RSSAmplifier

Blog

Irati Hurtado

Recent content on Irati Hurtado

irhuru.github.ioRSS feed ↗34 posts

Latest posts

Human-like isn't the same as good: Rethinking Conversational AI's North Star

There's an assumption baked into most conversational AI roadmaps that nobody really questions anymore: more human-like is better. Friendlier persona, more natural turn-taking, more emotional mirroring, all treated as obvious improvements, the same way 'more accurate' or 'faster' would be. But is that actually true? Or have we been chasing a proxy for quality and calling it the goal itself? Why…

Text is the new interface and not everyone speaks it well

For most of computing history, the interface stood between you and the machine, and good UX meant designing that interface so a user's skill level stopped mattering much. LLMs quietly broke that deal. The interface is now a text box, and 'prompt engineering' is mostly just writing well: knowing what you want, saying it clearly, structuring it so someone (or something) can follow your logic, etc.…

Stop blaming the interface: Conversational AI fails in the user's head first

A lot of conversations with AI products go wrong. But if you dig into why these interactions really fail, you'll often find that the problem isn't in the interface at all, but it's psychological and it starts in the user's head, before they've even typed a word. Everyone walks into a conversation with a chatbot, voice assistant, or AI companion already carrying a set of expectations: about how…

From Grice to GPT: What classic linguistics can teach conversation designers

Have you ever noticed how even the best chatbots sometimes sound off? Not because their grammar’s wrong, but because something about the conversation just doesn’t feel natural. It turns out that long before GPT and LLMs, linguists were already thinking about what makes communication flow (what keeps a dialogue cooperative, relevant, and satisfying). One of them, Paul Grice, offered a framework in…

How digital avatars can change the way we talk to machines

The way we interact with machines is evolving rapidly, and digital avatars are at the forefront of that transformation. As conversational AI systems become more powerful and human-like, avatars are reshaping not just how these systems look, but how we experience and engage with them. But what exactly is a digital avatar and why does it matter in the world of conversational AI? What is a digital…

Can LLMs understand emojis?

Emojis are graphic symbols or ideograms that represent not only facial expressions, but also ideas and concepts. Emojis are extremely popular worldwide, primarily in chat and social media. They can appear as an additional element in the sentence, as in 'I liked the cake 😊', or replacing a word, as in 'I liked the 🍰'. From the perspective of large language models (LLMs), emojis are interesting…

Talking to machines: A brief history of chatbots

Chatbots, although a trendy product now, have existed for decades. Advances in AI have transformed chatbots from simple and rule-based systems to more complex and sophisticated conversational agents. Let’s dive into the history of chatbots, from their beginnings to the modern AI-driven companions we see today. ELIZA: The first chatbot (1966) ELIZA was the first chatbot, developed by professor…

Do LLMs have a personality?

Personality is the enduring set of characteristics or qualities that form a person’s distinctive character. The science of personality has been long studied by psychologists, and multiple tests have been designed to assess personality constructs. Interestingly, many of these tests follow a linguistic approach, since they are based on a factor analysis of the language used to describe behavior. For…

Is learning a programming language similar to learning a natural language?

Programming languages are systems of notation used for writing computer programs (e.g., Java), whereas natural languages are systems of communication that have developed naturally in human communities (e.g., French). Despite both of them being languages, little is known about whether learning a programming language is similar to learning a natural language. What does the research say?…

Why is there a preference for female voices in AI virtual assistants?

If you ask Amazon Alexa if it is a woman, it will answer 'I'm not a woman, I'm an AI'. In fact, the majority of AI virtual assistants are designed to be genderless (think of AI virtual assistants like Siri, Cortana, or Google Assistant). After all, they are simply made of code. Surprisingly, most of these AI virtual assistants use clear female voices when interacting with users. But, why is that?

Have LLMs developed Theory of Mind?

If you are into psychology, you have probably heard about Theory of Mind before. Theory of Mind is the ability to attribute mental states, such as beliefs, emotions, or desires, to oneself and others. It basically is the human ability to interpret what is going on in someone else's head. As you can imagine, Theory of Mind is an important cognitive skill needed for proper social interaction.…

Can we make LLMs learn in a more human-like way?

The difference between artificial intelligence (AI) and human intelligence is well known: whereas humans use their brain, their memory, and other cognitive abilities, AI uses data provided by humans. In fact, AI uses a lot of data. For example, large language models (LLMs), a type of generative AI, are trained on huge amounts of data. The Common Crawl archive, which is LLMs' primary source of…

Are there standards for the annotation of language corpora?

Some weeks ago, I attended a conference where several of the presenters discussed language corpora they had created as part of different research projects. During the Q&A turn of the last presenter, an award-winning researcher and creator of various corpora, one of the attendees raised their hand and asked: “I have attended five talks today discussing language corpora and each corpus employed…

How human-like are the linguistic abilities of ChatGPT?

The arrival of ChatGPT at the end of last year has sparked a debate about how it could impact several different fields, ranging from education, to content creation, to finance and to many others. This NLP tool is the epitome of generative AI, a type of technology that “generates” content, hence the name, after being trained on large amounts of data. In this case, text data. From a linguistic…

Exploring textual data using R Shiny

In previous blog posts, I discussed the use of software such as AntConc or Voyant to analyze corpus data. Even though these are very helpful, you can only use them to perform the range of default analyses they offer. But, what if you need more customization? Maybe you want to create your own graphs or explore your data in ways that go beyond traditional KWIC analyses and basic word frequencies.…

An introduction to Voyant Tools for corpus analysis

If you're into the Digital Humanities, you probably know about the importance of text analysis. Whereas programming languages such as Python or R allow you to do this efficiently, some people prefer to use tools which don't require any coding skills. A good example of such a tool is Voyant Tools, an open-source web-based application for text analysis. In this blog post, I will show you the basics…

The little-known history of synthetic voices

Synthetic voices, or artificial voices, are part of our lives (who has never talked to iPhone's Siri, Amazon Alexa, or Microsoft Cortana?). Conversational AI, which largely relies on these synthetic voices, is currently experiencing a boom and its market is expected to grow even more in the next four years. However, even though synthetic voices might look like a very new invention, the truth is…

A simple guide to using AntConc for corpus analysis

If you're interested in corpus linguistics, you probably know that some of the things this field involves includes searching for patterns in a group of texts (e.g., find frequent words or collocations). This can be easily done using a programming language like Python or R. But, what if you don't have much experience with programming languages? Well, this is where a tool like AntConc becomes…

Why is there a need to anthropomorphize language technologies?

Some weeks ago, we heard about Blake Lemoine, a Google engineer who claimed that LaMDA, a chatbot that can carry unstructured conversations with people, had a 'human soul'. He came to this conclusion after having long, deep conversations with LaMDA, which he later shared with The Washington Post. What's striking about this case is that an engineer, who's supposed to have a thorough technical…

Phonetic transcription systems you should know

What is phonetic transcription? Phonetic transcription is the use of phonetic symbols to represent speech. We could say it's a way to visually represent speech sounds, which is important because orthography often doesn't represent pronunciation accurately. Each phonetic symbol represents a sound in a spoken utterance. However, which phonetic symbol we use depends on the phonetic transcription…

Can computers change the way we speak?

Some months ago, I bought my first smart speaker: an Amazon echo. Because I live in the U.S., Alexa's default language was English. Even though I'm a native speaker of Spanish, I can also speak English, so it wasn't really a problem for me. However, when pronouncing Spanish proper names (e.g., people or places), I would still use my Spanish pronunciation. It turns out Alexa couldn't understand me…

Four key steps for developing guidelines for language data annotation

Whether you want to build a corpus to document an interesting linguistic phenomenon or simply want to have language data to train your AI models, the process of data annotation is crucial. Language data annotation (or text data annotation) is the process of adding labels with relevant information or meta-data. Even though it might seem like a simple process, choosing the right labels for your data…

How to get Twitter data

Nowadays, a lot of linguistics research uses language data from Twitter. At every major linguistics conference, there's at least one or two people (if not more!) discussing the use of language on this popular social network. Even though doing research on language use on Twitter is quite trendy now, getting the data for research is not that cool. The reason behind this? There are several ways to…

An analysis of Peter Pan using the R package koRpus

If you are into language research, you probably use R to analyze your data. One of the good things about R is that you can install packages to extend R's basic functions. In this blog post, I'd like to introduce you to a very useful package for language research: koRpus. Despite its name, the R package koRpus was designed to work with individual texts, and not with a collection of texts (I know,…

Transcribing and analyzing speech samples with CLAN

In a previous blog post, I wrote about automatic transcription software that you can use to speed up your transcription process. However, chances are that you also want to analyze what you transcribe if you're working on a research project. Measures such as words per minute or lexical diversity scores are quite common in linguistics research. In this blog post, I introduce you to CLAN, a software…

A short guide to building a Twitter bot

I recently discovered the fascinating world of Twitter bots (what's that?). Isn't it amazing that a simple computer program can post random stuff on Twitter all by itself? It probably isn't, considering that we're in the middle of a tech revolution, but they're fun to build, so that's a good enough reason to waste your time creating one! In this guide, I explain you how to build a Twitter bot from…

Creating a Haiku generator with Python

What is a haiku? I'm sure you have heard the term haiku at least once in your life. Haiku is a type of short form poetry that originated in Japan. These pieces of poetry consist of three phrases, where the first and third line have 5 syllables each and the second one has 7 syllables. This makes a total of 17 syllables per poem which follow a 5-7-5 pattern. Haikus have become popular in other…

Using machine learning to build a dialect classifier

As a linguist, I have always been curious about how computers can make sense of texts that combine different languages or language varieties. Therefore, I decided to learn more about this and implemented a dialect classifier. That is, an algorithm that would take some text as input and would determine the language variety (or dialect) it represents. To find out how I did it, read this post or take…

Using Amazon Mechanical Turk to conduct linguistic research

As a linguist, I often need to find speakers from a particular country or region to gather data on how they say something. However, it can sometimes be difficult to find a large group of speakers from a region you have never been to. In those cases, Amazon Mechanical Turk can come in handy. In this blog post, I explain how you can start using this awesome tool for your linguistic research!

How to create stunning presentations using Reveal.js in R

I recently found out about an R package that allows you to create Reveal.js presentations in R using RMarkdown syntax. If you are like me and love to display cool graphs in your slides, don't miss this short tutorial! Why you need to start using Reveal.js in R Okay, so you already know how to use PowerPoint. Why using something different then? There are many reasons, actually, but one of them is…

Blog posts coming soon!

I will start blogging very soon.

Content Design

.custom-button { display: inline-block; background-color: #0081FB; color: white; padding: 12px 24px; border-radius: 999px; font-family: Verdana, Geneva, sans-serif; font-size: 16px; text-decoration: none; font-weight: bold; transition: all 0.3s ease; } .custom-button span { display: inline-block; transition: transform 0.3s ease; } .custom-button:hover span { transform: translateX(5px); } Content…

Conversation Design

.custom-button { display: inline-block; background-color: #0081FB; color: white; padding: 12px 24px; border-radius: 999px; font-family: Verdana, Geneva, sans-serif; font-size: 16px; text-decoration: none; font-weight: bold; transition: all 0.3s ease; } .custom-button span { display: inline-block; transition: transform 0.3s ease; } .custom-button:hover span { transform: translateX(5px); } Open D…

Language Data & Analytics

.custom-button { display: inline-block; background-color: #0081FB; color: white; padding: 12px 24px; border-radius: 999px; font-family: Verdana, Geneva, sans-serif; font-size: 16px; text-decoration: none; font-weight: bold; transition: all 0.3s ease; } .custom-button span { display: inline-block; transition: transform 0.3s ease; } .custom-button:hover span { transform: translateX(5px); } Meta AI I…