Speakers
Description
The rapid rise of generative AI has prompted claims that traditional dictionaries, and perhaps even dictionary use itself, are becoming obsolete. We acknowledge that generative language models are transforming many aspects of information access and language-related tasks. However, we argue that they do not remove the need for high‑quality, human‑curated lexical resources. Generative systems, which rely on stochastic sampling and lack explicit provenance and systematic source control, cannot guarantee information that is fully accurate, transparent, and verifiable, and this seems unlikely to change. In this landscape, carefully edited lexical data remain not only relevant but essential. The challenge for lexicography is therefore not survival, but adaptation: competing for users’ attention and integrating lexicographic strengths into an AI‑saturated environment.
In this poster, we present the redesign of a large national dictionary portal that now brings together twelve dictionaries, all meaning‑oriented rather than purely spelling‑oriented. Our central claim is that online dictionaries have relied too long on opaque lists of word forms, especially in search results and autocompletion interfaces. Conventional interfaces typically display bare lemma forms, at most accompanied by a part‑of‑speech label. Such lists presuppose that users already know the words, or else force them to click through to each article before they can judge relevance. This is not only inefficient; it underuses one of lexicography’s core assets: sense information.
We propose moving “beyond word lists” by integrating compact meaning information directly into all key navigational entry points. In the redesigned interface of our portal site, autocompletion results and other word lists are now always enriched with short snippets derived from dictionary definitions. Users thus see, from the outset, not only which headwords are attested in the dictionaries but also a concise indication of what each candidate means.
For instance, in our main contemporary dictionary, typing malle now automatically autocompletes with (cf. figure 1):
• malle1 sb. “1. benfisk der ofte har skægtråde og en lille fedtfinne …”
• malle2 sb. “lille bøjle eller ring af metal der sammen med en tilsvarende hægte bruges til at lukke fx en vest e. …”
• mallemuk sb. “ca. 45 cm lang, mågelignende stormfugl med grå ryg og tykhals”
This approach allows us to dispense with separate, formal disambiguation pages for homographs: instead of presenting a list of identical word forms that must be selected blindly, we display each homograph with an immediate sense‑based cue while typing, ideally shortening the path from the query to the relevant entry.
A further design challenge concerns inflected forms. Many existing systems exclude inflected forms from autocompletion, often out of concern that the same lemma will appear repeatedly. In our case, the autocompletion list is intended to assume much of the work of a separate search results page, so excluding inflected forms would significantly weaken its usefulness. We address this by allowing exact matches on inflected forms while still providing the same definition snippets, enabling users to navigate efficiently from the forms they actually type to the relevant lexical entries, cf. figure 2.
In cases where the user types a string that does not correspond to a lemma in the dictionary, the autocompletion list starts showing words that similar in spelling to the string, i.e. a Did you mean? result.
Implementing this snippet strategy across a heterogeneous set of twelve dictionaries, including several retro‑digitized works with only typographical markup and no explicit semantic structure, required developing methods for automatic snippet generation. In practice, we simply chose to present the first sense, or, in case of retro-digitized dictionaries, the beginning of the body of the entry, letting the sense number and the common abbreviation for ellipses, “...”, suggest that the entry include more sense information than presented in the autocompletion. We argue that even these automatically derived, simplistic sense cues represent a substantial usability gain. Even these support more informed navigation, reduce fruitless clicks, and make the semantic richness of the dictionaries visible earlier in the user journey.
We also utilize these snippets when supplying links to matches of the current query in the 11 other dictionaries, cf. figure 3.
We contend that the future of dictionaries lies not in mimicking generative AI, but in foregrounding what lexicography uniquely provides: curated, structured, and interpretable information that guides users through the lexicon more intelligently than word lists alone ever can. In the future, when the new platform has been live for a sufficient time, we hope to have data to support this.
In this poster, we present the motivation for the new design, and in-depth examples of the queries, before and after the redesign, and illustrative search scenarios highlighting how snippets support cross-dictionary navigation.