John Bohannon

Leila Khalili, Yao You, John Bohannon

Transformer language models provide superior accuracy over previous models but they are computationally and environmentally expensive. Borrowing the concept of model cascading from computer vision, we introduce BabyBear, a framework for cascading models for natural language processing (NLP) tasks to minimize cost. The core strategy is inference triage, exiting early when the least expensive model in the cascade achieves a sufficiently high-confidence prediction. We test BabyBear on several open source data sets related to document classification and entity recognition. We find that for common NLP tasks a high proportion of the inference load can be accomplished with cheap, fast models that have learned by observing a deep learning model. This allows us to reduce the compute cost of large-scale classification jobs by more than 50% while retaining overall accuracy. For named entity recognition, we save 33% of the deep learning compute while maintaining an F1 score higher than 95% on the CoNLL benchmark.

Oleg Vasilyev, Randy Sawaya, John Bohannon

A query encoder of a dual passage retrieval system can be tuned for specific types of queries or domains, while the precomputed and stored documents representations are kept intact. Switching from one query encoder to another when needed is easily feasible, unlike overhauling the embeddings of a whole knowledge base. In this work we raise a question: Can the generic, original qualities of the encoder be preserved or at least left not too degraded when it is tuned on a narrow domain? We conducted experiments on a high quality multilingual embedding model: Tuning it on a single English-only dataset, we observe that the tuning not only preserves the multilingual qualities, but even improves them. The embedding qualities on distinctly different data are also improved or at least preserved. Drawing on our observations, we suggest a more general hypothesis: Tuning with intentionally low learning rate can preserve or improve a system's properties acquired in training, but not specifically targeted by tuning. We call this adiabatic tuning and provide tentative explanations.

Oleg Vasilyev, John Bohannon

We propose a new kind of embedding for natural language text that deeply represents semantic meaning. Standard text embeddings use the outputs from hidden layers of a pretrained language model. In our method, we let a language model learn from the text and then literally pick its brain, taking the actual weights of the model's neurons to generate a vector. We call this representation of the text a neural embedding. We confirm the ability of this representation to reflect semantics of the text by an analysis of its behavior on several datasets, and by a comparison of neural embedding with state of the art sentence embeddings.

Oleg Vasilyev, Alex Dauenhauer, Vedant Dharnidharka et al.

We propose a simple and practical method for named entity linking (NEL), based on entity representation by multiple embeddings. To explore this method, and to review its dependency on parameters, we measure its performance on Namesakes, a highly challenging dataset of ambiguously named entities. Our observations suggest that the minimal number of mentions required to create a knowledge base (KB) entity is very important for NEL performance. The number of embeddings is less important and can be kept small, within as few as 10 or less. We show that our representations of KB entities can be adjusted using only KB data, and the adjustment can improve NEL performance. We also compare NEL performance of embeddings obtained from tuning language model on diverse news texts as opposed to tuning on more uniform texts from public datasets XSum, CNN / Daily Mail. We found that tuning on diverse news provides better embeddings.

Oleg Vasilyev, John Bohannon

We found that a simple property of clusters in a clustered dataset of news correlate strongly with importance and urgency of news (IUN) as assessed by LLM. We verified our finding across different news datasets, dataset sizes, clustering algorithms and embeddings. The found correlation should allow using clustering (as an alternative to LLM) for identifying the most important urgent news, or for filtering out unimportant articles.

Oleg Vasilyev, Fumika Isono, John Bohannon

Semantics of a sentence is defined with much less ambiguity than semantics of a single word, and we assume that it should be better preserved by translation to another language. If multilingual sentence embeddings intend to represent sentence semantics, then the similarity between embeddings of any two sentences must be invariant with respect to translation. Based on this suggestion, we consider a simple linear cross-lingual mapping as a possible improvement of the multilingual embeddings. We also consider deviation from orthogonality conditions as a measure of deficiency of the embeddings.

Oleg Vasilyev, John Bohannon

Factual consistency is one of important summary evaluation dimensions, especially as summary generation becomes more fluent and coherent. The ESTIME measure, recently proposed specifically for factual consistency, achieves high correlations with human expert scores both for consistency and fluency, while in principle being restricted to evaluating such text-summary pairs that have high dictionary overlap. This is not a problem for current styles of summarization, but it may become an obstacle for future summarization systems, or for evaluating arbitrary claims against the text. In this work we generalize the method, and make a variant of the measure applicable to any text-summary pairs. As ESTIME uses points of contextual similarity, it provides insights into usefulness of information taken from different BERT layers. We observe that useful information exists in almost all of the layers except the several lowest ones. For consistency and fluency - qualities focused on local text details - the most useful layers are close to the top (but not at the top); for coherence and relevance we found a more complicated and interesting picture.

Oleg Vasilyev, Aysu Altun, Nidhi Vyas et al.

We present Namesakes, a dataset of ambiguously named entities obtained from English-language Wikipedia and news articles. It consists of 58862 mentions of 4148 unique entities and their namesakes: 1000 mentions from news, 28843 from Wikipedia articles about the entity, and 29019 Wikipedia backlink mentions. Namesakes should be helpful in establishing challenging benchmarks for the task of named entity linking (NEL).

Spencer Braun, Oleg Vasilyev, Neslihan Iskender et al.

The creation of a quality summarization dataset is an expensive, time-consuming effort, requiring the production and evaluation of summaries by both trained humans and machines. If such effort is made in one language, it would be beneficial to be able to use it in other languages without repeating human annotations. To investigate how much we can trust machine translation of such a dataset, we translate the English dataset SummEval to seven languages and compare performance across automatic evaluation measures. We explore equivalence testing as the appropriate statistical paradigm for evaluating correlations between human and automated scoring of summaries. While we find some potential for dataset reuse in languages similar to the source, most summary evaluation methods are not found to be statistically equivalent across translations.

Neslihan Iskender, Oleg Vasilyev, Tim Polzehl et al.

Evaluating large summarization corpora using humans has proven to be expensive from both the organizational and the financial perspective. Therefore, many automatic evaluation metrics have been developed to measure the summarization quality in a fast and reproducible way. However, most of the metrics still rely on humans and need gold standard summaries generated by linguistic experts. Since BLANC does not require golden summaries and supposedly can use any underlying language model, we consider its application to the evaluation of summarization in German. This work demonstrates how to adjust the BLANC metric to a language other than English. We compare BLANC scores with the crowd and expert ratings, as well as with commonly used automatic metrics on a German summarization data set. Our results show that BLANC in German is especially good in evaluating informativeness.

Oleg Vasilyev, John Bohannon

We propose a new reference-free summary quality evaluation measure, with emphasis on the faithfulness. The measure is designed to find and count all possible minute inconsistencies of the summary with respect to the source document. The proposed ESTIME, Estimator of Summary-to-Text Inconsistency by Mismatched Embeddings, correlates with expert scores in summary-level SummEval dataset stronger than other common evaluation measures not only in Consistency but also in Fluency. We also introduce a method of generating subtle factual errors in human summaries. We show that ESTIME is more sensitive to subtle errors than other common evaluation measures.

Nicholas Egan, Oleg Vasilyev, John Bohannon

The goal of a summary is to concisely state the most important information in a document. With this principle in mind, we introduce new reference-free summary evaluation metrics that use a pretrained language model to estimate the information content shared between a document and its summary. These metrics are a modern take on the Shannon Game, a method for summary quality scoring proposed decades ago, where we replace human annotators with language models. We also view these metrics as an extension of BLANC, a recently proposed approach to summary quality measurement based on the performance of a language model with and without the help of a summary. Using transformer based language models, we empirically verify that our metrics achieve state-of-the-art correlation with human judgement of the summary quality dimensions of both coherence and relevance, as well as competitive correlation with human judgement of consistency and fluency.

Oleg Vasilyev, John Bohannon

Normally, summary quality measures are compared with quality scores produced by human annotators. A higher correlation with human scores is considered to be a fair indicator of a better measure. We discuss observations that cast doubt on this view. We attempt to show a possibility of an alternative indicator. Given a family of measures, we explore a criterion of selecting the best measure not relying on correlations with human scores. Our observations for the BLANC family of measures suggest that the criterion is universal across very different styles of summaries.

Nicholas Egan, John Bohannon

The prevalence of ambiguous acronyms make scientific documents harder to understand for humans and machines alike, presenting a need for models that can automatically identify acronyms in text and disambiguate their meaning. We introduce new methods for acronym identification and disambiguation: our acronym identification model projects learned token embeddings onto tag predictions, and our acronym disambiguation model finds training examples with similar sentence embeddings as test examples. Both of our systems achieve significant performance gains over previously suggested methods, and perform competitively on the SDU@AAAI-21 shared task leaderboard. Our models were trained in part on new distantly-supervised datasets for these tasks which we call AuxAI and AuxAD. We also identified a duplication conflict issue in the SciAD dataset, and formed a deduplicated version of SciAD that we call SciAD-dedupe. We publicly released all three of these datasets, and hope that they help the community make further strides in scientific document understanding.

Oleg Vasilyev, Vedant Dharnidharka, Nicholas Egan et al.

We explore the sensitivity of a document summary quality estimator, BLANC, to human assessment of qualities for the same summaries. In our human evaluations, we distinguish five summary qualities, defined by how fluent, understandable, informative, compact, and factually correct the summary is. We make the case for optimal BLANC parameters, at which the BLANC sensitivity to almost all of summary qualities is about as good as the sensitivity of a human annotator.

Oleg Vasilyev, Kathryn Evans, Anna Venancio-Marques et al.

We present an approach to generating topics using a model trained only for document title generation, with zero examples of topics given during training. We leverage features that capture the relevance of a candidate span in a document for the generation of a title for that document. The output is a weighted collection of the phrases that are most relevant for describing the document and distinguishing it within a corpus, without requiring access to the rest of the corpus. We conducted a double-blind trial in which human annotators scored the quality of our machine-generated topics along with original human-written topics associated with news articles from The Guardian and The Huffington Post. The results show that our zero-shot model generates topic labels for news documents that are on average equal to or higher quality than those written by humans, as judged by humans.

Oleg Vasilyev, Vedant Dharnidharka, John Bohannon

We present BLANC, a new approach to the automatic estimation of document summary quality. Our goal is to measure the functional performance of a summary with an objective, reproducible, and fully automated method. Our approach achieves this by measuring the performance boost gained by a pre-trained language model with access to a document summary while carrying out its language understanding task on the document's text. We present evidence that BLANC scores have as good correlation with human evaluations as do the ROUGE family of summary quality measurements. And unlike ROUGE, the BLANC method does not require human-written reference summaries, allowing for fully human-free summary quality estimation.

Oleg Vasilyev, Tom Grek, John Bohannon

We propose a novel method for generating titles for unstructured text documents. We reframe the problem as a sequential question-answering task. A deep neural network is trained on document-title pairs with decomposable titles, meaning that the vocabulary of the title is a subset of the vocabulary of the document. To train the model we use a corpus of millions of publicly available document-title pairs: news articles and headlines. We present the results of a randomized double-blind trial in which subjects were unaware of which titles were human or machine-generated. When trained on approximately 1.5 million news articles, the model generates headlines that humans judge to be as good or better than the original human-written headlines in the majority of cases.