Search This Blog

Sunday, 6 April 2014

Let’s not forget how to walk…

By Katherine Howard
 


http://farm9.staticflickr.com/8456/7941866470_00846a8127_o.jpg
Some of you may have seen the media release and subsequent blog posts from Swets in December last year about Natural Language Processing and academic literature search. 

Natural Language Processing – or NLP – is not a new field of study, but according to Swets it has seen renewed attention since Yahoo acquired the NLP technology company, SkyPhrase, in early December last year.  NLP draws on concepts and research from multiple fields, such as linguistics (including computer-mediated communication, or CMC), human-computer interaction (HCI), artificial intelligence (AI); and those more closely related to LIS such as information retrieval (IR) and knowledge representation (KR).

Broadly, NLP is about developing a computer language that is capable of understanding natural human languages such as English, with all of its idiosyncrasies: synonyms, homonyms, homographs for example that are largely incomprehensible to a computer.  However, advances have been and continue to be made, with perhaps the best-known current application of NLP being spelling and grammar check.

The blog posts relate the advantages of NLP to academic researchers in particular, whose role it is to “stay[…] abreast of recent developments” (Part 1, para. 1) in their field.  The case is made that while a researcher may be able to tell very quickly which search results are relevant, the sheer amount of information available makes it possible that other useful extant material may not be located by that researcher.  NLP seeks to understand the content/topic that the researcher is interested in by using “indicator phrases” to contextualise the surrounding words and sentences in order to return more highly relevant results. Or in ‘librarian speak,’ a higher precision to recall ratio, where the computer itself will be able to determine if the duck you want to carve is for dinner or a hobby.

This contextualisation is known as ‘semantic search’, and the second blog post briefly discusses this and the (until now) hypothetical semantic web.  The importance of metadata as an aid to this semantic discoverability is highlighted, and it is explained how NLP processing principles can be used to “automatically extract textual metadata and classify resources into established taxonomies” (Part 2, para. 5).

The premise underpinning this technological development is the realisation that
“Familiar keyword search as in Google […] is not always the ideal mechanism when looking for scholarly material […]. Pure keyword search may well miss crucial subtleties of context in academic literature, fail to return results rich in semantically similar keywords, or be incapable of tailoring results depending on whether the user is looking for discussion of methods, support for a hypothesis, background reading and literature review or a myriad of other specific uses for the material” (Part 1, para. 10).

As information professionals, we have known this for a while now. We have highly developed search skills that enable us to find the right information at the right time - but what do non-information professionals do? How do people obtain the skills to not only find information, but to evaluate it for its relevancy, accuracy and validity? The necessity for these skills in today’s information-rich environment I believe moves far beyond the academic researchers that are the subject of these two blog posts.  Anyone who seeks information electronically – whether via the World Wide Web or via subscription databases – should have these skills. This doesn’t mean I’m arguing for Grandma to have a skill-set equal to that of a qualified librarian – these skills can be learnt (and used) to varying degrees of expertise.  I believe we, as a profession, need to advocate for these skills to be viewed as essential in today’s world as both reading and writing.

“But the semantic web is coming soon – we won’t have to think about how to structure a search statement, or to evaluate results for relevancy because it will all be done for us,” I hear you cry!  As wonderful and exciting as these developments are – and I am looking forward to seeing the opportunities that NLP may bring to our profession – I am not sure that we should rely so heavily on technology to “[…] determine whether some, all or none [of the search results] are relevant to the user’s search” (Part 2, para. 2).  It is another tool that can be utilised to enhance the service that we offer our users.

After all, most of us know how to drive a car, but that doesn’t mean we have forgotten how to walk.

You can read the full posts at:

S.M. Das (2013, December 3). Natural language processing (NLP) and academic literature search (Part 1) [blog post].
S.M. Das (2013, December 17). NLP in academic literature search (Part 2) [blog post].

This post first appeared in the Research Column, March issue of InCite.






Katherine Howard is a PhD Candidate at the Queensland University of Technology

1 comment: