The thematic fit estimation task measures the compatibility between a predicate (typically a verb), an argument (typically a noun phrase), and a specific semantic role assigned to the argument. Previous state-of-the-art work has focused on modeling thematic fit through distributional or neural models of event representation, trained in a supervised fashion with indirect labels. In this work, we assess whether pre-trained autoregressive LLMs possess consistent, expressible knowledge about thematic fit. We evaluate both closed and open state-of-the-art LLMs on several psycholinguistic datasets, along three axes: (1) Reasoning Form: multi-step logical reasoning (chain-of-thought prompting) vs. simple prompting. (2) Input Form: providing context (generated sentences) vs. raw tuples . (3) Output Form: categorical vs. numeric. Our results show that chain-of-thought reasoning is more effective on datasets with self-explanatory semantic role labels, especially Location. Generated sentences helped only in few settings, and lowered results in many others. Predefined categorical (compared to numeric) output raised GPT's results across the board with few exceptions, but lowered Llama's. We saw that semantically incoherent generated sentences, which the models lack the ability to consistently filter out, hurt reasoning and overall performance too. Our GPT-powered methods set new state-of-the-art on all tested datasets.

Observing that for certain NLP tasks, such as semantic role prediction or thematic fit estimation, random embeddings perform as well as pretrained embeddings, we explore what settings allow for this and examine where most of the learning is encoded: the word embeddings, the semantic role embeddings, or ``the network''. We find nuanced answers, depending on the task and its relation to the training objective. We examine these representation learning aspects in multi-task learning, where role prediction and role-filling are supervised tasks, while several thematic fit tasks are outside the models' direct supervision. We observe a non-monotonous relation between some tasks' quality score and the training data size. In order to better understand this observation, we analyze these results using easier, per-verb versions of these tasks.

Andrew McRae, a Ph.D. student in the School of Electrical and Computer Engineering (ECE) has been recognized with the Colonel Oscar P. Cleaver Award for a second time. McRae is just one of two students to be a double Cleaver Award recipient. "A Ph.D. program is long and tedious and often seems futile. This award is a welcome reminder of how much the work and the graduate experience are worth, just as I hope the degree is for all Ph.D. students. This motivates me to work diligently and with the best of my ability as I study machine learning and as I pursue my future career," said McRae.

There's a lot of buzz about artificial intelligence being a game-changer for the telecom industry, but machine learning is paving the way in the short term. BT's Neil McRae, chief architect, said there are elements of artificial intelligence in today's machine learning, and that machine learning could enable more, and deeper automation in networks. "BT Labs have been doing some work with Cambridge University leveraging both machine learning and artificial intelligence to improve our ability to react to events on the network," McRae said. "The initial findings are very promising. Networks are more complex, customers demand more and more from the network and I want to ensure that the network is the strongest part of our customers supply change. Today though, more often than not, humans are the reason for problems in the network. Using machine learning to let the network learn rather than scripting that automation will accelerate automation and I believe will bring benefits quicker. "When you actually look at what artificial intelligence is and what you need to do to deploy and use it in a network function, that's not a trivial thing, but using approaches such as machine learning are going to be crucial to enable the end-to-end automation that we need and we need those ASAP." As cloud, 5G and IoT began to converge, ML adds value to operating models by helping to create "smart" software networks. Pushed by IoT platforms, automation and cloud-based technologies, the global machine learning as a service (MLaaS) market is projected to grow from $ 679.32 million in 2016 to $7620.18 million by 2023 with a CAGR of 41.2%, according to Stratistics MRC. McRae said that web-scale Internet companies, such as Google and Amazon Web Services, were among the current leaders for using machine learning (ML) and automation. "They're still at the very start of this journey, but I believe in the future, for sure, those things will play a huge part in network operations and network optimization," McRae said. As for current machine learning use cases, BT is trialing ML for deploying segment routing on its network. "We're using machine learning and artificial intelligence as part of our path computation engine to show what's the best path and what has the least risks in that path," McRae said. "What is the most bandwidth on that path?