We present a method for exploring regions around individual points in a contextualized vector space (particularly, BERT space), as a way to investigate how these regions correspond to word senses. By inducing a contextualized "pseudoword" as a stand-in for a static embedding in the input layer, and then performing masked prediction of a word in the sentence, we are able to investigate the geometry of the BERT-space in a controlled manner around individual instances. Using our method on a set of carefully constructed sentences targeting ambiguous English words, we find substantial regularity in the contextualized space, with regions that correspond to distinct word senses; but between these regions there are occasionally "sense voids" -- regions that do not correspond to any intelligible sense.
While corpora of child speech and child-directed speech (CDS) have enabled major contributions to the study of child language acquisition, semantic annotation for such corpora is still scarce and lacks a uniform standard. We compile two CDS corpora with sentential logical forms, one in English and the other in Hebrew. In compiling the corpora we employ a methodology that enforces a cross-linguistically consistent representation, building on recent advances in dependency representation and semantic parsing. The corpora are based on a sizable portion of Brown's Adam corpus from CHILDES (about 80% of its child-directed utterances), and to all child-directed utterances from Berman's Hebrew CHILDES corpus Hagar. We begin by annotating the corpora with the Universal Dependencies (UD) scheme for syntactic annotation, motivated by its applicability to a wide variety of domains and languages. We then proceed by applying an automatic method for transducing sentential logical forms (LFs) from UD structures. The two representations have complementary strengths: UD structures are language-neutral and support direct annotation, whereas LFs are neutral as to the interface between syntax and semantics, and transparently encode semantic distinctions. We verify the quality of the annotated UD annotation using an inter-annotator agreement study. We then demonstrate the utility of the compiled corpora through a longitudinal corpus study of the prevalence of different syntactic and semantic phenomena.
The learning trajectories of linguistic phenomena provide insight into the nature of linguistic representation, beyond what can be gleaned from inspecting the behavior of an adult speaker. To apply a similar approach to analyze neural language models (NLM), it is first necessary to establish that different models are similar enough in the generalizations they make. In this paper, we show that NLMs with different initialization, architecture, and training data acquire linguistic phenomena in a similar order, despite having different end performances over the data. Leveraging these findings, we compare the relative performance on different phenomena at varying learning stages with simpler reference models. Results suggest that NLMs exhibit consistent ``developmental'' stages. Initial analysis of these stages presents phenomena clusters (notably morphological ones), whose performance progresses in unison, suggesting potential links between their acquired representations.
In this research paper, I will elaborate on a method to evaluate machine translation models based on their performance on underlying syntactical phenomena between English and Arabic languages. This method is especially important as such "neural" and "machine learning" are hard to fine-tune and change. Thus, finding a way to evaluate them easily and diversely would greatly help the task of bettering them.
Neural knowledge-grounded generative models for dialogue often produce content that is factually inconsistent with the source text they rely on. As a consequence, such models are unreliable, limiting their real-world applicability. Inspired by recent work on evaluating factual consistency in abstractive summarization (Durmus et al., 2020; Wang et al., 2020), we propose an automatic evaluation metric for factual consistency in knowledge-grounded dialogue models using automatic question generation and question answering. Unlike previous works which use naive token-based comparison of answer spans, our metric makes use of co-reference resolution and natural language inference capabilities which greatly improve its performance. To foster proper evaluation, we curate a novel dataset of state-of-the-art dialogue system outputs for the Wizard-of-Wikipedia dataset (Dinan et al., 2019), which we manually annotate for factual consistency. We perform a thorough meta-evaluation of our metric against other metrics using the new dataset and two others, where it greatly outperforms the baselines.
Probing neural models for the ability to perform downstream tasks using their activation patterns is often used to localize what parts of the network specialize in performing what tasks. However, little work addressed potential mediating factors in such comparisons. As a test-case mediating factor, we consider the prediction's context length, namely the length of the span whose processing is minimally required to perform the prediction. We show that not controlling for context length may lead to contradictory conclusions as to the localization patterns of the network, depending on the distribution of the probing dataset. Indeed, when probing BERT with seven tasks, we find that it is possible to get 196 different rankings between them when manipulating the distribution of context lengths in the probing dataset. We conclude by presenting best practices for conducting such comparisons in the future.
SERRANT is a system and code for automatic classification of English grammatical errors that combines SErCl and ERRANT. SERRANT uses ERRANT's annotations when they are informative and those provided by SErCl otherwise.
While a number of works showed gains from incorporating source-side symbolic syntactic and semantic structure into neural machine translation (NMT), much fewer works addressed the decoding of such structure. We propose a general Transformer-based approach for tree and graph decoding based on generating a sequence of transitions, inspired by a similar approach that uses RNNs by Dyer (2016). Experiments with using the proposed decoder with Universal Dependencies syntax on English-German, German-English and English-Russian show improved performance over the standard Transformer decoder, as well as over ablated versions of the model.\tacltxt{\footnote{All code implementing the presented models will be released upon acceptance.
This is the annotation manual for Universal Conceptual Cognitive Annotation (UCCA; Abend and Rappoport, 2013), specifically the Foundational Layer. UCCA is a graph-based semantic annotation scheme based on typological linguistic principles. It has been applied to several languages; for ease of exposition these guidelines give examples mainly in English. New annotators may wish to start with the tutorial on the UCCA framework (Abend et al., 2020). Further resources are available at the project homepage: https://universalconceptualcognitiveannotation.github.io
Building robust natural language understanding systems will require a clear characterization of whether and how various linguistic meaning representations complement each other. To perform a systematic comparative analysis, we evaluate the mapping between meaning representations from different frameworks using two complementary methods: (i) a rule-based converter, and (ii) a supervised delexicalized parser that parses to one framework using only information from the other as features. We apply these methods to convert the STREUSLE corpus (with syntactic and lexical semantic annotations) to UCCA (a graph-structured full-sentence meaning representation). Both methods yield surprisingly accurate target representations, close to fully supervised UCCA parser quality---indicating that UCCA annotations are partially redundant with STREUSLE annotations. Despite this substantial convergence between frameworks, we find several important areas of divergence.