10

I'm currently investigating a subject that has theoretical and algorithmic aspects. I have my theoretical background that I think is novel enough. As for the algorithmic aspect, I found paper X (not written by me or anyone I know personally) to be genuinely very good, but I think that using my theoretical advances I can improve it further, especially for my own application which slightly differs from the one the authors of X had in mind.

Now, after I analyzed and scrutinized X's algorithm in practice (i.e. obtained the source code and performed numerous experiments), I found that some of their assumptions and declarations are inaccurate. [E.g., if their algorithm consists of stages A->B->C->end, and they claim A makes a big difference and is their novelty, I find that it is in fact a good implementation of C that makes that difference.]

I certainly don't mean to undermine or criticize them because it is a good paper. However, I do find some discrepancies. So, in my paper, should I also include some subsection in which I specify these issues (with suitable justifications), or will it just look as nitpicking or otherwise negative behaviour on my part?

Thanks.

EDIT (clarification): even though it seems to be a specialized case, it's really about etiquette regarding criticism of other authors' papers.

  • Does the theoretical novelty you bring to the table directly relate to the mistakes the original authors made? Are the advances you propose directly affected by the mistakes? – dimpol Nov 23 '16 at 14:39
  • First question - no. Second question - yes; I want to solve some specific problem P. Now, if their assumptions are wrong then I would need to improve a different part of their work, and I would need to explain why I am doing this (i.e. explain that it doesn't do what they claim and I therefore need to change it). – yoki Nov 23 '16 at 14:42
2

If you make choices that only make sense if the article contains mistakes, then point out those mistakes to justify your choices. Then it becomes a matter of necessity to point out the mistakes, not one of nitpicking.
However as a matter of courtesy and due diligence, you could contact the original authors to ask them about the mistakes. Be sure to approach the subject gently. Along the lines of: "in the article you state that A was the cause of the performance gain, but in my tests I found the performance gain in C. As you see in these results. Do you know what I did differently than you?"

  • I would add it may be worth trying to rephrase your thoughts to remove the need for conflict. More so over subjective things. I.e.: "X show a good way of implementing this, in which C plays a role in reducing overall cost". If you need to say more it could even be made to seem more like a compliment: "they were modest, C is really good". – drjpizzle Jan 31 at 18:09
1

There are ways of stating things which do not call the quality of the original work X or its authors into question. I have read many a paper containing inaccuracies.

If you cannot describe your enhancements to X without citing the discrepancies you found, then by all means include them. However, it would be good form to only state that which is necessary and make it clear that you discovered these discrepancies through your own empirical analysis.

  • "There are ways of stating things which do not call the quality of the original work X or its authors into question." That might not always be the case. For example we read a paper in which the authors claimed a certain substance can be used for a certain application. But their data suggested that this might not be the case, even though they presented it in a way that it looks like it would work. We tried it ourself and looked into it in more detail and it turns out that it indeed doesn't work, which they should have known or at least suspected from their data. – DSVA Jan 20 '17 at 11:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.