tag:blogger.com,1999:blog-3498267974301178302.post8783360734030000442..comments2023-04-26T08:52:25.787-04:00Comments on The StatMT Blog: Understanding mkclsChrishttp://www.blogger.com/profile/02873949286995651782noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-3498267974301178302.post-10417882516151018742016-05-30T05:14:41.150-04:002016-05-30T05:14:41.150-04:00Here it says "Franz extended mkcls to do bili...Here it says "Franz extended mkcls to do bilingual clustering", but I couldn't find how to pass a bilingual corpus to mkcls to do bilingual clustering. All the examples I found are using it for monolingual word clustering. Do you know how to run it for bilingual clustering?Anonymoushttps://www.blogger.com/profile/01857428317348317683noreply@blogger.comtag:blogger.com,1999:blog-3498267974301178302.post-71659687258247151562015-01-23T08:15:16.641-05:002015-01-23T08:15:16.641-05:00Thanks for the article. FYI, the source data is c...Thanks for the article. FYI, the source data is compressed twice.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3498267974301178302.post-4215183740034493482014-08-07T08:43:30.956-04:002014-08-07T08:43:30.956-04:00"Each algorithm starts with a heuristic parti..."Each algorithm starts with a heuristic partitioning of the vocabulary ..., and then considers updates to the current value of c by moving single words between its clusters." - isn't it an exchange algorithm used in (Martin et al, 1998) and (Och, 1999)? According to your explanation, mkcls seems to do what those papers (which most people cites) are doing, and adds combinatorial optimization strategies (e.g. threshold accepting) to improve the local optimum, as (Och, 1999) says.Yunsu Kimhttps://www.blogger.com/profile/04332347538641916395noreply@blogger.comtag:blogger.com,1999:blog-3498267974301178302.post-29731398546417026792014-07-14T14:16:28.611-04:002014-07-14T14:16:28.611-04:00I didn't try any of the other configurations, ...I didn't try any of the other configurations, but if you want to try them, <a href="http://demo.clab.cs.cmu.edu/cdyer/fbis.en.txt.gz" rel="nofollow">here's the source data</a>Chrishttps://www.blogger.com/profile/02873949286995651782noreply@blogger.comtag:blogger.com,1999:blog-3498267974301178302.post-21568282589129217042014-07-14T12:30:37.164-04:002014-07-14T12:30:37.164-04:00this is great, thanks! By any chance did you try a...this is great, thanks! By any chance did you try any of the other algs in mkcls (other than TA) to confirm Franz' findings?Unknownhttps://www.blogger.com/profile/05488651767446518757noreply@blogger.com