Word alignment was once a core unsupervised learning task in natural language processing because of its essential role in training statistical machine translation (MT) models. Although unnecessary for training neural MT models, word alignment still plays an important role in interactive applications of neural machine translation, such as annotation transfer and lexicon injection. While statistical MT methods have been replaced by neural approaches with superior performance, the twenty-year-old GIZA++ toolkit remains a key component of state-of-the-art word alignment systems. Prior work on neural word alignment has only been able to outperform GIZA++ by using its output during training. We present the first end-to-end neural word alignment method that consistently outperforms GIZA++ on three data sets. Our approach repurposes a Transformer model trained for supervised translation to also serve as an unsupervised word alignment model in a manner that is tightly integrated and does not affect translation quality.

Connectionist Temporal Classification has recently attracted a lot of interest as it offers an elegant approach to building acoustic models (AMs) for speech recognition. The CTC loss function maps an input sequence of observable feature vectors to an output sequence of symbols. Output symbols are conditionally independent of each other under CTC loss, so a language model (LM) can be incorporated conveniently during decoding, retaining the traditional separation of acoustic and linguistic components in ASR.For fixed vocabularies, Weighted Finite State Transducers provide a strong baseline for efficient integration of CTC AMs with n-gram LMs. Character-based neural LMs provide a straight forward solution for open vocabulary speech recognition and all-neural models, and can be decoded with beam search. Finally, sequence-to-sequence models can be used to translate a sequence of individual sounds into a word string.We compare the performance of these three approaches, and analyze their error patterns, which provides insightful guidance for future research and development in this important area.

This paper proposes a novel approach to create a unit set for CTC-based speech recognition systems. By using Byte-Pair Encoding we learn a unit set of arbitrary size on a given training text. In contrast to using characters or words as units, this allows us to find a good trade-off between the size of our unit set and the available training data. We investigate both crossword units, which may span multiple words, and subword units. By evaluating these unit sets with decoding methods using a separate language model, we are able to show improvements over a purely character-based unit set.

Traditionally systems for term extraction use a two stage approach of first identifying candiate terms, and the scoring them in a second process for identifying actual terms. Thus, research in this field has often mainly focused on refining and improving the scoring process of term candidates, which commonly are identified using linguistic and statistical features. Machine learning techniques and especially neural networks are currently only used in the second stage, that is to score candidates and classify them. In contrast to that we have built a system that identifies terms via directly performing sequence-labeling with a BILOU scheme on word sequences. To do so we have worked with different kinds of recurrent neural networks and word embeddings. In this paper we describe how one can built a state-of-theart term extraction systems with this single-stage technique and compare different network types and topologies and also examine the influence of the type of input embedding used for the task. We further investigated which network types and topologies are best suited when applying our term extraction systems to other domains than that of the training data of the networks.

No abstract