Simultaneous interpretation is an especially challenging form of translation because it requires converting speech from one language to another in real-time. Though prior work has relied on out-of-the-box machine translation metrics to evaluate interpretation data, we hypothesize that strategies common in high-quality human interpretations, such as summarization, may not be handled well by standard machine translation metrics. In this work, we examine both qualitatively and quantitatively four potential barriers to evaluation of interpretation: disfluency, summarization, paraphrasing, and segmentation. Our experiments reveal that, while some machine translation metrics correlate fairly well with human judgments of interpretation quality, much work is still needed to account for strategies of interpretation during evaluation. As a first step to address this, we develop a fine-tuned model for interpretation evaluation, and achieve better correlation with human judgments than the state-of-the-art machine translation metrics.

Translated texts bear several hallmarks distinct from texts originating in the language (“translationese”). Though individual translated texts are often fluent and preserve meaning, at a large scale, translated texts have statistical tendencies which distinguish them from text originally written in the language and can affect model performance. We frame the novel task of translationese reduction and hypothesize that Abstract Meaning Representation (AMR), a graph-based semantic representation which abstracts away from the surface form, can be used as an interlingua to reduce the amount of translationese in translated texts. By parsing English translations into an AMR and then generating text from that AMR, the result more closely resembles originally English text across three quantitative macro-level measures, without severely compromising fluency or adequacy. We compare our AMR-based approach against three other techniques based on machine translation or paraphrase generation. This work represents the first approach to reducing translationese in text and highlights the promise of AMR, given that our AMR-based approach outperforms more computationally intensive methods.

Among the problems with leaderboard culture in NLP has been the widespread lack of confidence estimation in reported results. In this work, we present a framework and simulator for estimating p-values for comparisons between the results of two systems, in order to understand the confidence that one is actually better (i.e. ranked higher) than the other. What has made this difficult in the past is that each system must itself be evaluated by comparison to a gold standard. We define a null hypothesis that each system’s metric scores are drawn from the same distribution, using variance found naturally (though rarely reported) in test set items and individual labels on an item (responses) to produce the metric distributions. We create a test set that evenly mixes the responses of the two systems under the assumption the null hypothesis is true. Exploring how to best estimate the true p-value from a single test set under different metrics, tests, and sampling methods, we find that the presence of response variance (from multiple raters or multiple model versions) has a profound impact on p-value estimates for model comparison, and that choice of metric and sampling method is critical to providing statistical guarantees on model comparisons.

Identifying semantically equivalent sentences is important for many NLP tasks. Current approaches to semantic equivalence take a loose, sentence-level approach to “equivalence,” despite evidence that fine-grained differences and implicit content have an effect on human understanding and system performance. In this work, we introduce a novel, more sensitive method of characterizing cross-lingual semantic equivalence that leverages Abstract Meaning Representation graph structures. We find that parsing sentences into AMRs and comparing the AMR graphs enables finer-grained equivalence measurement than comparing the sentences themselves. We demonstrate that when using gold or even automatically parsed AMR annotations, our solution is finer-grained than existing corpus filtering methods and more accurate at predicting strictly equivalent sentences than existing semantic similarity metrics.

The task of natural language inference (NLI) asks whether a given premise (expressed in NL) entails a given NL hypothesis. NLI benchmarks contain human ratings of entailment, but the meaning relationships driving these ratings are not formalized. Can the underlying sentence pair relationships be made more explicit in an interpretable yet robust fashion? We compare semantic structures to represent premise and hypothesis, including sets of *contextualized embeddings* and *semantic graphs* (Abstract Meaning Representations), and measure whether the hypothesis is a semantic substructure of the premise, utilizing interpretable metrics. Our evaluation on three English benchmarks finds value in both contextualized embeddings and semantic graphs; moreover, they provide complementary signals, and can be leveraged together in a hybrid model.

The term translationese describes the set of linguistic features unique to translated texts, which appear regardless of translation quality. Though automatic classifiers designed to distinguish translated texts achieve high accuracy and prior work has identified common hallmarks of translationese, human accuracy of identifying translated text is understudied. In this work, we perform a human evaluation of English original/translated texts in order to explore raters’ ability to classify texts as being original or translated English and the features that lead a rater to judge text as being translated. Ultimately, we find that, regardless of the annotators’ native language or the source language of the text, annotators are unable to distinguish translations from original English texts and also have low agreement. Our results provide critical insight into work in translation studies and context for assessments of translationese classifiers.

Abstract Meaning Representation (AMR) is a popular semantic annotation schema that presents sentence meaning as a graph while abstracting away from syntax. It was originally designed for English, but has since been extended to a variety of non-English versions of AMR. These cross-lingual adaptations, to varying degrees, incorporate language-specific features necessary to effectively capture the semantics of the language being annotated. Uniform Meaning Representation (UMR) on the other hand, the multilingual extension of AMR, was designed specifically for cross-lingual applications. In this work, we discuss these two approaches to extending AMR beyond English. We describe both approaches, compare the information they capture for a case language (Spanish), and outline implications for future work.

Rooted in AMR, Uniform Meaning Representation (UMR) is a graph-based formalism with nodes as concepts and edges as relations between them. When used to represent natural language semantics, UMR maps words in a sentence to concepts in the UMR graph. Multiword expressions (MWEs) pose a particular challenge to UMR annotation because they deviate from the default one-to-one mapping between words and concepts. There are different types of MWEs which require different kinds of annotation that must be specified in guidelines. This paper discusses the specific treatment for each type of MWE in UMR.

In this work, we use sentence similarity as a lens through which to investigate the representation of meaning in graphs vs. vectors. On semantic textual similarity data, we examine how similarity metrics based on vectors alone (SENTENCE-BERT and BERTSCORE) fare compared to metrics based on AMR graphs (SMATCH and S2MATCH). Quantitative and qualitative analyses show that the AMR-based metrics can better capture meanings dependent on sentence structures, but can also be distracted by structural differences—whereas the BERT-based metrics represent finer-grained meanings of individual words, but often fail to capture the ordering effect of words within sentences and suffer from interpretability problems. These findings contribute to our understanding of each approach to semantic representation and motivate distinct use cases for graph and vector-based representations.

Cross-lingual Abstract Meaning Representation (AMR) parsers are currently evaluated in comparison to gold English AMRs, despite parsing a language other than English, due to the lack of multilingual AMR evaluation metrics. This evaluation practice is problematic because of the established effect of source language on AMR structure. In this work, we present three multilingual adaptations of monolingual AMR evaluation metrics and compare the performance of these metrics to sentence-level human judgments. We then use our most highly correlated metric to evaluate the output of state-of-the-art cross-lingual AMR parsers, finding that Smatch may still be a useful metric in comparison to gold English AMRs, while our multilingual adaptation of S2match (XS2match) is best for comparison with gold in-language AMRs.

Abstract Meaning Representation (AMR), originally designed for English, has been adapted to a number of languages to facilitate cross-lingual semantic representation and analysis. We build on previous work and present the first sizable, general annotation project for Spanish AMR. We release a detailed set of annotation guidelines and a corpus of 486 gold-annotated sentences spanning multiple genres from an existing, cross-lingual AMR corpus. Our work constitutes the second largest non-English gold AMR corpus to date. Fine-tuning an AMR to-Spanish generation model with our annotations results in a BERTScore improvement of 8.8%, demonstrating initial utility of our work.

The careful design of a crowdsourcing protocol is critical to eliciting highly accurate annotations from untrained workers. In this work, we explore the development of crowdsourcing protocols for a challenging word sense disambiguation task. We find that (a) selecting a similar example usage can serve as a proxy for selecting an explicit definition of the sense, and (b) priming workers with an additional, related task within the HIT improves performance on the main proxy task. Ultimately, we demonstrate the usefulness of our crowdsourcing elicitation technique as an effective alternative to previously investigated training strategies, which can be used if agreement on a challenging task is low.

The Abstract Meaning Representation (AMR) annotation schema was originally designed for English. But the formalism has since been adapted for annotation in a variety of languages. Meanwhile, cross-lingual parsers have been developed to derive English AMR representations for sentences from other languages—implicitly assuming that English AMR can approximate an interlingua. In this work, we investigate the similarity of AMR annotations in parallel data and how much the language matters in terms of the graph structure. We set out to quantify the effect of sentence language on the structure of the parsed AMR. As a case study, we take parallel AMR annotations from Mandarin Chinese and English AMRs, and replace all Chinese concepts with equivalent English tokens. We then compare the two graphs via the Smatch metric as a measure of structural similarity. We find that source language has a dramatic impact on AMR structure, with Smatch scores below 50% between English and Chinese graphs in our sample—an important reference point for interpreting Smatch scores in cross-lingual AMR parsing.

Translation divergences are varied and widespread, challenging approaches that rely on parallel text. To annotate translation divergences, we propose a schema grounded in the Abstract Meaning Representation (AMR), a sentence-level semantic framework instantiated for a number of languages. By comparing parallel AMR graphs, we can identify specific points of divergence. Each divergence is labeled with both a type and a cause. We release a small corpus of annotated English-Spanish data, and analyze the annotations in our corpus.

We present the Prepositions Annotated with Supsersense Tags in Reddit International English (“PASTRIE”) corpus, a new dataset containing manually annotated preposition supersenses of English data from presumed speakers of four L1s: English, French, German, and Spanish. The annotations are comprehensive, covering all preposition types and tokens in the sample. Along with the corpus, we provide analysis of distributional patterns across the included L1s and a discussion of the influence of L1s on L2 preposition choice.

Prepositional supersense annotation is time-consuming and requires expert training. Here, we present two sensible methods for obtaining prepositional supersense annotations indirectly by eliciting surface substitution and similarity judgments. Four pilot studies suggest that both methods have potential for producing prepositional supersense annotations that are comparable in quality to expert annotations.

Most current state-of-the art systems for generating English text from Abstract Meaning Representation (AMR) have been evaluated only using automated metrics, such as BLEU, which are known to be problematic for natural language generation. In this work, we present the results of a new human evaluation which collects fluency and adequacy scores, as well as categorization of error types, for several recent AMR generation systems. We discuss the relative quality of these systems and how our results compare to those of automatic metrics, finding that while the metrics are mostly successful in ranking systems overall, collecting human judgments allows for more nuanced comparisons. We also analyze common errors made by these systems.

The Spanish Learner Language Oral Corpora (SPLLOC) of transcribed conversations between investigators and language learners contains a set of neologism tags. In this work, the utterances tagged as neologisms are broken down into three categories: true neologisms, loanwords, and errors. This work examines the relationships between neologism, loanword, and error production and both language learner level and conversation task. The results of this study suggest that loanwords and errors are produced most frequently by language learners with moderate experience, while neologisms are produced most frequently by native speakers. This study also indicates that tasks that require descriptions of images draw more neologism, loanword and error production. We ultimately present a unique analysis of the implications of neologism, loanword, and error production useful for further work in second language acquisition research, as well as for language educators.