize a combination of data likelihood and agreement between the models. Compared to the standard practice of intersecting predictions of independently-trained models, joint training provides a 32% reduction in AER. Moreover, a simple and efficient pair of HMM aligners provides a 29% reduction in AER over symmetrized IBM model 4 predictions.

Subfields: Artificial Intelligence, Molecular Biology, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Molecular Biology

In Natural Language Processing

Citations in the corpus

Deep Recurrent Models with Fast-Forward Connections for Neural Machine Translation (2016, methodology): “The full parallel corpus is used to obtain the word mappings (Liang et al., 2006).”
Effective Approaches to Attention-based Neural Machine Translation (2015, methodology): “Nevertheless, as shown in Table 6, we were able to achieve AER scores comparable to the one-to-many alignments obtained by the Berkeley aligner (Liang et al., 2006).16 We also found that the alignments produced by local attention models achieve lower AERs than those of the global one.”
Findings of the 2016 Conference on Machine Translation (2016, neutral): “, 2006) and Berkeley Aligner (Liang et al., 2006).”
Findings of the 2015 Workshop on Statistical Machine Translation (2015, neutral): “…combines different kinds of word alignment: GIZA++ word alignment with the grow-diag-ﬁnal-and (GDFA) heuristic (Koehn, 2010), SymGiza++ (Junczys-Dowmunt and Szal, 2011), the Berkeley aligner (Liang et al., 2006), and the edit distance-based aligners (Snover et al., 2006a; Lavie and Agarwal, 2007).”
Addressing the Rare Word Problem in Neural Machine Translation (2014, methodology): “The alignment is computed with the Berkeley aligner [16] using its default settings.”
Modeling Coverage for Neural Machine Translation (2016, background): “(2016) proposed agreement-based learning (Liang et al., 2006) to encourage bidirectional attention models to agree on parameterized alignment matrices.”
Learning Semantic Correspondences with Less Supervision (2009, background): “Many of the remaining errors are due to the garbage collection phenomenon familiar from word alignment models (Moore, 2004; Liang et al., 2006).”
Transfer Learning for Low-Resource Neural Machine Translation (2016, methodology): “To get these probabilities we compose Uzbek–English and English–French t-tables obtained from the Berkeley Aligner (Liang et al., 2006).”
Building a Semantic Parser Overnight (2015, methodology): “To obtain lexical features, we run the Berkeley Aligner (Liang et al., 2006) on the training set and compute conditional probabilities of aligning one word type to another.”
Bilingual Word Embeddings for Phrase-Based Machine Translation (2013, methodology): “In our experiments, we use MT word alignments extracted with the Berkeley Aligner (Liang et al., 2006) 1 .”
Sentence Compression Beyond Word Deletion (2008, methodology): “The former corpus was word-aligned using the Berkeley aligner (Liang et al., 2006) initialised with a lexicon of word identity mappings, and with Bikel’s (2002) parser.”
Neural Machine Translation with Reconstruction (2016, methodology): “It has been shown that combination of two directional models outperforms each model alone (Liang et al. 2006; Cheng et al. 2016a; Cheng et al. 2016b).”
Interpretese vs. Translationese: The Uniqueness of Human Strategies in Simultaneous Interpretation (2016, methodology): “We use the Berkeley aligner (Liang et al., 2006) for word alignment, the Stanford POS tagger (Toutanova et al.”
Recurrent Neural Networks for Word Alignment Model (2014, background): “It has been proven that the limitation may be overcome by encouraging two directional models to agree by training them concurrently (Ma-tusov et al., 2004; Liang et al., 2006; Grac¸a et al., 2008; Ganchev et al., 2008).”
Consistency by Agreement in Zero-Shot Neural Machine Translation (2019, methodology): “Our method is based on the idea of agreement-based learning initially proposed for learning consistent alignments in phrase-based statistical machine translation (SMT) systems (Liang et al., 2006, 2008).”
Bilingual Word Representations with Monolingual Quality in Mind (2015, methodology): “…languages: In our work, we experiment with two variants of our models: (a) BiSkip-UnsupAlign where we utilize unsupervised alignment information learned by the Berkeley aligner (Liang et al., 2006) and (b) BiSkip-MonoAlign where we simply assume monotonic alignments between words across languages.”
Yandex School of Data Analysis approach to English-Turkish translation at WMT16 News Translation Task (2016, methodology): “For a baseline, we trained a standard phrase-based system: Berkeley Aligner (IBM Model 1 and HMM, both for 5 iterations); phrase table with up to 5 tokens per phrase, 40-best translation options per source phrase, and Good-Turing smoothing; 5-gram lowercased LM with stupid backoff and pruning of sin”
Neural Machine Translation with Supervised Attention (2016, background): “• Multiplication (MUL) MUL is particularly designed for agreement in word alignment and it has been shown to be effective (Liang et al., 2006; Cheng et al., 2016).”
Posterior Regularization for Structured Latent Variable Models (2010, background): “The idea of training jointly has also been explored by Matusov et al. (2004) and Liang et al. (2006), although their formalization is quite different.”