Managing a comprehensive data products portfolio with scaling capabilities is one important task for an organization-wide analytics team. Those data products can be broken down into components that use serverless offerings from cloud service providers, allowing a team of modest size to manage company-wide analytics and data science solutions while improving productivity and promoting data-driven decisions. This work describes an architecture and tools used to speed up and manage data offerings, including data visualization, pipelines, models, and APIs. Considerations of component design include re-usability, integration, and maintainability, which are discussed along with their impact to team productivity. The components described are: Data ingestion using a containerized solution as a fundamental layer for all the applications, including its execution, orchestration, and monitoring. This solution is combined with traditional pipelines in order to enrich the data available; APIs for data and model serving using containerized solutions as a building block for data products that are powered by machine learning models, and for serving a unified data ontology; Data Visualization in the form of containerized web apps that provides fast solutions for data explorations, model predictions, visualization, and user insights. The architectures of three data products are then described as aggregations of the distinct building blocks (components) that were developed and how those can be repurposed for different applications. This includes continuous integration and delivery as well as pairing of each solution with the corresponding products from cloud service providers (e.g. Google Cloud Platform) in order to provide some real world examples. The scaling of each solution is discussed, as well as lessons learned and pitfalls we have encountered regarding security, usability, and maintenance. The proposed components and architecture allowed a team of 7 members to cater for analytics solutions for a section (equivalent to approx. 1500 employees), which provides a clear picture of the potential of serverless in rapid prototyping and empowering effective teams.

We address a challenging and practical task of labeling questions in speech in real time during telephone calls to emergency medical services in English, which embeds within a broader decision support system for emergency call-takers. We propose a novel multimodal approach to real-time sequence labeling in speech. Our model treats speech and its own textual representation as two separate modalities or views, as it jointly learns from streamed audio and its noisy transcription into text via automatic speech recognition. Our results show significant gains of jointly learning from the two modalities when compared to text or audio only, under adverse noise and limited volume of training data. The results generalize to medical symptoms detection where we observe a similar pattern of improvements with multimodal learning.

The two most common paradigms for end-to-end speech recognition are connectionist temporal classification (CTC) and attention-based encoder-decoder (AED) models. It has been argued that the latter is better suited for learning an implicit language model. We test this hypothesis by measuring temporal context sensitivity and evaluate how the models perform when we constrain the amount of contextual information in the audio input. We find that the AED model is indeed more context sensitive, but that the gap can be closed by adding self-attention to the CTC model. Furthermore, the two models perform similarly when contextual information is constrained. Finally, in contrast to previous research, our results show that the CTC model is highly competitive on WSJ and LibriSpeech without the help of an external language model.

Current methods of cross-lingual parser transfer focus on predicting the best parser for a low-resource target language globally, that is, "at treebank level". In this work, we propose and argue for a novel cross-lingual transfer paradigm: instance-level parser selection (ILPS), and present a proof-of-concept study focused on instance-level selection in the framework of delexicalized parser transfer. We start from an empirical observation that different source parsers are the best choice for different Universal POS sequences in the target language. We then propose to predict the best parser at the instance level. To this end, we train a supervised regression model, based on the Transformer architecture, to predict parser accuracies for individual POS-sequences. We compare ILPS against two strong single-best parser selection baselines (SBPS): (1) a model that compares POS n-gram distributions between the source and target languages (KL) and (2) a model that selects the source based on the similarity between manually created language vectors encoding syntactic properties of languages (L2V). The results from our extensive evaluation, coupling 42 source parsers and 20 diverse low-resource test languages, show that ILPS outperforms KL and L2V on 13/20 and 14/20 test languages, respectively. Further, we show that by predicting the best parser "at the treebank level" (SBPS), using the aggregation of predictions from our instance-level model, we outperform the same baselines on 17/20 and 16/20 test languages.

Viable cross-lingual transfer critically depends on the availability of parallel texts. Shortage of such resources imposes a development and evaluation bottleneck in multilingual processing. We introduce JW300, a parallel corpus of over 300 languages with around 100 thousand parallel sentences per language pair on average. In this paper, we present the resource and showcase its utility in experiments with cross-lingual word embedding induction and multi-source part-of-speech projection.

In natural language processing, the deep learning revolution has shifted the focus from conventional hand-crafted symbolic representations to dense inputs, which are adequate representations learned automatically from corpora. However, particularly when working with low-resource languages, small amounts of symbolic lexical resources such as user-generated lexicons are often available even when gold-standard corpora are not. Such additional linguistic information is though often neglected, and recent neural approaches to cross-lingual tagging typically rely only on word and subword embeddings. While these representations are effective, our recent work has shown clear benefits of combining the best of both worlds: integrating conventional lexical information improves neural cross-lingual part-of-speech (PoS) tagging. However, little is known on how complementary such additional information is, and to what extent improvements depend on the coverage and quality of these external resources. This paper seeks to fill this gap by providing the first thorough analysis on the contributions of lexical resources for cross-lingual PoS tagging in neural times.

Alonso Vasquez, Renzo Ego Aguirre, Candy Angulo, John Miller, Claudia Villanueva, Željko Agić, Roberto Zariquiey, Arturo Oncevay. Proceedings of the Second Workshop on Universal Dependencies (UDW 2018). 2018.

In this paper we present hr500k, a Croatian reference training corpus of 500 thousand tokens, segmented at document, sentence and word level, and annotated for morphosyntax, lemmas, dependency syntax, named entities, and semantic roles. We present each annotation layer via basic label statistics and describe the final encoding of the resource in CoNLL and TEI formats. We also give a description of the rather turbulent history of the resource and give insights into the topic and genre distribution in the corpus. Finally, we discuss further enrichments of the corpus with additional layers, which are already underway.

In this paper we present hr500k, a Croatian reference training corpus of 500 thousand tokens, segmented at document, sentence and word level, and annotated for morphosyntax, lemmas, dependency syntax, named entities, and semantic roles. We present each annotation layer via basic label statistics and describe the final encoding of the resource in CoNLL and TEI formats. We also give a description of the rather turbulent history of the resource and give insights into the topic and genre distribution in the corpus. Finally, we discuss further enrichments of the corpus with additional layers, which are already underway.

Projecting linguistic annotations through word alignments is one of the most prevalent approaches to cross-lingual transfer learning. Conventional wisdom suggests that annotation projection “just works” regardless of the task at hand. We carefully consider multi-source projection for named entity recognition. Our experiment with 17 languages shows that to detect named entities in true low-resource languages, annotation projection may not be the right way to move forward. On a more positive note, we also uncover the conditions that do favor named entity projection from multiple sources. We argue these are infeasible under noisy low-resource constraints.

The recent years have seen a revival of interest in textual entailment, sparked by i) the emergence of powerful deep neural network learners for natural language processing and ii) the timely development of large-scale evaluation datasets such as SNLI. Recast as natural language inference, the problem now amounts to detecting the relation between pairs of statements: they either contradict or entail one another, or they are mutually neutral. Current research in natural language inference is effectively exclusive to English. In this paper, we propose to advance the research in SNLI-style natural language inference toward multilingual evaluation. To that end, we provide test data for four major languages: Arabic, French, Spanish, and Russian. We experiment with a set of baselines. Our systems are based on cross-lingual word embeddings and machine translation. While our best system scores an average accuracy of just over 75%, we focus largely on enabling further research in multilingual inference.

We introduce DsDs: a cross-lingual neural part-of-speech tagger that learns from disparate sources of distant supervision, and realistically scales to hundreds of low-resource languages. The model exploits annotation projection, instance selection, tag dictionaries, morphological lexicons, and distributed representations, all in a uniform framework. The approach is simple, yet surprisingly effective, resulting in a new state of the art without access to any gold annotated data.

We address the challenge of cross-lingual POS tagger evaluation in absence of manually annotated test data. We put forth and evaluate two dictionary-based metrics. On the tasks of accuracy prediction and system ranking, we reveal that these metrics are reliable enough to approximate test set-based evaluation, and at the same time lean enough to support assessment for truly low-resource languages.

We propose UDP, the first training-free parser for Universal Dependencies (UD). Our algorithm is based on PageRank and a small set of head attachment rules. It features two-step decoding to guarantee that function words are attached as leaf nodes. The parser requires no training, and it is competitive with a delexicalized transfer system. UDP offers a linguistically sound unsupervised alternative to cross-lingual parsing for UD, which can be used as a baseline for such systems. The parser has very few parameters and is distinctly robust to domain change across languages.

In dependency parsing, jackknifing taggers is indiscriminately used as a simple adaptation strategy. Here, we empirically evaluate when and how (not) to use jackknifing in parsing. On 26 languages, we reveal a preference that conflicts with, and surpasses the ubiquitous ten-folding. We show no clear benefits of tagging the training data in cross-lingual parsing.

This release contains errors in several files. Please use http://hdl.handle.net/11234/1-1983 instead.