TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Have LLMs solved natural language parsing?

5 点作者 horsh1超过 1 年前
With all the recent advancements in LLM and transformers, has the goal of parsing natural languages and representing them as an AST been achieved?<p>Or is this task still considered to be a hard one?<p>LLMs seem to understand the text much better than any previous technologies, so anaphoric resolution, and complex tenses, and POS choice, and rare constructs, and cross-language boundaries all don&#x27;t seem to be hard issues for them.<p>There are so many research papers published on LLMs and transformer now. With all kinds of applications, but they wll not quite there at all.

6 条评论

lfciv超过 1 年前
It feels like it&#x27;s sort of it&#x27;s own thing. LLMs are really good at morphing or fuzzy finding.<p>An interesting example – I had a project where I needed to parse out addresses and dates in a document. However, the address and date formats were not standardized across documents. Utilizing LLMs was way easier then trying to regex or pattern match across the text.<p>But if you&#x27;re trying to take a text document and break it down into some sort of a structured output, the outcome using LLMs will be much more variable.
mikewarot超过 1 年前
No. Word2Vec takes in words and converts them to a high dimensional vector. The relationship between the vectors in terms of cosine distance generally indicates similarity of meaning. The vector difference in terms can be used to indicate some relationship, for example [father]-[mother] is close in distance to [male]-[female].<p>There&#x27;s nothing like an abstract syntax tree, nor anything programmatic in the traditional meaning of programming going on inside the math of an LLM. It&#x27;s all just weights and wibbly-wobbly &#x2F; timey-whimey <i>stuff</i> in there.
评论 #38700344 未加载
usgroup超过 1 年前
I think it’s useful to draw a Chomsky-esque distinction here between understanding and usefulness.<p>I think LLMs haven’t advanced our understanding of how human language syntax&#x2F;semantics work, but they’ve massively advanced our ability to work with it.
minimaxir超过 1 年前
Not perfect, but using pretrained embeddings from a LLM will handle &gt;80% of your NLP problems.
评论 #38699533 未加载
seydor超过 1 年前
I think they show that parsing is not needed, it&#x27;s a limited idealization. Why is parsing a goal?
i_have_an_idea超过 1 年前
Turns out, grammars and ASTs to represent natural language are a dead end in NLP.