Introduction to Computational Semantics

Aims

This course is an overview course of theoretical semantics and computational semantics. It covers truth-conditional logic-based formalisms and graph-based meaning representations, including the principle of semantic compositionality and scope. Based on these representations for the meaning of individual sentences, the course explores a range of semantic phenomena that extend beyond the sentence level. These include negation, some aspects of pragmatics, and cross-modal representations of meaning. The course also offers some psycholinguistic insights into how semantics is represented in the human mind, e.g., in early language acquisition and when language undergoes changes in a linguistic community.

This course is relevant in the light of interpretation of LLM output. LLMs are able to produce fluent and thematically relevant output. In order to assess whether the output is also semantically appropriate, students profit from a close study of the semantic properties of human language. Amongst other skills, students will learn to distinguish relevant semantic and pragmatic phenomena, find ambiguity in text and identify cases of distorted meaning. They will also learn how we can identify how meaning changes over time in a linguistic community. Knowledge of these aspects of semantics of language will provide a solid foundation for meaningful evaluation of modern LLM and other automatically produced text.
 

Objectives

• Give an operational definition of what is meant by “meaning” (for instance, above and beyond syntax);
• Name the types of phenomena in language that require semantic consideration, in terms of lexical, compositional and discourse/pragmatic aspects, in other words, argue why semantics is important;
• Demonstrate an understanding of the basics of various semantic representations, including logic-based and graph-based semantic representations, their properties, how they are used and why they are important, and how they are different from syntactic representations;
• Know how such semantic representations are derived during or after parsing, and how they can be analysed and mapped to surface strings;
• Understand applications of semantic representations e.g. reasoning, validation, and methods how these are approached.
• When designing NL tasks that clearly require semantic processing (e.g. knowledge-based QA), to be aware of and reuse semantic representations and algorithms when designing the task, rather than reinventing the wheel.

Practical advantages of this course for NLP students

• Knowledge of underlying semantic effects helps improve NLP evaluation, for instance by providing more meaningful error analysis. You will be able to link particular errors to design decisions inside your system.
• You will learn methods for better benchmarking of your system, whatever the task may be. Supervised ML systems (in particular black-box systems such as Deep Learning) are only as clever as the datasets they are based on. In this course, you will learn to design datasets so that they are harder to trick without real understanding, or critique existing datasets.
• You will be able to design tests for ML systems that better pinpoint which aspects of language an end-to-end system has “understood”.
• You will learn to detect ambiguity and ill-formed semantics in human-human communication. This can serve to write more clearly and logically.
• You will learn about decomposing complex semantics-reliant tasks sensibly so that you can reuse the techniques underlying semantic analyzers in a modular way. In this way, rather than being forced to treat complex tasks in an end-to-end manner, you will be able to profit from partial explanations and a better error analysis already built into the system.

Syllabus

1. Overview of the course
2. Events and semantic role labelling
3. Referentiality and coreference resolution
4. Truth-conditional semantics
5. Graph-based meaning representation
6. Compositional semantics
7. Context-free Graph Rewriting
8. Surface realisation
9. Negation and psychological approach to semantics
10. Dynamic semantics
11. Gricean pragmatics
12. Vector space models
13. Cross-modality
14. Acquisition of semantics
15. Diachronic change of semantics
16. Summary of the course

Pre-registration evaluation form 

Students who are interested in taking this module will be asked to complete a pre-registration evaluation form to determine whether they have sufficient background to take the module. 

This will take place during the week Monday 11 August - Friday 15 August. 

Assessment

The course will be assessed through two take-home assignments (each contributing 20% to the final grade) and one in-class, closed-book test (worth 60%).

The test will check basic knowledge of concepts taught such as terminology. Students may be asked to provide their own examples of linguistic phenomena, interpret given examples, and carry out mathematical derivations. Questions may also relate to assigned readings and may be presented in multiple-choice format or require longer written responses. The in-class test will include research-oriented questions and the students will be assigned cutting-edge research papers before the test.

To help students prepare for the exam, each lecture will highlight possible exam questions in the form of core 'take-home' knowledge and thinking-further questions which students can solve in their own time after the lecture. This will give students a good idea about which material is examinable.