The scarcity of parallel data is a major limitation for Neural Machine Translation (NMT) systems, in particular for translation into morphologically rich languages (MRLs). An important way to overcome the lack of parallel data is to leverage target monolingual data, which is typically more abundant and easier to collect. We evaluate a number of techniques to achieve this, ranging from back-translation to random token masking, on the challenging task of translating English into four typologically diverse MRLs, under low-resource settings. Additionally, we introduce Inflection Pre-Training (or PT-Inflect), a novel pre-training objective whereby the NMT system is pre-trained on the task of re-inflecting lemmatized target sentences before being trained on standard source-to-target language translation. We conduct our evaluation on four typologically diverse target MRLs, and find that PT-Inflect surpasses NMT systems trained only on parallel data. While PT-Inflect is outperformed by back-translation overall, combining the two techniques leads to gains in some of the evaluated language pairs.

We present VQA-MHUG – a novel 49-participant dataset of multimodal human gaze on both images and questions during visual question answering (VQA) collected using a high-speed eye tracker. We use our dataset to analyze the similarity between human and neural attentive strategies learned by five state-of-the-art VQA models: Modular Co-Attention Network (MCAN) with either grid or region features, Pythia, Bilinear Attention Network (BAN), and the Multimodal Factorized Bilinear Pooling Network (MFB). While prior work has focused on studying the image modality, our analyses show – for the first time – that for all models, higher correlation with human attention on text is a significant predictor of VQA performance. This finding points at a potential for improving VQA performance and, at the same time, calls for further research on neural text attention mechanisms and their integration into architectures for vision and language tasks, including but potentially also beyond VQA.

It is now established that modern neural language models can be successfully trained on multiple languages simultaneously without changes to the underlying architecture, providing an easy way to adapt a variety of NLP models to low-resource languages. But what kind of knowledge is really shared among languages within these models? Does multilingual training mostly lead to an alignment of the lexical representation spaces or does it also enable the sharing of purely grammatical knowledge? In this paper we dissect different forms of cross-lingual transfer and look for its most determining factors, using a variety of models and probing tasks. We find that exposing our LMs to a related language does not always increase grammatical knowledge in the target language, and that optimal conditions for lexical-semantic transfer may not be optimal for syntactic transfer.

Dravidian languages, such as Kannada and Tamil, are notoriously difficult to translate by state-of-the-art neural models. This stems from the fact that these languages are morphologically very rich as well as being low-resourced. In this paper, we focus on subword segmentation and evaluate Linguistically Motivated Vocabulary Reduction (LMVR) against the more commonly used SentencePiece (SP) for the task of translating from English into four different Dravidian languages. Additionally we investigate the optimal subword vocabulary size for each language. We find that SP is the overall best choice for segmentation, and that larger dictionary sizes lead to higher translation quality.

This paper describes our submission for the English-Tamil news translation task of WMT-2020. The various techniques and Neural Machine Translation (NMT) models used by our team are presented and discussed, including back-translation, fine-tuning and word dropout. Additionally, our experiments show that using a linguistically motivated subword segmentation technique (Ataman et al., 2017) does not consistently outperform the more widely used, non-linguistically motivated SentencePiece algorithm (Kudo and Richardson, 2018), despite the agglutinative nature of Tamil morphology.

We present work in progress on the temporal progression of compositionality in noun-noun compounds. Previous work has proposed computational methods for determining the compositionality of compounds. These methods try to automatically determine how transparent the meaning of the compound as a whole is with respect to the meaning of its parts. We hypothesize that such a property might change over time. We use the time-stamped Google Books corpus for our diachronic investigations, and first examine whether the vector-based semantic spaces extracted from this corpus are able to predict compositionality ratings, despite their inherent limitations. We find that using temporal information helps predicting the ratings, although correlation with the ratings is lower than reported for other corpora. Finally, we show changes in compositionality over time for a selection of compounds.

We introduce temporally and contextually-aware models for the novel task of predicting unseen but plausible concepts, as conveyed by noun-noun compounds in a time-stamped corpus. We train compositional models on observed compounds, more specifically the composed distributed representations of their constituents across a time-stamped corpus, while giving it corrupted instances (where head or modifier are replaced by a random constituent) as negative evidence. The model captures generalisations over this data and learns what combinations give rise to plausible compounds and which ones do not. After training, we query the model for the plausibility of automatically generated novel combinations and verify whether the classifications are accurate. For our best model, we find that in around 85% of the cases, the novel compounds generated are attested in previously unseen data. An additional estimated 5% are plausible despite not being attested in the recent corpus, based on judgments from independent human raters.

Recent work has shown that neural models can be successfully trained on multiple languages simultaneously. We investigate whether such models learn to share and exploit common syntactic knowledge among the languages on which they are trained. This extended abstract presents our preliminary results.