With the success of neural language models (LMs), their language acquisition has gained much attention. This work sheds light on the second language (L2) acquisition of LMs, while previous work has typically explored their first language (L1) acquisition. Specifically, we trained bilingual LMs with a scenario similar to human L2 acquisition and analyzed their cross-lingual transfer from linguistic perspectives. Our exploratory experiments demonstrated that the L1 pretraining accelerated their linguistic generalization in L2, and language transfer configurations (e.g., the L1 choice, and presence of parallel texts) substantially affected their generalizations. These clarify their (non-)human-like L2 acquisition in particular aspects.

In various natural language processing tasks, such as named entity recognition and machine translation, example-based approaches have been used to improve performance by leveraging existing knowledge. However, the effectiveness of this approach for Grammatical Error Correction (GEC) is unclear. In this work, we explore how an example-based approach affects the accuracy and interpretability of the output of GEC systems and the trade-offs involved. The approach we investigate has shown great promise in machine translation by using the $k$-nearest translation examples to improve the results of a pretrained Transformer model. We find that using this technique increases precision by reducing the number of false positives, but recall suffers as the model becomes more conservative overall. Increasing the number of example sentences in the datastore does lead to better performing systems, but with diminishing returns and a high decoding cost. Synthetic data can be used as examples, but the effectiveness varies depending on the base model. Finally, we find that finetuning on a set of data may be more effective than using that data during decoding as examples.

Lexical complexity prediction (LCP) is the task of predicting the complexity of words in a text on a continuous scale. It plays a vital role in simplifying or annotating complex words to assist readers. To study lexical complexity in Japanese, we construct the first Japanese LCP dataset. Our dataset provides separate complexity scores for Chinese/Korean annotators and others to address the readers’ L1-specific needs. In the baseline experiment, we demonstrate the effectiveness of a BERT-based system for Japanese LCP.

This paper presents our approach to the BEA 2023 shared task of generating teacher responses in educational dialogues, using the Teacher-Student Chatroom Corpus. Our system prompts GPT-3.5-turbo to generate initial suggestions, which are then subjected to reranking. We explore multiple strategies for candidate generation, including prompting for multiple candidates and employing iterative few-shot prompts with negative examples. We aggregate all candidate responses and rerank them based on DialogRPT scores. To handle consecutive turns in the dialogue data, we divide the task of generating teacher utterances into two components: teacher replies to the student and teacher continuations of previously sent messages. Through our proposed methodology, our system achieved the top score on both automated metrics and human evaluation, surpassing the reference human teachers on the latter.

k-nearest-neighbor machine translation (kNN-MT) (Khandelwal et al., 2021) boosts the translation performance of trained neural machine translation (NMT) models by incorporating example-search into the decoding algorithm. However, decoding is seriously time-consuming, i.e., roughly 100 to 1,000 times slower than standard NMT, because neighbor tokens are retrieved from all target tokens of parallel data in each timestep. In this paper, we propose “Subset kNN-MT”, which improves the decoding speed of kNN-MT by two methods: (1) retrieving neighbor target tokens from a subset that is the set of neighbor sentences of the input sentence, not from all sentences, and (2) efficient distance computation technique that is suitable for subset neighbor search using a look-up table. Our proposed method achieved a speed-up of up to 132.2 times and an improvement in BLEU score of up to 1.6 compared with kNN-MT in the WMT’19 De-En translation task and the domain adaptation tasks in De-En and En-Ja.

In this paper, we propose a table and image generation task to verify how the knowledge about entities acquired from natural language is retained in Vision & Language (V & L) models. This task consists of two parts: the first is to generate a table containing knowledge about an entity and its related image, and the second is to generate an image from an entity with a caption and a table containing related knowledge of the entity. In both tasks, the model must know the entities used to perform the generation properly. We created the Wikipedia Table and Image Generation (WikiTIG) dataset from about 200,000 infoboxes in English Wikipedia articles to perform the proposed tasks. We evaluated the performance on the tasks with respect to the above research question using the V & L model OFA, which has achieved state-of-the-art results in multiple tasks. Experimental results show that OFA forgets part of its entity knowledge by pre-training as a complement to improve the performance of image related tasks.

A Knowledge Graph (KG) is the directed graphical representation of entities and relations in the real world. KG can be applied in diverse Natural Language Processing (NLP) tasks where knowledge is required. The need to scale up and complete KG automatically yields Knowledge Graph Embedding (KGE), a shallow machine learning model that is suffering from memory and training time consumption issues. To mitigate the computational load, we propose a parameter-sharing method, i.e., using conjugate parameters for complex numbers employed in KGE models. Our method improves memory efficiency by 2x in relation embedding while achieving comparable performance to the state-of-the-art non-conjugate models, with faster, or at least comparable, training time. We demonstrated the generalizability of our method on two best-performing KGE models 5^★E (CITATION) and ComplEx (CITATION) on five benchmark datasets.

Zero-shot relation extraction (ZSRE) aims to predict target relations that cannot be observed during training. While most previous studies have focused on fully supervised relation extraction and achieved considerably high performance, less effort has been made towards ZSRE. This study proposes a new model incorporating discriminative embedding learning for both sentences and semantic relations. In addition, a self-adaptive comparator network is used to judge whether the relationship between a sentence and a relation is consistent. Experimental results on two benchmark datasets showed that the proposed method significantly outperforms the state-of-the-art methods.

Probing is popular to analyze whether linguistic information can be captured by a well-trained deep neural model, but it is hard to answer how the change of the encoded linguistic information will affect task performance. To this end, we study the dynamic relationship between the encoded linguistic information and task performance from the viewpoint of Pareto Optimality. Its key idea is to obtain a set of models which are Pareto-optimal in terms of both objectives. From this viewpoint, we propose a method to optimize the Pareto-optimal models by formalizing it as a multi-objective optimization problem. We conduct experiments on two popular NLP tasks, i.e., machine translation and language modeling, and investigate the relationship between several kinds of linguistic information and task performances. Experimental results demonstrate that the proposed method is better than a baseline method. Our empirical findings suggest that some syntactic information is helpful for NLP tasks whereas encoding more syntactic information does not necessarily lead to better performance, because the model architecture is also an important factor.

Deep learning has demonstrated performance advantages in a wide range of natural language processing tasks, including neural machine translation (NMT). Transformer NMT models are typically strengthened by deeper encoder layers, but deepening their decoder layers usually results in failure. In this paper, we first identify the cause of the failure of the deep decoder in the Transformer model. Inspired by this discovery, we then propose approaches to improving it, with respect to model structure and model training, to make the deep decoder practical in NMT. Specifically, with respect to model structure, we propose a cross-attention drop mechanism to allow the decoder layers to perform their own different roles, to reduce the difficulty of deep-decoder learning. For model training, we propose a collapse reducing training approach to improve the stability and effectiveness of deep-decoder training. We experimentally evaluated our proposed Transformer NMT model structure modification and novel training methods on several popular machine translation benchmarks. The results showed that deepening the NMT model by increasing the number of decoder layers successfully prevented the deepened decoder from degrading to an unconditional language model. In contrast to prior work on deepening an NMT model on the encoder, our method can deepen the model on both the encoder and decoder at the same time, resulting in a deeper model and improved performance.

N-gram language models (LM) has been largely superseded by neural LMs as the latter exhibits better performance. However, we find that n-gram models can achieve satisfactory performance on a large proportion of testing cases, indicating they have already captured abundant knowledge of the language with relatively low computational cost. With this observation, we propose to learn a neural LM that fits the residual between an n-gram LM and the real-data distribution. The combination of n-gram LMs and neural LMs not only allows the neural part to focus on deeper understanding of the language, but also provides a flexible way to customize a LM by switching the underlying n-gram model without changing the neural model. Experimental results on three typical language tasks (i.e., language modeling, machine translation, and summarization) demonstrate that our approach attains additional performance gains over popular standalone neural models consistently. We also show that our approach allows for effective domain adaptation by simply switching to a domain-specific n-gram model, without any extra training.

The user-dependency of Text Simplification makes its evaluation obscure. A targeted evaluation dataset clarifies the purpose of simplification, though its specification is hard to define. We built JADES (JApanese Dataset for the Evaluation of Simplification), a text simplification dataset targeted at non-native Japanese speakers, according to public vocabulary and grammar profiles. JADES comprises 3,907 complex-simple sentence pairs annotated by an expert. Analysis of JADES shows that wide and multiple rewriting operations were applied through simplification. Furthermore, we analyzed outputs on JADES from several benchmark systems and automatic and manual scores of them. Results of these analyses highlight differences between English and Japanese in operations and evaluations.

In this paper, we describe our NAIST-NICT-TIT submission to the WMT22 general machine translation task. We participated in this task for the English ↔ Japanese language pair. Our system is characterized as an ensemble of Transformer big models, k-nearest-neighbor machine translation (kNN-MT) (Khandelwal et al., 2021), and reranking.In our translation system, we construct the datastore for kNN-MT from back-translated monolingual data and integrate kNN-MT into the ensemble model. We designed a reranking system to select a translation from the n-best translation candidates generated by the translation system. We also use a context-aware model to improve the document-level consistency of the translation.

Recent years have witnessed rapid advancements in machine translation, but the state-of-the-art machine translation system still can not satisfy the high requirements in some rigorous translation scenarios. Computer-aided translation (CAT) provides a promising solution to yield a high-quality translation with a guarantee. Unfortunately, due to the lack of popular benchmarks, the research on CAT is not well developed compared with machine translation. In this year, we hold a new shared task called Word-level AutoCompletion (WLAC) for CAT in WMT. Specifically, we introduce some resources to train a WLAC model, and particularly we collect data from CAT systems as a part of test data for this shared task. In addition, we employ both automatic and human evaluations to measure the performance of the submitted systems, and our final evaluation results reveal some findings for the WLAC task.

This paper introduces a new Universal Dependencies treebank for the Tatar language named NMCTT. A significant feature of the corpus is that it includes code-switching (CS) information at a morpheme level, given the fact that Tatar texts contain intra-word CS between Tatar and Russian. We first outline NMCTT with a focus on differences from other treebanks of Turkic languages. Then, to evaluate the merit of the CS annotation, this study concisely reports the results of a language identification task implemented with Conditional Random Fields that considers POS tag information, which is readily available in treebanks in the CoNLL-U format. Experimenting on NMCTT and the Turkish-German CS treebank (SAGT), we demonstrate that the proposed annotation scheme introduced in NMCTT can improve the performance of the subword-level language identification. This annotation scheme for CS is not only universally applicable to languages with CS, but also shows a possibility to employ morphosyntactic information for CS-related downstream tasks.

Multilingual neural machine translation can translate unseen language pairs during training, i.e. zero-shot translation. However, the zero-shot translation is always unstable. Although prior works attributed the instability to the domination of central language, e.g. English, we supplement this viewpoint with the strict dependence of non-centered languages. In this work, we propose a simple, lightweight yet effective language-specific modeling method by adapting to non-centered languages and combining the shared information and the language-specific information to counteract the instability of zero-shot translation. Experiments with Transformer on IWSLT17, Europarl, TED talks, and OPUS-100 datasets show that our method not only performs better than strong baselines in centered data conditions but also can easily fit non-centered data conditions. By further investigating the layer attribution, we show that our proposed method can disentangle the coupled representation in the correct direction.

We introduce a Cyrillic-to-Latin transliterator for the Tatar language based on subword-level language identification. The transliteration is a challenging task due to the following two reasons. First, because modern Tatar texts often contain intra-word code-switching to Russian, a different transliteration set of rules needs to be applied to each morpheme depending on the language, which necessitates morpheme-level language identification. Second, the fact that Tatar is a low-resource language, with most of the texts in Cyrillic, makes it difficult to prepare a sufficient dataset. Given this situation, we proposed a transliteration method based on subword-level language identification. We trained a language classifier with monolingual Tatar and Russian texts, and applied different transliteration rules in accord with the identified language. The results demonstrate that our proposed method outscores other Tatar transliteration tools, and imply that it correctly transcribes Russian loanwords to some extent.

This paper presents a novel method for nested named entity recognition. As a layered method, our method extends the prior second-best path recognition method by explicitly excluding the influence of the best path. Our method maintains a set of hidden states at each time step and selectively leverages them to build a different potential function for recognition at each level. In addition, we demonstrate that recognizing innermost entities first results in better performance than the conventional outermost entities first scheme. We provide extensive experimental results on ACE2004, ACE2005, and GENIA datasets to show the effectiveness and efficiency of our proposed method.

It is reported that grammatical information is useful for machine translation (MT) task. However, the annotation of grammatical information requires the highly human resources. Furthermore, it is not trivial to adapt grammatical information to MT since grammatical annotation usually adapts tokenization standards which might not be suitable to capture the relation of two languages, and the use of sub-word tokenization, e.g., Byte-Pair-Encoding, to alleviate out-of-vocabulary problem might not be compatible with those annotations. In this work, we propose two methods to explicitly incorporate grammatical information without supervising annotation; first, latent phrase structure is induced in an unsupervised fashion from a multi-head attention mechanism; second, the induced phrase structures in encoder and decoder are synchronized so that they are compatible with each other using constraints during training. We demonstrate that our approach produces better performance and explainability in two tasks, translation and alignment tasks without extra resources. Although we could not obtain the high quality phrase structure in constituency parsing when evaluated monolingually, we find that the induced phrase structures enhance the explainability of translation through the synchronization constraint.

The presence of zero-pronoun (ZP) greatly affects the downstream tasks of NLP in pro-drop languages such as Japanese and Chinese. To tackle the problem, the previous works identified ZPs as sequence labeling on the word sequence or the linearlized tree nodes of the input. We propose a novel approach to ZP identification by casting it as a query-based argument span prediction task. Given a predicate as a query, our model predicts the omission with ZP. In the experiments, our model surpassed the sequence labeling baseline.

Unsupervised image captioning is a challenging task that aims at generating captions without the supervision of image-sentence pairs, but only with images and sentences drawn from different sources and object labels detected from the images. In previous work, pseudo-captions, i.e., sentences that contain the detected object labels, were assigned to a given image. The focus of the previous work was on the alignment of input images and pseudo-captions at the sentence level. However, pseudo-captions contain many words that are irrelevant to a given image. In this work, we investigate the effect of removing mismatched words from image-sentence alignment to determine how they make this task difficult. We propose a simple gating mechanism that is trained to align image features with only the most reliable words in pseudo-captions: the detected object labels. The experimental results show that our proposed method outperforms the previous methods without introducing complex sentence-level learning objectives. Combined with the sentence-level alignment method of previous work, our method further improves its performance. These results confirm the importance of careful alignment in word-level details.

Morphological analysis (MA) and lexical normalization (LN) are both important tasks for Japanese user-generated text (UGT). To evaluate and compare different MA/LN systems, we have constructed a publicly available Japanese UGT corpus. Our corpus comprises 929 sentences annotated with morphological and normalization information, along with category information we classified for frequent UGT-specific phenomena. Experiments on the corpus demonstrated the low performance of existing MA/LN methods for non-general words and non-standard forms, indicating that the corpus would be a challenging benchmark for further research on UGT.

Abstract Meaning Representation (AMR) is a sentence-level meaning representation based on predicate argument structure. One of the challenges we find in AMR parsing is to capture the structure of complex sentences which expresses the relation between predicates. Knowing the core part of the sentence structure in advance may be beneficial in such a task. In this paper, we present a list of dependency patterns for English complex sentence constructions designed for AMR parsing. With a dedicated pattern matcher, all occurrences of complex sentence constructions are retrieved from an input sentence. While some of the subordinators have semantic ambiguities, we deal with this problem through training classification models on data derived from AMR and Wikipedia corpus, establishing a new baseline for future works. The developed complex sentence patterns and the corresponding AMR descriptions will be made public.

Lexical normalization, in addition to word segmentation and part-of-speech tagging, is a fundamental task for Japanese user-generated text processing. In this paper, we propose a text editing model to solve the three task jointly and methods of pseudo-labeled data generation to overcome the problem of data deficiency. Our experiments showed that the proposed model achieved better normalization performance when trained on more diverse pseudo-labeled data.

We propose a simple method for nominal coordination boundary identification. As the main strength of our method, it can identify the coordination boundaries without training on labeled data, and can be applied even if coordination structure annotations are not available. Our system employs pre-trained word embeddings to measure the similarities of words and detects the span of coordination, assuming that conjuncts share syntactic and semantic similarities. We demonstrate that our method yields good results in identifying coordinated noun phrases in the GENIA corpus and is comparable to a recent supervised method for the case when the coordinator conjoins simple noun phrases.

Measuring domain relevance of data and identifying or selecting well-fit domain data for machine translation (MT) is a well-studied topic, but denoising is not yet. Denoising is concerned with a different type of data quality and tries to reduce the negative impact of data noise on MT training, in particular, neural MT (NMT) training. This paper generalizes methods for measuring and selecting data for domain MT and applies them to denoising NMT training. The proposed approach uses trusted data and a denoising curriculum realized by online data selection. Intrinsic and extrinsic evaluations of the approach show its significant effectiveness for NMT to train on data with severe noise.

In this paper, we propose a new decoding method for phrase-based statistical machine translation which directly uses multiple preordering candidates as a graph structure. Compared with previous phrase-based decoding methods, our method is based on a simple left-to-right dynamic programming in which no decoding-time reordering is performed. As a result, its runtime is very fast and implementing the algorithm becomes easy. Our system does not depend on specific preordering methods as long as they output multiple preordering candidates, and it is trivial to employ existing preordering methods into our system. In our experiments for translating diverse 11 languages into English, the proposed method outperforms conventional phrase-based decoder in terms of translation qualities under comparable or faster decoding time.

This paper describes NICT’s participation in the IWSLT 2014 evaluation campaign for the TED Chinese-English translation shared-task. Our approach used a combination of phrase-based and hierarchical statistical machine translation (SMT) systems. Our focus was in several areas, specifically system combination, word alignment, and various language modeling techniques including the use of neural network joint models. Our experiments on the test set from the 2013 shared task, showed that an improvement in BLEU score can be gained in translation performance through all of these techniques, with the largest improvements coming from using large data sizes to train the language model.

This paper describes NICT’s participation in the IWSLT 2010 evaluation campaign for the DIALOG translation (Chinese-English) and the BTEC (French-English) translation shared-tasks. For the DIALOG translation, the main challenge to this task is applying context information during translation. Context information can be used to decide on word choice and also to replace missing information during translation. We applied discriminative reranking using contextual information as additional features. In order to provide more choices for re-ranking, we generated n-best lists from multiple phrase-based statistical machine translation systems that varied in the type of Chinese word segmentation schemes used. We also built a model that merged the phrase tables generated by the different segmentation schemes. Furthermore, we used a lattice-based system combination model to combine the output from different systems. A combination of all of these systems was used to produce the n-best lists for re-ranking. For the BTEC task, a general approach that used latticebased system combination of two systems, a standard phrasebased system and a hierarchical phrase-based system, was taken. We also tried to process some unknown words by replacing them with the same words but different inflections that are known to the system.

Minimum error rate training (MERT) is a widely used learning method for statistical machine translation. In this paper, we present a SVM-based training method to enhance generalization ability. We extend MERT optimization by maximizing the margin between the reference and incorrect translations under the L2-norm prior to avoid overfitting problem. Translation accuracy obtained by our proposed methods is more stable in various conditions than that obtained by MERT. Our experimental results on the French-English WMT08 shared task show that degrade of our proposed methods is smaller than that of MERT in case of small training data or out-of-domain test data.

The NTT Statistical Machine Translation System consists of two primary components: a statistical machine translation decoder and a reranker. The decoder generates k-best translation canditates using a hierarchical phrase-based translation based on synchronous context-free grammar. The decoder employs a linear feature combination among several real-valued scores on translation and language models. The reranker reorders the k-best translation candidates using Ranking SVMs with a large number of sparse features. This paper describes the two components and presents the results for the evaluation campaign of IWSLT 2008.

The NTT Statistical Machine Translation System employs a large number of feature functions. First, k-best translation candidates are generated by an efficient decoding method of hierarchical phrase-based translation. Second, the k-best translations are reranked. In both steps, sparse binary features — of the order of millions — are integrated during the search. This paper gives the details of the two steps and shows the results for the Evaluation campaign of the International Workshop on Spoken Language Translation (IWSLT) 2007.

This paper presents a decoder for statistical machine translation that can take advantage of the example-based machine translation framework. The decoder presented here is based on the greedy approach to the decoding problem, but the search is initiated from a similar translation extracted from a bilingual corpus. The experiments on multilingual translations showed that the proposed method was far superior to a word-by-word generation beam search algorithm.