Pedro Szekely

Fei Wang, Zhewei Xu, Pedro Szekely et al.

Controlled table-to-text generation seeks to generate natural language descriptions for highlighted subparts of a table. Previous SOTA systems still employ a sequence-to-sequence generation method, which merely captures the table as a linear structure and is brittle when table layouts change. We seek to go beyond this paradigm by (1) effectively expressing the relations of content pieces in the table, and (2) making our model robust to content-invariant structural transformations. Accordingly, we propose an equivariance learning framework, which encodes tables with a structure-aware self-attention mechanism. This prunes the full self-attention structure into an order-invariant graph attention that captures the connected graph structure of cells belonging to the same row or column, and it differentiates between relevant cells and irrelevant cells from the structural perspective. Our framework also modifies the positional encoding mechanism to preserve the relative position of tokens in the same cell but enforce position invariance among different cells. Our technology is free to be plugged into existing table-to-text generation models, and has improved T5-based models to offer better performance on ToTTo and HiTab. Moreover, on a harder version of ToTTo, we preserve promising performance, while previous SOTA systems, even with transformation-based data augmentation, have seen significant performance drops. Our code is available at https://github.com/luka-group/Lattice.

Hyundong Cho, Nicolaas Jedema, Leonardo F. R. Ribeiro et al. · amazon-science

Current instruction-tuned language models are exclusively trained with textual preference data and thus are often not aligned with the unique requirements of other modalities, such as speech. To better align language models with the speech domain, we explore (i) prompting strategies grounded in radio-industry best practices and (ii) preference learning using a novel speech-based preference data of 20K samples, generated with a wide spectrum of prompts that induce varying dimensions of speech-suitability and labeled by annotators who listen to response pairs. Both human and automatic evaluation show that both prompting and preference learning increase the speech-suitability of popular instruction-tuned LLMs. Interestingly, we find that prompting and preference learning can be additive; combining them achieves the best win rates in head-to-head comparison, resulting in responses that are preferred or tied to the base model in 76.2% of comparisons on average. Lastly, we share lexical, syntactical, and qualitative analyses to showcase how each method contributes to improving the speech-suitability of generated responses.

Jiang Wang, Filip Ilievski, Pedro Szekely et al.

Embedding methods have demonstrated robust performance on the task of link prediction in knowledge graphs, by mostly encoding entity relationships. Recent methods propose to enhance the loss function with a literal-aware term. In this paper, we propose KGA: a knowledge graph augmentation method that incorporates literals in an embedding model without modifying its loss function. KGA discretizes quantity and year values into bins, and chains these bins both horizontally, modeling neighboring values, and vertically, modeling multiple levels of granularity. KGA is scalable and can be used as a pre-processing step for any existing knowledge graph embedding model. Experiments on legacy benchmarks and a new large benchmark, DWD, show that augmenting the knowledge graph with quantities and years is beneficial for predicting both entities and numbers, as KGA outperforms the vanilla models and other relevant baselines. Our ablation studies confirm that both quantities and years contribute to KGA's performance, and that its performance depends on the discretization and binning settings. We make the code, models, and the DWD benchmark publicly available to facilitate reproducibility and future research.

Bohui Zhang, Filip Ilievski, Pedro Szekely

Large public knowledge graphs, like Wikidata, contain billions of statements about tens of millions of entities, thus inspiring various use cases to exploit such knowledge graphs. However, practice shows that much of the relevant information that fits users' needs is still missing in Wikidata, while current linked open data (LOD) tools are not suitable to enrich large graphs like Wikidata. In this paper, we investigate the potential of enriching Wikidata with structured data sources from the LOD cloud. We present a novel workflow that includes gap detection, source selection, schema alignment, and semantic validation. We evaluate our enrichment method with two complementary LOD sources: a noisy source with broad coverage, DBpedia, and a manually curated source with a narrow focus on the art domain, Getty. Our experiments show that our workflow can enrich Wikidata with millions of novel statements from external LOD sources with high quality. Property alignment and data quality are key challenges, whereas entity alignment and source selection are well-supported by existing Wikidata mechanisms. We make our code and data available to support future work.

Fei Wang, Kexuan Sun, Jay Pujara et al.

Tables provide valuable knowledge that can be used to verify textual statements. While a number of works have considered table-based fact verification, direct alignments of tabular data with tokens in textual statements are rarely available. Moreover, training a generalized fact verification model requires abundant labeled training data. In this paper, we propose a novel system to address these problems. Inspired by counterfactual causality, our system identifies token-level salience in the statement with probing-based salience estimation. Salience estimation allows enhanced learning of fact verification from two perspectives. From one perspective, our system conducts masked salient token prediction to enhance the model for alignment and reasoning between the table and the statement. From the other perspective, our system applies salience-aware data augmentation to generate a more diverse set of training instances by replacing non-salient terms. Experimental results on TabFact show the effective improvement by the proposed salience-aware learning techniques, leading to the new SOTA performance on the benchmark. Our code is publicly available at https://github.com/luka-group/Salience-aware-Learning .

Mayank Kejriwal, Jiayuan Ding, Runqi Shao et al.

In this paper, we describe and study the indicator mining problem in the online sex advertising domain. We present an in-development system, FlagIt (Flexible and adaptive generation of Indicators from text), which combines the benefits of both a lightweight expert system and classical semi-supervision (heuristic re-labeling) with recently released state-of-the-art unsupervised text embeddings to tag millions of sentences with indicators that are highly correlated with human trafficking. The FlagIt technology stack is open source. On preliminary evaluations involving five indicators, FlagIt illustrates promising performance compared to several alternatives. The system is being actively developed, refined and integrated into a domain-specific search system used by over 200 law enforcement agencies to combat human trafficking, and is being aggressively extended to mine at least six more indicators with minimal programming effort. FlagIt is a good example of a system that operates in limited label settings, and that requires creative combinations of established machine learning techniques to produce outputs that could be used by real-world non-technical analysts.

Daniel M. Benjamin, Fred Morstatter, Ali E. Abbas et al. · stanford

Sound decision-making relies on accurate prediction for tangible outcomes ranging from military conflict to disease outbreaks. To improve crowdsourced forecasting accuracy, we developed SAGE, a hybrid forecasting system that combines human and machine generated forecasts. The system provides a platform where users can interact with machine models and thus anchor their judgments on an objective benchmark. The system also aggregates human and machine forecasts weighting both for propinquity and based on assessed skill while adjusting for overconfidence. We present results from the Hybrid Forecasting Competition (HFC) - larger than comparable forecasting tournaments - including 1085 users forecasting 398 real-world forecasting problems over eight months. Our main result is that the hybrid system generated more accurate forecasts compared to a human-only baseline which had no machine generated predictions. We found that skilled forecasters who had access to machine-generated forecasts outperformed those who only viewed historical data. We also demonstrated the inclusion of machine-generated forecasts in our aggregation algorithms improved performance, both in terms of accuracy and scalability. This suggests that hybrid forecasting systems, which potentially require fewer human resources, can be a viable approach for maintaining a competitive level of accuracy over a larger number of forecasting questions.

Ehsan Qasemi, Lee Kezar, Jay Pujara et al.

Language models (LMs) show state of the art performance for common sense (CS) question answering, but whether this ability implies a human-level mastery of CS remains an open question. Understanding the limitations and strengths of LMs can help researchers improve these models, potentially by developing novel ways of integrating external CS knowledge. We devise a series of tests and measurements to systematically quantify their performance on different aspects of CS. We propose the use of cloze testing combined with word embeddings to measure the LM's robustness and confidence. Our results show than although language models tend to achieve human-like accuracy, their confidence is subpar. Future work can leverage this information to build more complex systems, such as an ensemble of symbolic and distributed knowledge.

Yixiang Yao, Tanmay Ghai, Srivatsan Ravi et al.

Entity resolution is the task of identifying records in different datasets that refer to the same entity in the real world. In sensitive domains (e.g. financial accounts, hospital health records), entity resolution must meet privacy requirements to avoid revealing sensitive information such as personal identifiable information to untrusted parties. Existing solutions are either too algorithmically-specific or come with an implicit trade-off between accuracy of the computation, privacy, and run-time efficiency. We propose AMMPERE, an abstract computation model for performing universal privacy-preserving entity resolution. AMPPERE offers abstractions that encapsulate multiple algorithmic and platform-agnostic approaches using variants of Jaccard similarity to perform private data matching and entity resolution. Specifically, we show that two parties can perform entity resolution over their data, without leaking sensitive information. We rigorously compare and analyze the feasibility, performance overhead and privacy-preserving properties of these approaches on the Sharemind multi-party computation (MPC) platform as well as on PALISADE, a lattice-based homomorphic encryption library. The AMPPERE system demonstrates the efficacy of privacy-preserving entity resolution for real-world data while providing a precise characterization of the induced cost of preventing information leakage.

Filip Ilievski, Pedro Szekely, Gleb Satyukov et al.

While the similarity between two concept words has been evaluated and studied for decades, much less attention has been devoted to algorithms that can compute the similarity of nodes in very large knowledge graphs, like Wikidata. To facilitate investigations and head-to-head comparisons of similarity algorithms on Wikidata, we present a user-friendly interface that allows flexible computation of similarity between Qnodes in Wikidata. At present, the similarity interface supports four algorithms, based on: graph embeddings (TransE, ComplEx), text embeddings (BERT), and class-based similarity. We demonstrate the behavior of the algorithms on representative examples about semantically similar, related, and entirely unrelated entity pairs. To support anticipated applications that require efficient similarity computations, like entity linking and recommendation, we also provide a REST API that can compute most similar neighbors for any Qnode in Wikidata.

Hans Chalupsky, Pedro Szekely, Filip Ilievski et al.

Application developers today have three choices for exploiting the knowledge present in Wikidata: they can download the Wikidata dumps in JSON or RDF format, they can use the Wikidata API to get data about individual entities, or they can use the Wikidata SPARQL endpoint. None of these methods can support complex, yet common, query use cases, such as retrieval of large amounts of data or aggregations over large fractions of Wikidata. This paper introduces KGTK Kypher, a query language and processor that allows users to create personalized variants of Wikidata on a laptop. We present several use cases that illustrate the types of analyses that Kypher enables users to run on the full Wikidata KG on a laptop, combining data from external resources such as DBpedia. The Kypher queries for these use cases run much faster on a laptop than the equivalent SPARQL queries on a Wikidata clone running on a powerful server with 24h time-out limits.

Kartik Shenoy, Filip Ilievski, Daniel Garijo et al.

Wikidata has been increasingly adopted by many communities for a wide variety of applications, which demand high-quality knowledge to deliver successful results. In this paper, we develop a framework to detect and analyze low-quality statements in Wikidata by shedding light on the current practices exercised by the community. We explore three indicators of data quality in Wikidata, based on: 1) community consensus on the currently recorded knowledge, assuming that statements that have been removed and not added back are implicitly agreed to be of low quality; 2) statements that have been deprecated; and 3) constraint violations in the data. We combine these indicators to detect low-quality statements, revealing challenges with duplicate entities, missing triples, violated type rules, and taxonomic distinctions. Our findings complement ongoing efforts by the Wikidata community to improve data quality, aiming to make it easier for users and editors to find and correct mistakes.

Ehsan Qasemi, Filip Ilievski, Muhao Chen et al.

Humans can seamlessly reason with circumstantial preconditions of commonsense knowledge. We understand that a glass is used for drinking water, unless the glass is broken or the water is toxic. Despite state-of-the-art (SOTA) language models' (LMs) impressive performance on inferring commonsense knowledge, it is unclear whether they understand the circumstantial preconditions. To address this gap, we propose a novel challenge of reasoning with circumstantial preconditions. We collect a dataset, called PaCo, consisting of 12.4 thousand preconditions of commonsense statements expressed in natural language. Based on this dataset, we create three canonical evaluation tasks and use them to examine the capability of existing LMs to understand situational preconditions. Our results reveal a 10-30% gap between machine and human performance on our tasks, which shows that reasoning with preconditions is an open challenge.

Avijit Thawani, Jay Pujara, Pedro A. Szekely et al.

NLP systems rarely give special consideration to numbers found in text. This starkly contrasts with the consensus in neuroscience that, in the brain, numbers are represented differently from words. We arrange recent NLP work on numeracy into a comprehensive taxonomy of tasks and methods. We break down the subjective notion of numeracy into 7 subtasks, arranged along two dimensions: granularity (exact vs approximate) and units (abstract vs grounded). We analyze the myriad representational choices made by 18 previously published number encoders and decoders. We synthesize best practices for representing numbers in text and articulate a vision for holistic numeracy in NLP, comprised of design trade-offs and a unified evaluation.

Filip Ilievski, Alessandro Oltramari, Kaixin Ma et al.

Commonsense knowledge is essential for many AI applications, including those in natural language processing, visual processing, and planning. Consequently, many sources that include commonsense knowledge have been designed and constructed over the past decades. Recently, the focus has been on large text-based sources, which facilitate easier integration with neural (language) models and application to textual tasks, typically at the expense of the semantics of the sources and their harmonization. Efforts to consolidate commonsense knowledge have yielded partial success, with no clear path towards a comprehensive solution. We aim to organize these sources around a common set of dimensions of commonsense knowledge. We survey a wide range of popular commonsense sources with a special focus on their relations. We consolidate these relations into 13 knowledge dimensions. This consolidation allows us to unify the separate sources and to compute indications of their coverage, overlap, and gaps with respect to the knowledge dimensions. Moreover, we analyze the impact of each dimension on downstream reasoning tasks that require commonsense knowledge, observing that the temporal and desire/goal dimensions are very beneficial for reasoning on current downstream tasks, while distinctness and lexical knowledge have little impact. These results reveal preferences for some dimensions in current evaluation, and potential neglect of others.

Filip Ilievski, Pedro Szekely, Bin Zhang

Sources of commonsense knowledge support applications in natural language understanding, computer vision, and knowledge graphs. Given their complementarity, their integration is desired. Yet, their different foci, modeling approaches, and sparse overlap make integration difficult. In this paper, we consolidate commonsense knowledge by following five principles, which we apply to combine seven key sources into a first integrated CommonSense Knowledge Graph (CSKG). We analyze CSKG and its various text and graph embeddings, showing that CSKG is well-connected and that its embeddings provide a useful entry point to the graph. We demonstrate how CSKG can provide evidence for generalizable downstream reasoning and for pre-training of language models. CSKG and all its embeddings are made publicly available to support further research on commonsense knowledge integration and reasoning.

Filip Ilievski, Pedro Szekely, Daniel Schwabe

Wikidata and Wikipedia have been proven useful for reason-ing in natural language applications, like question answering or entitylinking. Yet, no existing work has studied the potential of Wikidata for commonsense reasoning. This paper investigates whether Wikidata con-tains commonsense knowledge which is complementary to existing commonsense sources. Starting from a definition of common sense, we devise three guiding principles, and apply them to generate a commonsense subgraph of Wikidata (Wikidata-CS). Within our approach, we map the relations of Wikidata to ConceptNet, which we also leverage to integrate Wikidata-CS into an existing consolidated commonsense graph. Our experiments reveal that: 1) albeit Wikidata-CS represents a small portion of Wikidata, it is an indicator that Wikidata contains relevant commonsense knowledge, which can be mapped to 15 ConceptNet relations; 2) the overlap between Wikidata-CS and other commonsense sources is low, motivating the value of knowledge integration; 3) Wikidata-CS has been evolving over time at a slightly slower rate compared to the overall Wikidata, indicating a possible lack of focus on commonsense knowledge. Based on these findings, we propose three recommended actions to improve the coverage and quality of Wikidata-CS further.

Filip Ilievski, Pedro Szekely, Jingwei Cheng et al.

Commonsense reasoning is an important aspect of building robust AI systems and is receiving significant attention in the natural language understanding, computer vision, and knowledge graphs communities. At present, a number of valuable commonsense knowledge sources exist, with different foci, strengths, and weaknesses. In this paper, we list representative sources and their properties. Based on this survey, we propose principles and a representation model in order to consolidate them into a Common Sense Knowledge Graph (CSKG). We apply this approach to consolidate seven separate sources into a first integrated CSKG. We present statistics of CSKG, present initial investigations of its utility on four QA datasets, and list learned lessons.

Filip Ilievski, Daniel Garijo, Hans Chalupsky et al.

Knowledge graphs (KGs) have become the preferred technology for representing, sharing and adding knowledge to modern AI applications. While KGs have become a mainstream technology, the RDF/SPARQL-centric toolset for operating with them at scale is heterogeneous, difficult to integrate and only covers a subset of the operations that are commonly needed in data science applications. In this paper we present KGTK, a data science-centric toolkit designed to represent, create, transform, enhance and analyze KGs. KGTK represents graphs in tables and leverages popular libraries developed for data science applications, enabling a wide audience of developers to easily construct knowledge graph pipelines for their applications. We illustrate the framework with real-world scenarios where we have used KGTK to integrate and manipulate large KGs, such as Wikidata, DBpedia and ConceptNet.

Peifeng Wang, Nanyun Peng, Filip Ilievski et al.

Commonsense question answering (QA) requires background knowledge which is not explicitly stated in a given context. Prior works use commonsense knowledge graphs (KGs) to obtain this knowledge for reasoning. However, relying entirely on these KGs may not suffice, considering their limited coverage and the contextual dependence of their knowledge. In this paper, we augment a general commonsense QA framework with a knowledgeable path generator. By extrapolating over existing paths in a KG with a state-of-the-art language model, our generator learns to connect a pair of entities in text with a dynamic, and potentially novel, multi-hop relational path. Such paths can provide structured evidence for solving commonsense questions without fine-tuning the path generator. Experiments on two datasets show the superiority of our method over previous works which fully rely on knowledge from KGs (with up to 6% improvement in accuracy), across various amounts of training data. Further evaluation suggests that the generated paths are typically interpretable, novel, and relevant to the task.

Majid Ghasemi-Gol, Pedro Szekely

There are hundreds of millions of tables in Web pages that contain useful information for many applications. Leveraging data within these tables is difficult because of the wide variety of structures, formats and data encoded in these tables. TabVec is an unsupervised method to embed tables into a vector space to support classification of tables into categories (entity, relational, matrix, list, and non-data) with minimal user intervention. TabVec deploys syntax and semantics of table cells, and embeds the structure of tables in a table vector space. This enables superior classification of tables even in the absence of domain annotations. Our evaluations in four real world domains show that TabVec improves classification accuracy by more than 20% compared to three state of the art systems, and that those systems require significant in domain training to achieve good results.

Mayank Kejriwal, Pedro Szekely

We propose a supervised algorithm for generating type embeddings in the same semantic vector space as a given set of entity embeddings. The algorithm is agnostic to the derivation of the underlying entity embeddings. It does not require any manual feature engineering, generalizes well to hundreds of types and achieves near-linear scaling on Big Graphs containing many millions of triples and instances by virtue of an incremental execution. We demonstrate the utility of the embeddings on a type recommendation task, outperforming a non-parametric feature-agnostic baseline while achieving 15x speedup and near-constant memory usage on a full partition of DBpedia. Using state-of-the-art visualization, we illustrate the agreement of our extensionally derived DBpedia type embeddings with the manually curated domain ontology. Finally, we use the embeddings to probabilistically cluster about 4 million DBpedia instances into 415 types in the DBpedia ontology.

Mayank Kejriwal, Pedro Szekely

Extracting useful entities and attribute values from illicit domains such as human trafficking is a challenging problem with the potential for widespread social impact. Such domains employ atypical language models, have `long tails' and suffer from the problem of concept drift. In this paper, we propose a lightweight, feature-agnostic Information Extraction (IE) paradigm specifically designed for such domains. Our approach uses raw, unlabeled text from an initial corpus, and a few (12-120) seed annotations per domain-specific attribute, to learn robust IE models for unobserved pages and websites. Empirically, we demonstrate that our approach can outperform feature-centric Conditional Random Field baselines by over 18\% F-Measure on five annotated sets of real-world human trafficking datasets in both low-supervision and high-supervision settings. We also show that our approach is demonstrably robust to concept drift, and can be efficiently bootstrapped even in a serial computing environment.

Mohsen Taheriyan, Craig A. Knoblock, Pedro Szekely et al.

Information sources such as relational databases, spreadsheets, XML, JSON, and Web APIs contain a tremendous amount of structured data that can be leveraged to build and augment knowledge graphs. However, they rarely provide a semantic model to describe their contents. Semantic models of data sources represent the implicit meaning of the data by specifying the concepts and the relationships within the data. Such models are the key ingredients to automatically publish the data into knowledge graphs. Manually modeling the semantics of data sources requires significant effort and expertise, and although desirable, building these models automatically is a challenging problem. Most of the related work focuses on semantic annotation of the data fields (source attributes). However, constructing a semantic model that explicitly describes the relationships between the attributes in addition to their semantic types is critical. We present a novel approach that exploits the knowledge from a domain ontology and the semantic models of previously modeled sources to automatically learn a rich semantic model for a new source. This model represents the semantics of the new source in terms of the concepts and relationships defined by the domain ontology. Given some sample data from the new source, we leverage the knowledge in the domain ontology and the known semantic models to construct a weighted graph that represents the space of plausible semantic models for the new source. Then, we compute the top k candidate semantic models and suggest to the user a ranked list of the semantic models for the new source. The approach takes into account user corrections to learn more accurate semantic models on future data sources. Our evaluation shows that our method generates expressive semantic models for data sources and services with minimal user input. ...

Reza Nourjou, Michinori Hatayama, Stephen F. Smith et al.

Problem: This paper addresses the design of an intelligent software system for the IC (incident commander) of a team in order to coordinate actions of agents (field units or robots) in the domain of emergency/crisis response operations. Objective: This paper proposes GICoordinator. It is a GIS-based assistant software agent that assists and collaborates with the human planner in strategic planning and macro tasks assignment for centralized multi-agent coordination. Method: Our approach to design GICoordinator was to: analyze the problem, design a complete data model, design an architecture of GICoordinator, specify required capabilities of human and system in coordination problem solving, specify development tools, and deploy. Result: The result was an architecture/design of GICoordinator that contains system requirements. Findings: GICoordinator efficiently integrates geoinformatics with artifice intelligent techniques in order to provide a spatial intelligent coordinator system for an IC to efficiently coordinate and control agents by making macro/strategic decisions. Results define a framework for future works to develop this system.