We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering and language inference, without substantial task-specific architecture modifications. 
BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural language processing tasks, including pushing the GLUE score to 80.5% (7.7% point absolute improvement), MultiNLI accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models (Peters et al., 2018a; Radford et al., 2018), BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering and language inference, without substantial task-specific architecture modifications. BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural language processing tasks, including pushing the GLUE score to 80.5 (7.7 point absolute improvement), MultiNLI accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).

Data Mining Practical Machine Learning Tools and Techniques

How to Do Things With Words

Transfer learning, where a model is first pre-trained on a data-rich task before being fine-tuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts all text-based language problems into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled data sets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new ``Colossal Clean Crawled Corpus'', we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our data set, pre-trained models, and code.

Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

Semantic Textual Similarity (STS) measures the meaning similarity of sentences. Applications include machine translation (MT), summarization, generation, question answering (QA), short answer grading, semantic search, dialog and conversational systems. The STS shared task is a venue for assessing the current state-of-the-art. The 2017 task focuses on multilingual and cross-lingual pairs with one sub-track exploring MT quality estimation (MTQE) data. The task obtained strong participation from 31 teams, with 17 participating in all language tracks. We summarize performance and review a selection of well performing methods. Analysis highlights common errors, providing insight into the limitations of existing models. To support ongoing work on semantic representations, the STS Benchmark is introduced as a new shared training and evaluation set carefully selected from the corpus of English STS shared task data (2012-2017).

/pdf/semeval-2017-task-1-semantic-textual-similarity-multilingual-1su3vk2wkg.pdf

SemEval-2017 Task 1: Semantic Textual Similarity - Multilingual and Cross-lingual Focused Evaluation

/pdf/semeval-2017-task-1-semantic-textual-similarity-multilingual-51ojcf0apx.pdf

SemEval-2017 Task 1: Semantic Textual Similarity Multilingual and Crosslingual Focused Evaluation

Semantic Textual Similarity (STS) measures the degree of semantic equivalence between two texts. This paper presents the results of the STS pilot task in Semeval. The training data contained 2000 sentence pairs from previously existing paraphrase datasets and machine translation evaluation resources. The test data also comprised 2000 sentences pairs for those datasets, plus two surprise datasets with 400 pairs from a different machine translation evaluation corpus and 750 pairs from a lexical resource mapping exercise. The similarity of pairs of sentences was rated on a 0-5 scale (low to high similarity) by human judges using Amazon Mechanical Turk, with high Pearson correlation scores, around 90%. 35 teams participated in the task, submitting 88 runs. The best results scored a Pearson correlation >80%, well above a simple lexical baseline that only scored a 31% correlation. This pilot task opens an exciting way ahead, although there are still open issues, specially the evaluation metric.

/pdf/semeval-2012-task-6-a-pilot-on-semantic-textual-similarity-nphj9ju4gz.pdf

SemEval-2012 Task 6: A Pilot on Semantic Textual Similarity

In this paper, we present MADAMIRA, a system for morphological analysis and disambiguation of Arabic that combines some of the best aspects of two previously commonly used systems for Arabic processing, MADA (Habash and Rambow, 2005; Habash et al., 2009; Habash et al., 2013) and AMIRA (Diab et al., 2007). MADAMIRA improves upon the two systems with a more streamlined Java implementation that is more robust, portable, extensible, and is faster than its ancestors by more than an order of magnitude. We also discuss an online demo (see http://nlp.ldeo.columbia.edu/madamira/) that highlights these aspects.

/pdf/madamira-a-fast-comprehensive-tool-for-morphological-pt9qsrhep1.pdf

MADAMIRA: A Fast, Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic

In Semantic Textual Similarity (STS), systems rate the degree of semantic equivalence, on a graded scale from 0 to 5, with 5 being the most similar. This year we set up two tasks: (i) a core task (CORE), and (ii) a typed-similarity task (TYPED). CORE is similar in set up to SemEval STS 2012 task with pairs of sentences from sources related to those of 2012, yet different in genre from the 2012 set, namely, this year we included newswire headlines, machine translation evaluation datasets and multiple lexical resource glossed sets. TYPED, on the other hand, is novel and tries to characterize why two items are deemed similar, using cultural heritage items which are described with metadata such as title, author or description. Several types of similarity have been defined, including similar author, similar time period or similar location. The annotation for both tasks leverages crowdsourcing, with relative high interannotator correlation, ranging from 62% to 87%. The CORE task attracted 34 participants with 89 runs, and the TYPED task attracted 6 teams with 14 runs.

Mona Diab

Papers

SemEval-2017 Task 1: Semantic Textual Similarity - Multilingual and Cross-lingual Focused Evaluation

SemEval-2017 Task 1: Semantic Textual Similarity Multilingual and Crosslingual Focused Evaluation

SemEval-2012 Task 6: A Pilot on Semantic Textual Similarity

MADAMIRA: A Fast, Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic

*SEM 2013 shared task: Semantic Textual Similarity