women@CL Talklets - NLIP

Add to your list(s) Download to your calendar using vCal

• Computer Laboratory, University of Cambridge
• Thursday 12 February 2015, 13:00-14:00
• Computer Laboratory, William Gates Building, Room FW26.

If you have a question about this talk, please contact Ekaterina Kochmar.

Title: Learning from Text Whether a Lion is an Animal

Speaker: Laura Rimell

Abstract: The ability to detect entailment relations between natural language sentences is a crucial Artificial Intelligence task. A prerequisite is the ability to automatically identify entailment relations between words: for example, being a lion entails being an animal, and kicking something entails touching it. This talk will discuss how such relations can be learned in an unsupervised way from large amounts of text. We make use of distributional semantics, the idea that a word’s contexts of use characterise its meaning. The talk will begin with an introduction to distributional semantics and describe how the contexts of words like “lion” and “animal” relate to each other in surprising ways.

Title: Words as Functions: Composing Meanings

Speaker: Tamara Polajnar

Abstract: Although we are able to adequately, if not perfectly, represent the meaning of words in a way that a computer can process, longer constructions such as phrases and sentences offer new challenges. In this talk, I will describe how we take into account grammatical structure in order to encode the meaning of simple sentences in a vector space. There are many tasks where the word order, grammatical structure, and semantics are key. For example, we may want to be able to distinguish that in a sentence like “Animals eat plants”, the animals are the animate subjects with teeth doing the eating and not the other way around. Likewise, we may be concerned about plausibility and want to reject sentences like “Animals eat planets”.

Title: Error Detection in Non-Native English Writing

Speaker: Ekaterina Kochmar

Abstract: Even though English is used as the international language in many fields, the majority of people who use it are not native speakers of English. As a result, they inevitably commit errors. Errors in the use of content words (for example, adjectives, nouns and verbs) can change the meaning of the original expression or even result in implausible word combinations. For example, if non-native speakers write “Animals eat planets” they probably mean “plants“, and when they write “big history“ they ptobably mean “long history“. This talk will discuss error detection and correction in the use of content words by non-native speakers of English. We model the meaning of the words computationally using the methods of compostional and distributional semantics overviewed in the previous talks. The implemented error detection and correction algorithm can further be integrated in a self-assessment and tutoring system.

Title: Write & Improve!

Speaker: Helen Yannakoudakis

Abstract: The task of automated assessment (AA) of text focuses on automatically analysing and assessing the quality of writing. A number of AA systems have been developed, aimed particularly at second-language learners and providing feedback on their writing. Automated writing feedback may be a useful complement to teacher comments in the process of learning a foreign language, while recent studies have shown that automated writing evaluation can lead to increased learner autonomy and higher writing accuracy. In this talk, we will discuss how we can leverage natural language processing and machine learning techniques to develop an accurate self-assessment and tutoring system efficiently. We will describe Write & Improve, a system that provides automated feedback on learners’ writing at three different levels of granularity: 1) an overall assessment of their proficiency, 2) a qualitative assessment of each individual sentence, making the language learner aware of potentially problematic areas rather than providing a panacea, and 3) diagnostic feedback on local issues including spelling and word choice. Pedagogically useful feedback requires high precision (few false positives in the detection of errors) and reasonable recall (not too many errors undetected), which makes this a challenging task.

This talk is part of the Women@CL Events series.

This talk is included in these lists:

• All Talks (aka the CURE list)
• Cambridge Centre for Data-Driven Discovery (C2D3)
• Cambridge talks
• Chris Davis' list
• Computer Laboratory, William Gates Building, Room FW26
• Department of Computer Science and Technology talks and seminars
• Guy Emerson's list
• Interested Talks
• School of Technology
• Trust & Technology Initiative - interesting events
• Women@CL Events
• bld31
• ndk22's list
• ob366-ai4er
• rp587
• women@CL all

Note that ex-directory lists are not shown.