Teleradiology is the transmission of radiological patient images from procedures such as x-rays, Computed tomography (CT), and MRI imaging, from one location to another for the purposes of sharing studies with other radiologists and physicians. Teleradiology allows radiologists to provide services without actually having to be at the location of the patient. This is particularly important when a sub-specialist such as an MRI radiologist, neuroradiologist, pediatric radiologist, or musculoskeletal radiologist is needed, since these professionals are generally only located in large metropolitan areas working during daytime hours. Teleradiology allows for specialists to be available at all times. Teleradiology utilizes standard network technologies such as the Internet, telephone lines, wide area networks, local area networks (LAN) and the latest advanced technologies such as medical cloud computing. Specialized software is used to transmit the images and enable the radiologist to effectively analyze potentially hundreds of images of a given study. Technologies such as advanced graphics processing, voice recognition, artificial intelligence, and image compression are often used in teleradiology. Through teleradiology and mobile DICOM viewers, images can be sent to another part of the hospital or to other locations around the world with equal effort. Teleradiology is a growth technology given that imaging procedures are growing approximately 15% annually against an increase of only 2% in the radiologist population. == Reports == Teleradiology services commonly provide either preliminary or final interpretations of medical imaging studies. Preliminary reads are frequently used in emergency settings to support immediate clinical decisions and may include direct communication of critical findings to the referring physician. Some providers report turnaround times of approximately 30 minutes for emergency cases, with faster processing for time-sensitive conditions such as stroke. Final reads are definitive and used in official patient records and billing. These reports typically include all relevant findings and may require access to prior imaging and clinical data. Teleradiology is also employed to provide off-hour or overflow coverage for healthcare institutions lacking continuous on-site radiology staffing. == Subspecialties == Some teleradiologists are fellowship trained and have a wide variety of subspecialty expertise including such difficult-to-find areas as neuroradiology, pediatric neuroradiology, thoracic imaging, musculoskeletal radiology, mammography, and nuclear cardiology. There are also various medical practitioners who are not radiologists that take on studies in radiology to become sub specialists in their respected fields, an example of this is dentistry where oral and maxillofacial radiology allows those in dentistry to specialize in the acquisition and interpretation of radiographic imaging studies performed for diagnosis of treatment guidance for conditions affecting the maxillofacial region. == Teleultrasound == Teleradiology infrastructure has also been adapted to support point-of-care ultrasound (POCUS) in remote and austere environments. In teleultrasound—also known as telementored ultrasound—a remote expert guides a non-specialist in real time during image acquisition. This technique has been successfully demonstrated in extreme settings, including aboard the International Space Station, on Mount Everest, and during helicopter flight. == Regulations == In the United States, Medicare and Medicaid laws require the teleradiologist to be on U.S. soil in order to qualify for reimbursement of the Final Read. In addition, advanced teleradiology systems must also be HIPAA compliant, which helps to ensure patients' privacy. HIPAA (Health Insurance Portability and Accountability Act of 1996) is a uniform, federal floor of privacy protections for consumers. It limits the ways that entities can use patients' personal information and protects the privacy of all medical information no matter what form it is in. Quality teleradiology must abide by important HIPAA rules to ensure patients' privacy is protected. Also State laws governing the licensing requirements and medical malpractice insurance coverage required for physicians vary from state to state. Ensuring compliance with these laws is a significant overhead expense for larger multi-state teleradiology groups. Medicare (Australia) has identical requirements to that of the United States, where the guidelines are provided by the Department of Health and Ageing, and government based payments fall under the Health Insurance Act. The regulations in Australia are also conducted at both federal and state levels, ensuring that strict guidelines are adhered to at all times, with regular yearly updates and amendments are introduced (usually around March and November of every year), ensuring that the legislation is kept up to date with changes in the industry. One of the most recent changes to Medicare and radiology / teleradiology in Australia was the introduction of the Diagnostic Imaging Accreditation Scheme (DIAS) on 1 July 2008. DIAS was introduced to further improve the quality of Diagnostic Imaging and to amend the Health Insurance Act. == Industry growth == Until the late 1990s teleradiology was primarily used by individual radiologists to interpret occasional emergency studies from offsite locations, often in the radiologists home. The connections were made through standard analog phone lines. Teleradiology expanded rapidly as the growth of the internet and broad band combined with new CT scanner technology to become an essential tool in trauma cases in emergency rooms throughout the country. The occasional 2–3 x-ray studies a week soon became 3–10 CT scans, or more, a night. Because ER physicians are not trained to read CT scans or MRIs, radiologists went from working 8–10 hours a day, five and half days a week to a schedule of 24 hours a day, 7 days a week coverage. This became a particularly acute challenge in smaller rural facilities that only had one solo radiologist with no other to share call. These circumstances spawned a post-dot.com boom of firms and groups that provided medical outsourcing, off-site teleradiology on-call services to hospitals and Radiology Groups around the country. As an example, a teleradiology firm might cover trauma at a hospital in Indiana with doctors based in Texas. Some firms even used overseas doctors in locations like Australia and India. Nighthawk, founded by Paul Berger, was the first to station U.S. licensed radiologists overseas (initially Australia and later Switzerland) to maximize the time zone difference to provide nightcall in U.S. hospitals. Currently, teleradiology firms are facing pricing pressures. Industry consolidation is likely as there are more than 500 of these firms, large and small, throughout the United States.
Dropbox Paper
Dropbox Paper, or simply Paper, is a collaborative document-editing service developed by Dropbox. Originating from the company's acquisition of document collaboration company Hackpad in April 2014, Dropbox Paper was officially announced in October 2015, and launched in January 2017. It offers a web application, as well as mobile apps for Android and iOS. Dropbox Paper was described in the official announcement post as "a flexible workspace that brings people and ideas together. With Paper, teams can create, review, revise, manage, and organize — all in shared documents". Reception of Dropbox Paper has been mixed. Critics praised collaboration functionality, including content available immediately, the ability to mention specific collaborators, assign tasks, write comments, as well as editing attribution, and revision history. It received particular praise for its support for rich media from a variety of sources, with one reviewer noting that the Paper's support for rich media exceeds the capabilities of most of its competitors. However, it was criticized for a lack of formatting options and editing features. While the user interface was liked for being minimal, reviewers cited the lack of a fixed formatting bar and missing features present in competitors' products as making Dropbox Paper seem like a "light" tool. == History == Dropbox acquired document collaboration company Hackpad in April 2014. A year later, Dropbox launched a Dropbox Notes note-taking product in beta testing phase. Dropbox Paper was officially announced on October 15, 2015, followed by an open beta and release of mobile Android and iOS apps in August 2016. Dropbox Paper was officially released on January 30, 2017. == Reception == In a comparison between Dropbox Paper and Evernote, PC World's Michael Ansaldo wrote that "With its emphasis on document creation, you might expect formatting to be front and center in Dropbox Paper. That's not the case." Ansaldo noted the lack of a "fixed formatting toolbar as you'd find in Evernote or a word processor like Google Docs or Microsoft Word. Instead, the text editor appears as a floating ribbon only when you highlight selected text." The only formatting options available for emphasis were bolding, strikethrough, bulleted and numbered lists, and H1 and H2 tags. Users can also add links, convert text to checklists, and add comments. Ansaldo wrote that "Both Evernote and Dropbox Paper make it easy to add images to a document", but also noted that "Dropbox Paper doesn't support any image editing". Paper supports rich media, and users can "add rich content to your document just by pasting a link to the file. In addition to Dropbox, Paper supports media from a variety of popular services including YouTube, Spotify, Vimeo, SoundCloud, Facebook, and Google's productivity suite. Once the file appears, you can delete the link for a cleaner display." To start working with other people, Paper "allows you to invite people via email from within a document", with sharing options for who can view the link (anyone with the link or just the invited person), and action permissions (edit or only comment). Regarding collaboration, Ansaldo wrote that "Creative collaboration is Paper’s marquee feature, and it provides a variety of ways to work effectively with others in real time". Users can "make any content immediately visible and accessible to a specific collaborator with "@mentions"", and "You can also use @mentions to create and assign task lists within a document." Paper also "boasts essential collaboration tools including comments, editing attribution, and revision history." Writing for TechRadar, John Brandon wrote that Dropbox Paper "might be a 'light' tool for now without the extensive templates of Microsoft Office or the integration with other apps in the Zoho suite, but it does work well with the Dropbox storage service that's so popular with office workers these days." Kyle Wiggers of Digital Trends wrote that Paper is "all about minimizing distractions. Its interface is quite literally a big, blank canvas on which you tap out your agenda. You can organize notes by title and create to-do lists, but even basic formatting tools are obscured from view", noting Paper's "floating box above words and phrases highlighted by your cursor". Wiggers stated that "Paper is not a to-do organizer", but that it's "well suited to the purpose thanks to a bevy of labor-saving conveniences", highlighting that Paper "supports more media than most of its to-do and note-taking counterparts". He praised the collaboration tools, writing that they "are as extensive as you'd hope, and then some", citing its invitation system with permission controls, lists of changes and revision history, comment and chat support, and "perhaps best of all", the ability to assign tasks with a "@" mention. Business Insider's Alex Heath praised that "Paper's interface is spotless and friendly to write in. You don't feel overwhelmed with formatting options", but criticized the available features, writing that "Google Docs is much more full-featured in the formatting department, so Paper has some catching up to do if it wants to be on par with the competition". Writing for The Verge, Casey Newton praised Paper's handling of rich media, complimenting it for being "great", and added that "I imagine that creative types who work on teams will appreciate having rich media embedded in the documents they're working on rather than in a series of infinite tabs".
Small language model
Small language models or compact language models are artificial intelligence language models designed for human natural language processing including language and text generation. They are smaller in scale and scope than large language models. A large language model typically contains hundreds of billions of training parameters, with some models exceeding a trillion parameters. This substantial parameter count enables the model to encode vast amounts of information, thereby improving the generalizability and accuracy of its outputs. However, training such models demands enormous computational resources, rendering it infeasible for an individual to do so using a single computer and graphics processing unit. Small language models, on the other hand, use far fewer parameters, typically ranging from a few thousand to a few hundred million. This make them more feasible to train and host in resource-constrained environments such as a single computer or even a mobile device. Most contemporary (2020s) small language models use the same architecture as a large language model, but with a smaller parameter count and sometimes lower arithmetic precision. Parameter count is reduced by a combination of knowledge distillation and pruning. Precision can be reduced by quantization. Work on large language models mostly translate to small language models: pruning and quantization are also widely used to speed up large language models. == Models == Some notable models are: Below 1B parameters: Llama-Prompt-Guard-2-22M (detects prompt injection and jailbreaking, based on DeBERTa-xsmall), SmolLM2-135M, SmolLM2-360M 1–4B parameters: Llama3.2-1B, Qwen2.5-1.5B, DeepSeek-R1-1.5B, SmolLM2-1.7B, SmolVLM-2.25B, Phi-3.5-Mini-3.8B, Phi-4-Mini-3.8B, Gemma3-4B; closed-weights ones include Gemini Nano 4–14B parameters: Mistral 7B, Gemma 9B, Phi-4 14B. Phi-4 14B is marginally "small" at best, but Microsoft does market it as a small model. == Language model with small pre-training dataset == Traditional AI language systems need enormous computers and vast amounts of data. Pre-training matters, even tiny models show significant performance improvements when pre-trained performance increases with larger pre-training datasets. Classification accuracy improves when pre-training and test datasets share similar tokens. Shallow architectures can replicate deep model performance through collaborative learning.
Nick Frosst
Nicholas M. W. Frosst is a Canadian computer scientist and musician. He co-founded Cohere, a Toronto-based artificial intelligence company. He is also the lead singer in the indie rock band Good Kid. == Early life and education == Frosst was born on January 5, 1993. Frosst earned a Bachelor of Science degree in computer science and cognitive science from the University of Toronto in 2015. He was a student of Geoffrey Hinton, who also hired Frosst at Google Brain. == Career == Frosst was among Geoffrey Hinton's earliest hires at Google Brain in Toronto, working as a machine learning researcher on deep learning and neural network architectures. He worked there from 2016 to 2020. Frosst co-founded Cohere with Aidan Gomez and Ivan Zhang in 2019. The company builds large language models and enterprise AI tools. Frosst has publicly explained Cohere's focus on industries like finance and health, where there are privacy and other regulatory considerations. Frosst has also spoken openly about his belief that artificial intelligence will not replace humans, but rather streamline and automate mundane tasks, and his belief that AGI is less "imminent" than many in the field claim. Frosst and the other Cohere co-founders were listed first on Maclean's AI Trailblazers Power List and The Logic's Innovation Leaders. == Music == After spending time in a prior band which played "weird" music featuring a glockenspiel, Frosst and fellow computer science students at the University of Toronto formed the indie rock band Good Kid in 2015. Frosst is the lead vocalist for the band. While on tour with the band, Frosst continues his work in the tech industry remotely. Frosst has described the band as way for him to relax and not constantly think about tech. His vocals have been compared to that of Kele Okereke. As of 2026, the band, which has performed at Lollapalooza, has 3.1 million monthly Spotify listeners. In 2024, the band was nominated for the Juno Awards Breakthrough Group of the Year. == Discography == === Good Kid === Can We Hang Out Sometime? (2026)
Multiple sequence alignment
Multiple sequence alignment (MSA) is the process or the result of sequence alignment of three or more biological sequences, generally protein, DNA, or RNA. These alignments are used to infer evolutionary relationships via phylogenetic analysis and can highlight homologous features between sequences. Alignments highlight mutation events such as point mutations (single amino acid or nucleotide changes), insertion mutations and deletion mutations, and alignments are used to assess sequence conservation and infer the presence and activity of protein domains, tertiary structures, secondary structures, and individual amino acids or nucleotides. Multiple sequence alignments require more sophisticated methodologies than pairwise alignments, as they are more computationally complex. Most multiple sequence alignment programs use heuristic methods rather than global optimization because identifying the optimal alignment between more than a few sequences of moderate length is prohibitively computationally expensive. However, heuristic methods generally cannot guarantee high-quality solutions and have been shown to fail to yield near-optimal solutions on benchmark test cases. == Problem statement == Given m {\displaystyle m} sequences S i {\displaystyle S_{i}} , i = 1 , ⋯ , m {\displaystyle i=1,\cdots ,m} similar to the form below: S := { S 1 = ( S 11 , S 12 , … , S 1 n 1 ) S 2 = ( S 21 , S 22 , ⋯ , S 2 n 2 ) ⋮ S m = ( S m 1 , S m 2 , … , S m n m ) {\displaystyle S:={\begin{cases}S_{1}=(S_{11},S_{12},\ldots ,S_{1n_{1}})\\S_{2}=(S_{21},S_{22},\cdots ,S_{2n_{2}})\\\,\,\,\,\,\,\,\,\,\,\vdots \\S_{m}=(S_{m1},S_{m2},\ldots ,S_{mn_{m}})\end{cases}}} A multiple sequence alignment is taken of this set of sequences S {\displaystyle S} by inserting any amount of gaps needed into each of the S i {\displaystyle S_{i}} sequences of S {\displaystyle S} until the modified sequences, S i ′ {\displaystyle S'_{i}} , all conform to length L ≥ max { n i ∣ i = 1 , … , m } {\displaystyle L\geq \max\{n_{i}\mid i=1,\ldots ,m\}} and no values in the sequences of S {\displaystyle S} of the same column consists of only gaps. The mathematical form of an MSA of the above sequence set is shown below: S ′ := { S 1 ′ = ( S 11 ′ , S 12 ′ , … , S 1 L ′ ) S 2 ′ = ( S 21 ′ , S 22 ′ , … , S 2 L ′ ) ⋮ S m ′ = ( S m 1 ′ , S m 2 ′ , … , S m L ′ ) {\displaystyle S':={\begin{cases}S'_{1}=(S'_{11},S'_{12},\ldots ,S'_{1L})\\S'_{2}=(S'_{21},S'_{22},\ldots ,S'_{2L})\\\,\,\,\,\,\,\,\,\,\,\vdots \\S'_{m}=(S'_{m1},S'_{m2},\ldots ,S'_{mL})\end{cases}}} To return from each particular sequence S i ′ {\displaystyle S'_{i}} to S i {\displaystyle S_{i}} , remove all gaps. == Graphing approach == A general approach when calculating multiple sequence alignments is to use graphs to identify all of the different alignments. When finding alignments via graph, a complete alignment is created in a weighted graph that contains a set of vertices and a set of edges. Each of the graph edges has a weight based on a certain heuristic that helps to score each alignment or subset of the original graph. === Tracing alignments === When determining the best suited alignments for each MSA, a trace is usually generated. A trace is a set of realized, or corresponding and aligned, vertices that has a specific weight based on the edges that are selected between corresponding vertices. When choosing traces for a set of sequences it is necessary to choose a trace with a maximum weight to get the best alignment of the sequences. == Alignment methods == There are various alignment methods used within multiple sequence to maximize scores and correctness of alignments. Each is usually based on a certain heuristic with an insight into the evolutionary process. Most try to replicate evolution to get the most realistic alignment possible to best predict relations between sequences. === Dynamic programming === A direct method for producing an MSA uses the dynamic programming technique to identify the globally optimal alignment solution. For proteins, this method usually involves two sets of parameters: a gap penalty and a substitution matrix assigning scores or probabilities to the alignment of each possible pair of amino acids based on the similarity of the amino acids' chemical properties and the evolutionary probability of the mutation. For nucleotide sequences, a similar gap penalty is used, but a much simpler substitution matrix, wherein only identical matches and mismatches are considered, is typical. The scores in the substitution matrix may be either all positive or a mix of positive and negative in the case of a global alignment, but must be both positive and negative, in the case of a local alignment. For n individual sequences, the naive method requires constructing the n-dimensional equivalent of the matrix formed in standard pairwise sequence alignment. The search space thus increases exponentially with increasing n and is also strongly dependent on sequence length. Expressed with the big O notation commonly used to measure computational complexity, a naïve MSA takes O(LengthNseqs) time to produce. To find the global optimum for n sequences this way has been shown to be an NP-complete problem. In 1989, based on Carrillo-Lipman Algorithm, Altschul introduced a practical method that uses pairwise alignments to constrain the n-dimensional search space. In this approach pairwise dynamic programming alignments are performed on each pair of sequences in the query set, and only the space near the n-dimensional intersection of these alignments is searched for the n-way alignment. The MSA program optimizes the sum of all of the pairs of characters at each position in the alignment (the so-called sum of pair score) and has been implemented in a software program for constructing multiple sequence alignments. In 2019, Hosseininasab and van Hoeve showed that by using decision diagrams, MSA may be modeled in polynomial space complexity. === Progressive alignment construction === The most widely used approach to multiple sequence alignments uses a heuristic search known as progressive technique (also known as the hierarchical or tree method) developed by Da-Fei Feng and Doolittle in 1987. Progressive alignment builds up a final MSA by combining pairwise alignments beginning with the most similar pair and progressing to the most distantly related. All progressive alignment methods require two stages: a first stage in which the relationships between the sequences are represented as a phylogenetic tree, called a guide tree, and a second step in which the MSA is built by adding the sequences sequentially to the growing MSA according to the guide tree. The initial guide tree is determined by an efficient clustering method such as neighbor-joining or unweighted pair group method with arithmetic mean (UPGMA), and may use distances based on the number of identical two-letter sub-sequences (as in FASTA rather than a dynamic programming alignment). Progressive alignments are not guaranteed to be globally optimal. The primary problem is that when errors are made at any stage in growing the MSA, these errors are then propagated through to the final result. Performance is also particularly bad when all of the sequences in the set are rather distantly related. Most modern progressive methods modify their scoring function with a secondary weighting function that assigns scaling factors to individual members of the query set in a nonlinear fashion based on their phylogenetic distance from their nearest neighbors. This corrects for non-random selection of the sequences given to the alignment program. Progressive alignment methods are efficient enough to implement on a large scale for many (100s to 1000s) sequences. A popular progressive alignment method has been the Clustal family. ClustalW is used extensively for phylogenetic tree construction, in spite of the author's explicit warnings that unedited alignments should not be used in such studies and as input for protein structure prediction by homology modeling. European Bioinformatics Institute (EMBL-EBI) announced that CLustalW2 will expire in August 2015. They recommend Clustal Omega which performs based on seeded guide trees and HMM profile-profile techniques for protein alignments. An alternative tool for progressive DNA alignments is multiple alignment using fast Fourier transform (MAFFT). Another common progressive alignment method named T-Coffee is slower than Clustal and its derivatives but generally produces more accurate alignments for distantly related sequence sets. T-Coffee calculates pairwise alignments by combining the direct alignment of the pair with indirect alignments that aligns each sequence of the pair to a third sequence. It uses the output from Clustal as well as another local alignment program LALIGN, which finds multiple regions of local alignment between two sequences. The resulting alignment and phylogenetic tree are used as a guide to produce new and more accurate w
Pronunciation assessment
Automatic pronunciation assessment uses computer speech recognition to determine how accurately speech has been pronounced, instead of relying on a human instructor or proctor. It is also called speech verification, pronunciation evaluation, and pronunciation scoring. This technology is used to grade speech quality, for language testing, for computer-aided pronunciation teaching (CAPT) in computer-assisted language learning (CALL), for speaking skill remediation, and for accent reduction. Pronunciation assessment is different from dictation or automatic transcription, because instead of determining unknown speech, it verifies learners' pronunciation of known word(s), often from prior transcription of the same utterance; ideally scoring the intelligibility of the learners' speech. Sometimes pronunciation assessment evaluates the prosody of the learners' speech, such as intonation, pitch, tempo, rhythm, and syllable and word stress, although those are usually not essential for being understood in most languages. Pronunciation assessment is also used in reading tutoring, for example in products from Google, Microsoft, and Amira Learning. Automatic pronunciation assessment can also be used to help diagnose and treat speech disorders such as apraxia. == Intelligibility == Intelligibility refers to how well a learner's utterance is understood by a listener, rather than how much it sounds like a native speaker. This is separate from measures of fluency, such as so-called "Goodness of Pronunciation" (GoP) scores, which estimate how closely an utterance aligns with those of native speakers. Intelligibility is widely regarded as the most important communicative goal in pronunciation teaching and assessment. For example, in the Common European Framework of Reference for Languages (CEFR) assessment criteria for "overall phonological control", intelligibility outweighs formally correct pronunciation at all levels. Studies in applied linguistics have shown that accent reduction does not always increase intelligibility because listeners can often comprehend heavily accented speech without difficulty. Pronunciation assessment systems often rely on acoustic methods such as GoP which compare learner speech to reference models to produce phoneme-level scores, which are in turn aggregated to produce word and phrase scores. While these methods are effective for identifying deviations from native speakers' utterances, they do not effectively measure how understandable speech is to human listeners. Intelligibility is influenced by broader linguistic and contextual factors such as stress placement, speech rate, and coarticulation, which are not represented in purely segmental scores. The earliest work on pronunciation assessment avoided measuring genuine listener intelligibility, a shortcoming corrected in 2011 at the Toyohashi University of Technology, and included in the Versant high-stakes English fluency assessment from Pearson and mobile apps from 17zuoye Education & Technology, but still missing in 2023 products from Google Search, Microsoft, Educational Testing Service, Speechace, and ELSA. Assessing authentic listener intelligibility is essential for avoiding inaccuracies from accent bias, especially in high-stakes assessments; from words with multiple correct pronunciations; and from phoneme coding errors in machine-readable pronunciation dictionaries. In 2022, researchers found that some newer speech-to-text systems, based on end-to-end reinforcement learning to map audio signals directly into words, produce word and phrase confidence scores (from 10-25ms audio frame logit aggregation) closely correlated with genuine listener intelligibility. Others have been able to assess intelligibility using Levenshtein or dynamic time warping distance measures from Wav2Vec2 representation of good speech. Further work through 2025 has focused specifically on measuring intelligibility. A 2025 study of 42 pronunciation and speech coaching apps (32 mobile and 10 web) found that none offered intelligibility assessment. Instead, most provided only segmental and accent-focused scoring. About two-thirds of the apps provided some form of specific pronunciation feedback, usually with phonetic transcriptions, but accompanied by visual cues (such as animations of the vocal tract or the lips and tongue from the front) in only about 5% of the apps. Less than a third provided feedback on learner perception of exemplar speech. == Evaluation == Although there are as yet no industry-standard benchmarks for evaluating pronunciation assessment accuracy, researchers occasionally release evaluation speech corpuses for others to use for improving assessment quality. Such evaluation databases often emphasize formally unaccented pronunciation to the exclusion of genuine intelligibility evident from blinded listener transcriptions. As of mid-2025, state of the art approaches for automatically transcribing phonemes typically achieve an error rate of about 10% from known good speech. The International Speech Communication Association (ISCA) 2025 Workshop on Speech and Language Technology in Education (SLaTE) administered a Speak & Improve Challenge: Spoken Language Assessment and Feedback, introducing benchmarks for evaluating pronunciation assessment and remediation systems across languages, accents, and learner populations. The challenge emphasized cross-lingual generalization and alignment with human intelligibility judgments, for more robust and interpretable assessment systems. Ethical issues in pronunciation assessment are present in both human and automatic methods. Authentic validity, fairness, and mitigating bias in evaluation are all crucial. Diverse speech data should be included in automatic pronunciation assessment models. Combining human judgments, especially blinded transcriptions from a wide diversity of listeners, with automated feedback can improve accuracy and fairness. Second language learners benefit substantially from their use of widely available speech recognition systems for dictation, virtual assistants, and AI chatbots. In such systems, users naturally try to correct their own errors evident in speech recognition results that they notice. Such use improves their grammar and vocabulary development along with their pronunciation skills. The extent to which explicit pronunciation assessment and remediation approaches improve on such self-directed interactions remains an open question. Similarly, automatic dictation results have been shown to reflect intelligibility about as well as human scorers. == Recent developments == During 2021–22, a smartphone-based CAPT system was used to sense articulation through both audible and inaudible signals, providing feedback at the phoneme level. Some promising areas for improvement which were being developed in 2024 include articulatory feature extraction and transfer learning to suppress unnecessary corrections. Other interesting advances under development include "augmented reality" interfaces for mobile devices using optical character recognition to provide pronunciation training on text found in user environments. In 2024, audio multimodal large language models were first described as assessing pronunciation. That work has been carried forward by other researchers in 2025 who report positive results. Subsequently, researchers demonstrated pronunciation scoring by providing a language model with textual descriptions of speech, including the speech-to-text transcript, phoneme sequences, pauses, and phoneme sequence matching; this approach can achieve performance similar to multimodal LLMs that analyze raw audio while avoiding their higher computational cost. In 2025, the Duolingo English Test authors published a description of their pronunciation assessment method, purportedly built to measure intelligibility rather than accent imitation. While achieving a correlation of 0.82 with expert human ratings, very close to inter-rater agreement and outperforming alternative methods, the method is nonetheless based on experts' scores along the six-point CEFR common reference levels scale, instead of actual blinded listener transcriptions. Further promising work in 2025 includes assessment feedback aligning learner speech to synthetic utterances using interpretable features, identifying continuous spans of words for remediation feedback; synthesizing corrected speech matching learners' self-perceived voices, which they prefer and imitate more accurately as corrections; and streaming such interactions. On January 21, 2026, Educational Testing Service's TOEFL iBT high-stakes English language test, required by US university admissions and employers from English as a foreign language applicants more often than all other internet-based tests combined, changed its speaking assessments. While official rubrics claim that the new scoring will be based primarily on intelligibility, the new test's technical description indicates that it ju
Yorick Wilks
Yorick Alexander Wilks FBCS (27 October 1939 – 14 April 2023) was a British computer scientist. He was an emeritus professor of artificial intelligence at the University of Sheffield, visiting professor of artificial intelligence at Gresham College (a post created especially for him), senior research fellow at the Oxford Internet Institute, senior scientist at the Florida Institute for Human and Machine Cognition, and a member of the Epiphany Philosophers. In February 2023, Wilks joined WiredVibe as Director of AI and a Board Member, with the goal of commercializing his previous research and ideas. He remained in this role until his death, which occurred shortly before WiredVibe was acquired by AKY X, a company that continues to build on his legacy and contributions. == Biography == Wilks was born in Gerrards Cross, Buckinghamshire in England. He was educated at Torquay Boys' Grammar School, followed by Pembroke College, Cambridge, where he read Philosophy, joined the Epiphany Philosophers and obtained his Doctor of Philosophy degree (1968) under Professor R. B. Braithwaite for the thesis 'Argument and Proof'; he was an early pioneer in meaning-based approaches to the understanding of natural language content by computers. His main early contribution in the 1970s was called "Preference Semantics" (Wilks, 1973; Wilks and Fass, 1992), an algorithmic method for assigning the "most coherent" interpretation to a sentence in terms of having the maximum number of internal preferences of its parts (normally verbs or adjectives) satisfied. That early work was hand-coded with semantic entries (of the order of some hundreds) as was normal at the time, but since then has led to the empirical determinations of preferences (chiefly of English verbs) in the 1980s and 1990s. A key component of the notion of preference in semantics was that the interpretation of an utterance is not a well- or ill-formed notion, as was argued in Chomskyan approaches, such as those of Jerry Fodor and Jerrold Katz. It was rather that a semantic interpretation was the best available, even though some preferences might not be satisfied. So, in "The machine answered the question with a low whine" the agent of "answer" does not satisfy that verb's preference for a human answerer—which would cause it to be deemed ill-formed by Fodor and Katz—but is accepted as sub-optimal or metaphorical, and, now, conventional. The function of the algorithm is not to determine well-formedness at all but to make the optimal selection of word-senses to participate in the overall interpretation. Thus, in "The Pole answered..." the system will always select the human sense of the agent and not the inanimate one if it gives a more coherent interpretation overall. Preference Semantics is thus some of the earliest computational work—with programs run at Systems Development Corporation in Santa Monica in 1967 in LISP on an IBM360—in the now established field of word sense disambiguation. This approach was used in the first operational machine translation system based principally on meaning structures and built by Wilks at Stanford Artificial Intelligence Laboratory in the early 1970s (Wilks, 1973) at the same time and place as Roger Schank was applying his "Conceptual Dependency" approach to machine translation. The LISP code of Wilks' system was in The Computer Museum, Boston. Wilks was elected a fellow of the American and European Associations for Artificial Intelligence, of the British Computer Society, a member of the UK Computing Research Committee, and a permanent member of ICCL, the International Committee on Computational Linguistics. He was professor of artificial intelligence at the University of Sheffield and a senior research fellow at the Oxford Internet Institute. In 1991 he received a Defense Advanced Projects Agency grant on interlingual pragmatics-based machine translation and in 1994 he received a grant by the Engineering and Physical Sciences Research Council to investigate in the field of large-scale information extraction (LaSIE); in the following years he would obtain more grants to carry on exploring the field of information extraction (AVENTINUS, ECRAN, PASTA...). In the 1990s Wilks also became interested in modelling human-computer dialogue and the team led by David Levy and him as chief researcher won the Loebner Prize in 1997. He was the founding director of the EU funded Companions Project on creating long-term computer companions for people. At his Festschrift in 2007 at the British Computer Society in London a volume of his own papers was presented along with a volume of essays in his honour. He was awarded the Antonio Zampolli prize in honour of his lifetime work at the LREC 2008 conference on 28 May 2008, and the Lifetime Achievement Award at the ACL 2008 conference on 18 June 2008. In 2009, he was awarded the British Computer Society's Lovelace Medal, its annual award for research achievement, and was awarded the Fellowship of the Association for Computing Machinery. In 1998, Wilks became head of the Department of Computer Science of the University of Sheffield, where he had started working in the year 1993 as professor of artificial intelligence, a post he still held. In 1993 he became the founding director of the Institute of Language, Speech and Hearing (ILASH). Wilks also set up the Natural Language Processing Group of the University of Sheffield. In 1994 he (along with Rob Gaizauskas and Hamish Cunningham) designed GATE, an advanced NLP architecture that has been widely distributed. National Life Stories conducted an oral history interview (C1672/24) with Yorick Wilks in 2016 for its Science and Religion collection held by the British Library. Wilks died on 14 April 2023, at the age of 83. == Awards == Wilks received many awards: (2009) Elected Fellow of the Association for Computing Machinery (2009) Lovelace Medal by the British Computer Society (2008) Zampolli Prize (ELRA, awarded at LREC in Marrakech, Morocco) (2008) Lifetime Achievement Award (Association for Computational Linguistics, in Columbus) (2006) Visiting Professor, University of Oxford (2004) Elected to UK Computing Research Committee (2004) Elected Fellow, British Computer Society (2003) Visiting Fellow, Oxford Internet Institute (1998) Elected Fellow of European Association for Artificial Intelligence (1997) Elected Fellow, EPSRC College of Computing (1991) Visiting Fellow, Trinity Hall, Cambridge (1991) Elected Fellow of the American Association for Artificial Intelligence (1983) Royal Society Travel Fellowship (1983) Commonwealth of Australia Visiting Professor (1981) Visiting Sloan Fellow, University of California, Berkeley (1980) Invited Participant in the Nobel Symposium on Language, Stockholm (1979) NATO Senior Scientist Fellowship (1979) Visiting Sloan Fellow, Yale University (1975) SRC Senior Visiting Fellowship, University of Edinburgh == Membership == Wilks was an active member of the following associations: Association for Computational Linguistics Society for the Study of AI and Simulation of Behaviour Association for Computing Machinery Cognitive Science Society British Society for the Philosophy of Science American Association for Artificial Intelligence Aristotelian Society == Selected works == === Books === Wilks, Y. (2019) Artificial Intelligence: Modern Magic or Dangerous Future?.Icon Books. New illustrated edition, 2023, MIT Press. Wilks, Y. (2015) Machine Translation: its scope and limits. Springer Wilks, Y (ed.) (2010) Close Engagements with Artificial Companions: Key Social, Psychological and Design issues. John Benjamins; Amsterdam Wilks, Y., Brewster, C. (2009) Natural Language Processing as a Foundation of the Semantic Web. Now Press: London. Wilks, Y. (2007) Words and Intelligence I, Selected papers by Yorick Wilks. In K. Ahmad, C. Brewster & M. Stevenson (eds.), Springer: Dordrecht. Wilks, Y. (ed. and with introduction and commentaries). (2006) Language, cohesion and form: selected papers of Margaret Masterman. Cambridge: Cambridge University Press. Wilks, Y., Nirenburg, S., Somers, H. (eds.) (2003) Readings in Machine Translation. Cambridge, MA: MIT Press. Wilks, Y.(ed.). (1999) Machine Conversations. Kluwer: New York. Wilks, Y., Slator, B., Guthrie, L. (1996) Electric Words: dictionaries, computers and meanings. Cambridge, MA: MIT Press. Ballim, A., Wilks, Y. (1991) Artificial Believers. Norwood, NJ: Erlbaum. Wilks, Y.(ed.). (1990) Theoretical Issues in Natural Language Processing. Norwood, NJ: Erlbaum. Wilks, Y., Partridge, D. (eds. plus three YW chapters and an introduction). (1990) The Foundations of Artificial Intelligence: a sourcebook. Cambridge: Cambridge University Press. Wilks, Y., Sparck-Jones, K.(eds.). (1984) Automatic Natural Language Processing, paperback edition. New York: Wiley. Originally published by Ellis Horwood. Wilks, Y., Charniak, E. (eds and principal authors). (1976) Computational Semantics—an Introduction to Artificial Intelligence and