<> "The repository administrator has not yet configured an RDF license."^^ . <> . . "Response-Based and Counterfactual Learning for Sequence-to-Sequence Tasks in NLP"^^ . "Many applications nowadays rely on statistical machine-learnt models, such as a rising\r\nnumber of virtual personal assistants. To train statistical models, typically large amounts\r\nof labelled data are required which are expensive and difficult to obtain. In this thesis, we\r\ninvestigate two approaches that alleviate the need for labelled data by leveraging feedback to model outputs instead. Both scenarios are applied to two sequence-to-sequence\r\ntasks for Natural Language Processing (NLP): machine translation and semantic parsing\r\nfor question-answering. Additionally, we define a new question-answering task based on\r\nthe geographical database OpenStreetMap (OSM) and collect a corpus, NLmaps v2, with\r\n28,609 question-parse pairs. With the corpus, we build semantic parsers for subsequent experiments. Furthermore, we are the first to design a natural language interface to OSM, for\r\nwhich we specifically tailor a parser.\r\nThe first approach to learn from feedback given to model outputs, considers a scenario\r\nwhere weak supervision is available by grounding the model in a downstream task for\r\nwhich labelled data has been collected. Feedback obtained from the downstream task is\r\nused to improve the model in a response-based on-policy learning setup. We apply this\r\napproach to improve a machine translation system, which is grounded in a multilingual\r\nsemantic parsing task, by employing ramp loss objectives. Next, we improve a neural semantic parser where only gold answers, but not gold parses, are available, by lifting ramp\r\nloss objectives to non-linear neural networks. In the second approach to learn from feedback, instead of collecting expensive labelled data, a model is deployed and user-model\r\ninteractions are recorded in a log. This log is used to improve a model in a counterfactual\r\noff-policy learning setup. We first exemplify this approach on a domain adaptation task for\r\nmachine translation. Here, we show that counterfactual learning can be applied to tasks\r\nwith large output spaces and, in contrast to prevalent theory, deterministic logs can successfully be used on sequence-to-sequence tasks for NLP. Next, we demonstrate on a semantic parsing task that counterfactual learning can also be applied when the underlying\r\nmodel is a neural network and feedback is collected from human users. Applying both approaches to the same semantic parsing task, allows us to draw a direct comparison between\r\nthem. Response-based on-policy learning outperforms counterfactual off-policy learning,\r\nbut requires expensive labelled data for the downstream task, whereas interaction logs for\r\ncounterfactual learning can be easier to obtain in various scenarios."^^ . "2019" . . . . . . . "Carolin"^^ . "Lawrence"^^ . "Carolin Lawrence"^^ . . . . . . "Response-Based and Counterfactual Learning for Sequence-to-Sequence Tasks in NLP (PDF)"^^ . . . "20190510_Thesis_Carolin.pdf"^^ . . . "Response-Based and Counterfactual Learning for Sequence-to-Sequence Tasks in NLP (Other)"^^ . . . . . . "indexcodes.txt"^^ . . "HTML Summary of #26477 \n\nResponse-Based and Counterfactual Learning for Sequence-to-Sequence Tasks in NLP\n\n" . "text/html" . . . "000 Allgemeines, Wissenschaft, Informatik"@de . "000 Generalities, Science"@en . . . "004 Informatik"@de . "004 Data processing Computer science"@en . . . "310 Statistik"@de . "310 General statistics"@en . . . "400 Sprachwissenschaft"@de . "400 Linguistics"@en . . . "420 Englisch"@de . "420 English"@en . . . "490 Andere Sprachen"@de . "490 Other languages"@en . . . "500 Naturwissenschaften und Mathematik"@de . "500 Natural sciences and mathematics"@en . .