Despite LLMs’ recent advancements, they still suffer from factual inconsistency and hallucination. An often-opted remedy is retrieval-augmented generation – however, there is no guarantee that the model will strictly adhere to retrieved grounding. Fundamentally, LLMs need to be aligned to be more faithful to grounding, which will require high-quality preference annotations. This paper investigates whether we can create high-quality grounded preference data for model alignment without using annotations from humans or large proprietary models. We experimented with existing entailment data and proposed approaches to generate synthetic grounded preference data, with which we train a Grounded Preference Model(GPM). We demonstrate through Proximal Policy Optimization(PPO) training of Mistral-7B-Instruct that our GPM model can successfully align powerful LLMs to generate much better grounded responses as judged by GPT4. Moreover, we show that our GPM is also a great faithfulness classifier, achieving SoTA in dialogue sub-tasks of the TRUE faithfulness Benchmark. We will release our GPM under the Apache 2.0 license.
Neural information retrieval (IR) systems have progressed rapidly in recent years, in large part due to the release of publicly available benchmarking tasks. Unfortunately, some dimensions of this progress are illusory: the majority of the popular IR benchmarks today focus exclusively on downstream task accuracy and thus conceal the costs incurred by systems that trade away efficiency for quality. Latency, hardware cost, and other efficiency considerations are paramount to the deployment of IR systems in user-facing settings. We propose that IR benchmarks structure their evaluation methodology to include not only metrics of accuracy, but also efficiency considerations such as a query latency and the corresponding cost budget for a reproducible hardware setting. For the popular IR benchmarks MS MARCO and XOR-TyDi, we show how the best choice of IR system varies according to how these efficiency considerations are chosen and weighed. We hope that future benchmarks will adopt these guidelines toward more holistic IR evaluation.
Using in-context learning (ICL) for data generation, techniques such as Self-Instruct (Wang et al., 2023) or the follow-up Alpaca (Taori et al., 2023) can train strong conversational agents with only a small amount of human supervision. One limitation of these approaches is that they resort to very large language models (around 175B parameters) that are also proprietary and non-public. Here we explore the application of such techniques to language models that are much smaller (around 10B–40B parameters) and have permissive licenses. We find the Self-Instruct approach to be less effective at these sizes and propose new ICL methods that draw on two main ideas: (a) categorization and simplification of the ICL templates to make prompt learning easier for the LM, and (b) ensembling over multiple LM outputs to help select high-quality synthetic examples. Our algorithm leverages the 175 Self-Instruct seed tasks and employs separate pipelines for instructions that require an input and instructions that do not. Empirical investigations with different LMs show that: (1) Our proposed method yields higher-quality instruction tuning data than Self-Instruct, (2) It improves performances of both vanilla and instruction-tuned LMs by significant margins, and (3) Smaller instruction-tuned LMs generate more useful examples than their larger un-tuned counterparts.
Many information retrieval tasks require large labeled datasets for fine-tuning. However, such datasets are often unavailable, and their utility for real-world applications can diminish quickly due to domain shifts. To address this challenge, we develop and motivate a method for using large language models (LLMs) to generate large numbers of synthetic queries cheaply. The method begins by generating a small number of synthetic queries using an expensive LLM. After that, a much less expensive one is used to create large numbers of synthetic queries, which are used to fine-tune a family of reranker models. These rerankers are then distilled into a single efficient retriever for use in the target domain. We show that this technique boosts zero-shot accuracy in long-tail domains and achieves substantially lower latency than standard reranking methods.
The field of Question Answering (QA) has made remarkable progress in recent years, thanks to the advent of large pre-trained language models, newer realistic benchmark datasets with leaderboards, and novel algorithms for key components such as retrievers and readers. In this paper, we introduce PrimeQA: a one-stop and open-source QA repository with an aim to democratize QA research and facilitate easy replication of state-of-the-art (SOTA) QA methods. PrimeQA supports core QA functionalities like retrieval and reading comprehension as well as auxiliary capabilities such as question generation. It has been designed as an end-to-end toolkit for various use cases: building front-end applications, replicating SOTA methods on public benchmarks, and expanding pre-existing methods. PrimeQA is available at: https://github.com/primeqa.
Machine learning models are prone to overfitting their training (source) domains, which is commonly believed to be the reason why they falter in novel target domains. Here we examine the contrasting view that multi-source domain generalization (DG) is first and foremost a problem of mitigating source domain underfitting: models not adequately learning the signal already present in their multi-domain training data. Experiments on a reading comprehension DG benchmark show that as a model learns its source domains better—using familiar methods such as knowledge distillation (KD) from a bigger model—its zero-shot out-of-domain utility improves at an even faster pace. Improved source domain learning also demonstrates superior out-of-domain generalization over three popular existing DG approaches that aim to limit overfitting. Our implementation of KD-based domain generalization is available via PrimeQA at: https://ibm.biz/domain-generalization-with-kd.
Transition-based parsers for Abstract Meaning Representation (AMR) rely on node-to-word alignments. These alignments are learned separately from parser training and require a complex pipeline of rule-based components, pre-processing, and post-processing to satisfy domain-specific constraints. Parsers also train on a point-estimate of the alignment pipeline, neglecting the uncertainty due to the inherent ambiguity of alignment. In this work we explore two avenues for overcoming these limitations. First, we propose a neural aligner for AMR that learns node-to-word alignments without relying on complex pipelines. We subsequently explore a tighter integration of aligner and parser training by considering a distribution over oracle action sequences arising from aligner uncertainty. Empirical results show this approach leads to more accurate alignments and generalization better from the AMR2.0 to AMR3.0 corpora. We attain a new state-of-the art for gold-only trained models, matching silver-trained performance without the need for beam search on AMR3.0.
Despite extensive research on parsing of English sentences into Abstract Meaning Representation (AMR) graphs, which are compared to gold graphs via the Smatch metric, full-document parsing into a unified graph representation lacks well-defined representation and evaluation. Taking advantage of a super-sentential level of coreference annotation from previous work, we introduce a simple algorithm for deriving a unified graph representation, avoiding the pitfalls of information loss from over-merging and lack of coherence from under merging. Next, we describe improvements to the Smatch metric to make it tractable for comparing document-level graphs and use it to re-evaluate the best published document-level AMR parser. We also present a pipeline approach combining the top-performing AMR parser and coreference resolution systems, providing a strong baseline for future research.
AMR parsing has experienced an unprecendented increase in performance in the last three years, due to a mixture of effects including architecture improvements and transfer learning. Self-learning techniques have also played a role in pushing performance forward. However, for most recent high performant parsers, the effect of self-learning and silver data augmentation seems to be fading. In this paper we propose to overcome this diminishing returns of silver data by combining Smatch-based ensembling techniques with ensemble distillation. In an extensive experimental setup, we push single model English parser performance to a new state-of-the-art, 85.9 (AMR2.0) and 84.3 (AMR3.0), and return to substantial gains from silver data augmentation. We also attain a new state-of-the-art for cross-lingual AMR parsing for Chinese, German, Italian and Spanish. Finally we explore the impact of the proposed technique on domain adaptation, and show that it can produce gains rivaling those of human annotated data for QALD-9 and achieve a new state-of-the-art for BioAMR.
Abstract Meaning Representation parsing is a sentence-to-graph prediction task where target nodes are not explicitly aligned to sentence tokens. However, since graph nodes are semantically based on one or more sentence tokens, implicit alignments can be derived. Transition-based parsers operate over the sentence from left to right, capturing this inductive bias via alignments at the cost of limited expressiveness. In this work, we propose a transition-based system that combines hard-attention over sentences with a target-side action pointer mechanism to decouple source tokens from node representations and address alignments. We model the transitions as well as the pointer mechanism through straightforward modifications within a single Transformer architecture. Parser state and graph structure information are efficiently encoded using attention heads. We show that our action-pointer approach leads to increased expressiveness and attains large gains (+1.6 points) against the best transition-based AMR parser in very similar conditions. While using no graph re-categorization, our single model yields the second best Smatch score on AMR 2.0 (81.8), which is further improved to 83.4 with silver data and ensemble decoding.
We develop high performance multilingual Abstract Meaning Representation (AMR) systems by projecting English AMR annotations to other languages with weak supervision. We achieve this goal by bootstrapping transformer-based multilingual word embeddings, in particular those from cross-lingual RoBERTa (XLM-R large). We develop a novel technique for foreign-text-to-English AMR alignment, using the contextual word alignment between English and foreign language tokens. This word alignment is weakly supervised and relies on the contextualized XLM-R word embeddings. We achieve a highly competitive performance that surpasses the best published results for German, Italian, Spanish and Chinese.
In this paper, we present the event detection models and systems we have developed for Multilingual Protest News Detection - Shared Task 1 at CASE 2021. The shared task has 4 subtasks which cover event detection at different granularity levels (from document level to token level) and across multiple languages (English, Hindi, Portuguese and Spanish). To handle data from multiple languages, we use a multilingual transformer-based language model (XLM-R) as the input text encoder. We apply a variety of techniques and build several transformer-based models that perform consistently well across all the subtasks and languages. Our systems achieve an average F_1 score of 81.2. Out of thirteen subtask-language tracks, our submissions rank 1st in nine and 2nd in four tracks.
Supervised models can achieve very high accuracy for fine-grained text classification. In practice, however, training data may be abundant for some types but scarce or even non-existent for others. We propose a hybrid architecture that uses as much labeled data as available for fine-tuning classification models, while also allowing for types with little (few-shot) or no (zero-shot) labeled data. In particular, we pair a supervised text classification model with a Natural Language Inference (NLI) reranking model. The NLI reranker uses a textual representation of target types that allows it to score the strength with which a type is implied by a text, without requiring training data for the types. Experiments show that the NLI model is very sensitive to the choice of textual representation, but can be effective for classifying unseen types. It can also improve classification accuracy for the known types of an already highly accurate supervised model.
Predicting linearized Abstract Meaning Representation (AMR) graphs using pre-trained sequence-to-sequence Transformer models has recently led to large improvements on AMR parsing benchmarks. These parsers are simple and avoid explicit modeling of structure but lack desirable properties such as graph well-formedness guarantees or built-in graph-sentence alignments. In this work we explore the integration of general pre-trained sequence-to-sequence language models and a structure-aware transition-based approach. We depart from a pointer-based transition system and propose a simplified transition set, designed to better exploit pre-trained language models for structured fine-tuning. We also explore modeling the parser state within the pre-trained encoder-decoder architecture and different vocabulary strategies for the same purpose. We provide a detailed comparison with recent progress in AMR parsing and show that the proposed parser retains the desirable properties of previous transition-based approaches, while being simpler and reaching the new parsing state of the art for AMR 2.0, without the need for graph re-categorization.
Existing models on Machine Reading Comprehension (MRC) require complex model architecture for effectively modeling long texts with paragraph representation and classification, thereby making inference computationally inefficient for production use. In this work, we propose VAULT: a light-weight and parallel-efficient paragraph representation for MRC based on contextualized representation from long document input, trained using a new Gaussian distribution-based objective that pays close attention to the partially correct instances that are close to the ground-truth. We validate our VAULT architecture showing experimental results on two benchmark MRC datasets that require long context modeling; one Wikipedia-based (Natural Questions (NQ)) and the other on TechNotes (TechQA). VAULT can achieve comparable performance on NQ with a state-of-the-art (SOTA) complex document modeling approach while being 16 times faster, demonstrating the efficiency of our proposed model. We also demonstrate that our model can also be effectively adapted to a completely different domain – TechQA – with large improvement over a model fine-tuned on a previously published large PLM.
Modeling the parser state is key to good performance in transition-based parsing. Recurrent Neural Networks considerably improved the performance of transition-based systems by modelling the global state, e.g. stack-LSTM parsers, or local state modeling of contextualized features, e.g. Bi-LSTM parsers. Given the success of Transformer architectures in recent parsing systems, this work explores modifications of the sequence-to-sequence Transformer architecture to model either global or local parser states in transition-based parsing. We show that modifications of the cross attention mechanism of the Transformer considerably strengthen performance both on dependency and Abstract Meaning Representation (AMR) parsing tasks, particularly for smaller models or limited training data.
Abstract Meaning Representation (AMR) parsing has experienced a notable growth in performance in the last two years, due both to the impact of transfer learning and the development of novel architectures specific to AMR. At the same time, self-learning techniques have helped push the performance boundaries of other natural language processing applications, such as machine translation or question answering. In this paper, we explore different ways in which trained models can be applied to improve AMR parsing performance, including generation of synthetic text and AMR annotations as well as refinement of actions oracle. We show that, without any additional human annotations, these techniques improve an already performant parser and achieve state-of-the-art results on AMR 1.0 and AMR 2.0.
Transfer learning is a popular technique to learn a task using less training data and fewer compute resources. However, selecting the correct source model for transfer learning is a challenging task. We demonstrate a novel predictive method that determines which existing source model would minimize error for transfer learning to a given target. This technique does not require learning for prediction, and avoids computational costs of trail-and-error. We have evaluated this technique on nine datasets across diverse domains, including newswire, user forums, air flight booking, cybersecurity news, etc. We show that it per-forms better than existing techniques such as fine-tuning over vanilla BERT, or curriculum learning over the largest dataset on top of BERT, resulting in average F1 score gains in excess of 3%. Moreover, our technique consistently selects the best model using fewer tries.
This paper presents M-GAAMA, a Multilingual Question Answering architecture and demo system. This is the first multilingual machine reading comprehension (MRC) demo which is able to answer questions in over 100 languages. M-GAAMA answers questions from a given passage in the same or different language. It incorporates several existing multilingual models that can be used interchangeably in the demo such as M-BERT and XLM-R. The M-GAAMA demo also improves language accessibility by incorporating the IBM Watson machine translation widget to provide additional capabilities to the user to see an answer in their desired language. We also show how M-GAAMA can be used in downstream tasks by incorporating it into an END-TO-END-QA system using CFO (Chakravarti et al., 2019). We experiment with our system architecture on the Multi-Lingual Question Answering (MLQA) and the COVID-19 CORD (Wang et al., 2020; Tang et al., 2020) datasets to provide insights into the performance of the system.
Industry-scale NLP systems necessitate two features. 1. Robustness: “zero-shot transfer learning” (ZSTL) performance has to be commendable and 2. Efficiency: systems have to train efficiently and respond instantaneously. In this paper, we introduce the development of a production model called GAAMA (Go Ahead Ask Me Anything) which possess the above two characteristics. For robustness, it trains on the recently introduced Natural Questions (NQ) dataset. NQ poses additional challenges over older datasets like SQuAD: (a) QA systems need to read and comprehend an entire Wikipedia article rather than a small passage, and (b) NQ does not suffer from observation bias during construction, resulting in less lexical overlap between the question and the article. GAAMA consists of Attention-over-Attention, diversity among attention heads, hierarchical transfer learning, and synthetic data augmentation while being computationally inexpensive. Building on top of the powerful BERTQA model, GAAMA provides a ∼2.0% absolute boost in F1 over the industry-scale state-of-the-art (SOTA) system on NQ. Further, we show that GAAMA transfers zero-shot to unseen real life and important domains as it yields respectable performance on two benchmarks: the BioASQ and the newly introduced CovidQA datasets.
We present scalable Universal Dependency (UD) treebank synthesis techniques that exploit advances in language representation modeling which leverage vast amounts of unlabeled general-purpose multilingual text. We introduce a data augmentation technique that uses synthetic treebanks to improve production-grade parsers. The synthetic treebanks are generated using a state-of-the-art biaffine parser adapted with pretrained Transformer models, such as Multilingual BERT (M-BERT). The new parser improves LAS by up to two points on seven languages. The production models’ LAS performance improves as the augmented treebanks scale in size, surpassing performance of production models trained on originally annotated UD treebanks.
We introduce TECHQA, a domain-adaptation question answering dataset for the technical support domain. The TECHQA corpus highlights two real-world issues from the automated customer support domain. First, it contains actual questions posed by users on a technical forum, rather than questions generated specifically for a competition or a task. Second, it has a real-world size – 600 training, 310 dev, and 490 evaluation question/answer pairs – thus reflecting the cost of creating large labeled datasets with actual data. Hence, TECHQA is meant to stimulate research in domain adaptation rather than as a resource to build QA systems from scratch. TECHQA was obtained by crawling the IBMDeveloper and DeveloperWorks forums for questions with accepted answers provided in an IBM Technote—a technical document that addresses a specific technical issue. We also release a collection of the 801,998 Technotes available on the web as of April 4, 2019 as a companion resource that can be used to learn representations of the IT domain language.
Abstract Meaning Representations (AMRs) are broad-coverage sentence-level semantic graphs. Existing approaches to generating text from AMR have focused on training sequence-to-sequence or graph-to-sequence models on AMR annotated data only. In this paper, we propose an alternative approach that combines a strong pre-trained language model with cycle consistency-based re-scoring. Despite the simplicity of the approach, our experimental results show these models outperform all previous techniques on the English LDC2017T10 dataset, including the recent use of transformer architectures. In addition to the standard evaluation metrics, we provide human evaluation experiments that further substantiate the strength of our approach.
Transfer learning techniques are particularly useful for NLP tasks where a sizable amount of high-quality annotated data is difficult to obtain. Current approaches directly adapt a pretrained language model (LM) on in-domain text before fine-tuning to downstream tasks. We show that extending the vocabulary of the LM with domain-specific terms leads to further gains. To a bigger effect, we utilize structure in the unlabeled data to create auxiliary synthetic tasks, which helps the LM transfer to downstream tasks. We apply these approaches incrementally on a pretrained Roberta-large LM and show considerable performance gain on three tasks in the IT domain: Extractive Reading Comprehension, Document Ranking and Duplicate Question Detection.
We introduce ARES (A Reading Comprehension Ensembling Service): a novel Machine Reading Comprehension (MRC) demonstration system which utilizes an ensemble of models to increase F1 by 2.3 points. While many of the top leaderboard submissions in popular MRC benchmarks such as the Stanford Question Answering Dataset (SQuAD) and Natural Questions (NQ) use model ensembles, the accompanying papers do not publish their ensembling strategies. In this work, we detail and evaluate various ensembling strategies using the NQ dataset. ARES leverages the CFO (Chakravarti et al., 2019) and ReactJS distributed frameworks to provide a scalable interactive Question Answering experience that capitalizes on the agreement (or lack thereof) between models to improve the answer visualization experience.
Our work involves enriching the Stack-LSTM transition-based AMR parser (Ballesteros and Al-Onaizan, 2017) by augmenting training with Policy Learning and rewarding the Smatch score of sampled graphs. In addition, we also combined several AMR-to-text alignments with an attention mechanism and we supplemented the parser with pre-processed concept identification, named entities and contextualized embeddings. We achieve a highly competitive performance that is comparable to the best published results. We show an in-depth study ablating each of the new components of the parser.
Relation extraction (RE) seeks to detect and classify semantic relationships between entities, which provides useful information for many NLP applications. Since the state-of-the-art RE models require large amounts of manually annotated data and language-specific resources to achieve high accuracy, it is very challenging to transfer an RE model of a resource-rich language to a resource-poor language. In this paper, we propose a new approach for cross-lingual RE model transfer based on bilingual word embedding mapping. It projects word embeddings from a target language to a source language, so that a well-trained source-language neural network RE model can be directly applied to the target language. Experiment results show that the proposed approach achieves very good performance for a number of target languages on both in-house and open datasets, using a small bilingual dictionary with only 1K word pairs.
This paper introduces a novel orchestration framework, called CFO (Computation Flow Orchestrator), for building, experimenting with, and deploying interactive NLP (Natural Language Processing) and IR (Information Retrieval) systems to production environments. We then demonstrate a question answering system built using this framework which incorporates state-of-the-art BERT based MRC (Machine Reading Com- prehension) with IR components to enable end-to-end answer retrieval. Results from the demo system are shown to be high quality in both academic and industry domain specific settings. Finally, we discuss best practices when (pre-)training BERT based MRC models for production systems. Screencast links: - Short video (< 3 min): http: //ibm.biz/gaama_demo - Supplementary long video (< 13 min): http://ibm.biz/gaama_cfo_demo
We propose an entity-centric neural crosslingual coreference model that builds on multi-lingual embeddings and language independent features. We perform both intrinsic and extrinsic evaluations of our model. In the intrinsic evaluation, we show that our model, when trained on English and tested on Chinese and Spanish, achieves competitive results to the models trained directly on Chinese and Spanish respectively. In the extrinsic evaluation, we show that our English model helps achieve superior entity linking accuracy on Chinese and Spanish test sets than the top 2015 TAC system without using any annotated data from Chinese or Spanish.
Slot Filling (SF) aims to extract the values of certain types of attributes (or slots, such as person:cities_of_residence) for a given entity from a large collection of source documents. In this paper we propose an effective DNN architecture for SF with the following new strategies: (1). Take a regularized dependency graph instead of a raw sentence as input to DNN, to compress the wide contexts between query and candidate filler; (2). Incorporate two attention mechanisms: local attention learned from query and candidate filler, and global attention learned from external knowledge bases, to guide the model to better select indicative contexts to determine slot type. Experiments show that this framework outperforms state-of-the-art on both relation extraction (16% absolute F-score gain) and slot filling validation for each individual system (up to 8.5% absolute F-score gain).
The state-of-the-art named entity recognition (NER) systems are supervised machine learning models that require large amounts of manually annotated data to achieve high accuracy. However, annotating NER data by human is expensive and time-consuming, and can be quite difficult for a new language. In this paper, we present two weakly supervised approaches for cross-lingual NER with no human annotation in a target language. The first approach is to create automatically labeled NER data for a target language via annotation projection on comparable corpora, where we develop a heuristic scheme that effectively selects good-quality projection-labeled data from noisy data. The second approach is to project distributed representations of words (word embeddings) from a target language to a source language, so that the source-language NER system can be applied to the target language without re-training. We also design two co-decoding schemes that effectively combine the outputs of the two projection-based approaches. We evaluate the performance of the proposed approaches on both in-house and open NER data for several target languages. The results show that the combined systems outperform three other weakly supervised approaches on the CoNLL data.
Answer sentence ranking and answer extraction are two key challenges in question answering that have traditionally been treated in isolation, i.e., as independent tasks. In this article, we (1) explain how both tasks are related at their core by a common quantity, and (2) propose a simple and intuitive joint probabilistic model that addresses both via joint computation but task-specific application of that quantity. In our experiments with two TREC datasets, our joint model substantially outperforms state-of-the-art systems in both tasks.