Treebanks annotated with Universal Dependencies (UD) are currently available for over 100 languages and are widely utilized by the community. However, their inherent characteristics are hard to measure and are only partially reflected in parser evaluations via accuracy metrics like LAS. In this study, we analyze a large subset of the UD treebanks using three recently proposed accuracy-free dataset analysis methods: dataset cartography, 𝒱-information, and minimum description length. Each method provides insights about UD treebanks that would remain undetected if only LAS was considered. Specifically, we identify a number of treebanks that, despite yielding high LAS, contain very little information that is usable by a parser to surpass what can be achieved by simple heuristics. Furthermore, we make note of several treebanks that score consistently low across numerous metrics, indicating a high degree of noise or annotation inconsistency present therein.

Recent studies have shown that language models pretrained and/or fine-tuned on randomly permuted sentences exhibit competitive performance on GLUE, putting into question the importance of word order information. Somewhat counter-intuitively, some of these studies also report that position embeddings appear to be crucial for models’ good performance with shuffled text. We probe these language models for word order information and investigate what position embeddings learned from shuffled text encode, showing that these models retain a notion of word order information. We show this is in part due to a subtlety in how shuffling is implemented in previous work – before rather than after subword segmentation. Surprisingly, we find even Language models trained on text shuffled after subword segmentation retain some semblance of information about word order because of the statistical dependencies between sentence length and unigram probabilities. Finally, we show that beyond GLUE, a variety of language understanding tasks do require word order information, often to an extent that cannot be learned through fine-tuning.

In this work, we demonstrate that the contextualized word vectors derived from pretrained masked language model-based encoders share a common, perhaps undesirable pattern across layers. Namely, we find cases of persistent outlier neurons within BERT and RoBERTa’s hidden state vectors that consistently bear the smallest or largest values in said vectors. In an attempt to investigate the source of this information, we introduce a neuron-level analysis method, which reveals that the outliers are closely related to information captured by positional embeddings. We also pre-train the RoBERTa-base models from scratch and find that the outliers disappear without using positional embeddings. These outliers, we find, are the major cause of anisotropy of encoders’ raw vector spaces, and clipping them leads to increased similarity across vectors. We demonstrate this in practice by showing that clipped vectors can more accurately distinguish word senses, as well as lead to better sentence embeddings when mean pooling. In three supervised tasks, we find that clipping does not affect the performance.

Pretrained language models have been shown to encode relational information, such as the relations between entities or concepts in knowledge-bases — (Paris, Capital, France). However, simple relations of this type can often be recovered heuristically and the extent to which models implicitly reflect topological structure that is grounded in world, such as perceptual structure, is unknown. To explore this question, we conduct a thorough case study on color. Namely, we employ a dataset of monolexemic color terms and color chips represented in CIELAB, a color space with a perceptually meaningful distance metric. Using two methods of evaluating the structural alignment of colors in this space with text-derived color term representations, we find significant correspondence. Analyzing the differences in alignment across the color spectrum, we find that warmer colors are, on average, better aligned to the perceptual color space than cooler ones, suggesting an intriguing connection to findings from recent work on efficient communication in color naming. Further analysis suggests that differences in alignment are, in part, mediated by collocationality and differences in syntactic usage, posing questions as to the relationship between color perception and usage and context.

Since the popularization of the Transformer as a general-purpose feature encoder for NLP, many studies have attempted to decode linguistic structure from its novel multi-head attention mechanism. However, much of such work focused almost exclusively on English — a language with rigid word order and a lack of inflectional morphology. In this study, we present decoding experiments for multilingual BERT across 18 languages in order to test the generalizability of the claim that dependency syntax is reflected in attention patterns. We show that full trees can be decoded above baseline accuracy from single attention heads, and that individual relations are often tracked by the same heads across languages. Furthermore, in an attempt to address recent debates about the status of attention as an explanatory mechanism, we experiment with fine-tuning mBERT on a supervised parsing objective while freezing different series of parameters. Interestingly, in steering the objective to learn explicit linguistic structure, we find much of the same structure represented in the resulting attention patterns, with interesting differences with respect to which parameters are frozen.

Recent work on the interpretability of deep neural language models has concluded that many properties of natural language syntax are encoded in their representational spaces. However, such studies often suffer from limited scope by focusing on a single language and a single linguistic formalism. In this study, we aim to investigate the extent to which the semblance of syntactic structure captured by language models adheres to a surface-syntactic or deep syntactic style of analysis, and whether the patterns are consistent across different languages. We apply a probe for extracting directed dependency trees to BERT and ELMo models trained on 13 different languages, probing for two different syntactic annotation styles: Universal Dependencies (UD), prioritizing deep syntactic relations, and Surface-Syntactic Universal Dependencies (SUD), focusing on surface structure. We find that both models exhibit a preference for UD over SUD — with interesting variations across languages and layers — and that the strength of this preference is correlated with differences in tree shape.

We present Køpsala, the Copenhagen-Uppsala system for the Enhanced Universal Dependencies Shared Task at IWPT 2020. Our system is a pipeline consisting of off-the-shelf models for everything but enhanced graph parsing, and for the latter, a transition-based graph parser adapted from Che et al. (2019). We train a single enhanced parser model per language, using gold sentence splitting and tokenization for training, and rely only on tokenized surface forms and multilingual BERT for encoding. While a bug introduced just before submission resulted in a severe drop in precision, its post-submission fix would bring us to 4th place in the official ranking, according to average ELAS. Our parser demonstrates that a unified pipeline is effective for both Meaning Representation Parsing and Enhanced Universal Dependencies.

Transition-based and graph-based dependency parsers have previously been shown to have complementary strengths and weaknesses: transition-based parsers exploit rich structural features but suffer from error propagation, while graph-based parsers benefit from global optimization but have restricted feature scope. In this paper, we show that, even though some details of the picture have changed after the switch to neural networks and continuous representations, the basic trade-off between rich features and global optimization remains essentially the same. Moreover, we show that deep contextualized word embeddings, which allow parsers to pack information about global sentence structure into local feature representations, benefit transition-based parsers more than graph-based parsers, making the two approaches virtually equivalent in terms of both accuracy and error profile. We argue that the reason is that these representations help prevent search errors and thereby allow transition-based parsers to better exploit their inherent strength of making accurate local decisions. We support this explanation by an error analysis of parsing experiments on 13 languages.

Representational Similarity Analysis (RSA) is a technique developed by neuroscientists for comparing activity patterns of different measurement modalities (e.g., fMRI, electrophysiology, behavior). As a framework, RSA has several advantages over existing approaches to interpretation of language encoders based on probing or diagnostic classification: namely, it does not require large training samples, is not prone to overfitting, and it enables a more transparent comparison between the representational geometries of different models and modalities. We demonstrate the utility of RSA by establishing a previously unknown correspondence between widely-employed pretrained language encoders and human processing difficulty via eye-tracking data, showcasing its potential in the interpretability toolbox for neural models.

We investigate the effects of multi-task learning using the recently introduced task of semantic tagging. We employ semantic tagging as an auxiliary task for three different NLP tasks: part-of-speech tagging, Universal Dependency parsing, and Natural Language Inference. We compare full neural network sharing, partial neural network sharing, and what we term the learning what to share setting where negative transfer between tasks is less likely. Our findings show considerable improvements for all tasks, particularly in the learning what to share setting which shows consistent gains across all tasks.

In this paper we describe our submission to SemEval-2018 Task 1: Affects in Tweets. The model which we present is an ensemble of various neural architectures and gradient boosted trees, and employs three different types of vectorial tweet representations. Furthermore, our system is language-independent and ranked first in 5 out of the 12 subtasks in which we participated, while achieving competitive results in the remaining ones. Comparatively remarkable performance is observed on both the Arabic and Spanish languages.

We participated to the SemEval-2018 shared task on capturing discriminative attributes (Task 10) with a simple system that ranked 8th amongst the 26 teams that took part in the evaluation. Our final score was 0.67, which is competitive with the winning score of 0.75, particularly given that our system is a zero-shot system that requires no training and minimal parameter optimisation. In addition to describing the submitted system, and discussing the implications of the relative success of such a system on this task, we also report on other, more complex models we experimented with.

In this paper, we explore the performance of a linear SVM trained on language independent character features for the NLI Shared Task 2017. Our basic system (GRONINGEN) achieves the best performance (87.56 F1-score) on the evaluation set using only 1-9 character n-grams as features. We compare this against several ensemble and meta-classifiers in order to examine how the linear system fares when combined with other, especially non-linear classifiers. Special emphasis is placed on the topic bias that exists by virtue of the assessment essay prompt distribution.