2013Q3 Reports: Program Chairs
Program Chairs (Pascale Fung and Massimo Poesio)
New Practices
This year's ACL introduces a number of innovations, the main among of which is that this is the first years in which papers from TACL will be presented at ACL. This was handled by adding an extra track to the program. A stricter policy requiring at least one author of an accepted paper to register in order for that paper to be published was also introduced. Third, we introduced an Industry Keynote, inviting the Chief Engineer from Facebook to speak this year. Last but not least, mentoring was implemented again with positive results, and several innovations from previous years were maintained (e.g., including supplemental materials with the papers). Handling such innovations as well as the greatly increased number of submissions proved challenging, but in the end not really problematic.
Submissions
ACL 2013 received a total of 1286 submissions, of which 662 long papers and 624 short papers. (This number of long paper submissions is in line with previous years and in fact slightly lower than in 2011, but the number of short paper submissions is the highest ever - almost twice the 369 of last year, and 20% higher than in 2011.) These submissions were managed by a program committee of 45 area chairs assisted by XXX reviewers.
The submission process in general worked reasonably well, except for the delay caused by Softconf moving to a new platform in October 2012. The new platform aims at reducing the number of distinct Softconf accounts each of us has to keep track of, which is a very good idea, but the fact it only came live in October meant that both us and the area chairs had to invite people twice, first via email and then through Softconf, which was quite painful. (But apart from this difficulty we wish to stress that we were very impressed with the level and speed of support provided by Softconf, which went way beyond the terms of the contract.) The timetable for submissions, developed in close collaboration with both the chairs of NAACL and with ACL exec, worked quite well and we recommend future chairs to follow it quite closely as it will not be easy to find a different solution satisfying all the constraints.
Presentations
174 (26%) long papers and 154 (24%) short papers were selected for presentation in the conference (12 long paper submissions and 35 short paper submissions were either withdrawn or rejected without review for violating anonymity or page length rules). Among the accepted papers, 111 of the long papers will have an oral presentation (in 37 sessions of 3 papers each) as well as 56 of the short papers (in 14 sessions of 4 papers each). In addition, the authors of 16 TACL papers chose to present their paper at ACL; 9 of these were given an oral presentation slot and 7 a poster presentation slot. In total, 344 papers will be presented at ACL 2013: 176 orally and 168 as posters. The number of poster presentations is much higher than the 75 of the 2012 edition and the 128 of 2011. We believe the increase in the number of posters is a positive development in the ACL community; but it will have to be managed in a number of ways. For example, we recommend installing all-day parallel poster sessions, on all three days of the conference, so that the audience can have more time to look at the posters they are interested in. To enable this, the PC and the local arrangement need to choose venues amenable to multiple, parallel oral and poster sessions. (How to handle the unexpected large number of posters was one of the main issues we had to face as program chairs. More in general, we recommend keeping program chairs more in the loop regarding physical constraints of the venue - we only learned about the specifics of this year's venue after coming up with a first version of the program, which then had to be substantially revised.)
Areas and Area Statistics
The list of areas was substantially revised compared to previous years. The traditional area 'Dialogue and Discourse' was split in two separate areas- Dialogue and Interactive Systems and Discourse Coreference and Pragmatics - as there is very little interaction between these communities nowadays. Information Retrieval and Information Extraction, unified in a single area in 2012, have been separated again, as in 2011, due to the large number of accepted papers in 2012. Phonology/Morphology, Tagging/Chunking, and Word Segmentation were treated as separate areas. Conversely, the areas 'Lexical Semantics' and 'Lexicon and Ontologies' introduced in 2012 were merged into a single area 'Lexical Semantics and Ontology'. New areas for 'Evaluation Methods', 'Low Resource Language Processing' and 'NLP for the Languages of Central and Eastern Europe and the Balkans' were introduced. These changes resulted in a fairly large number of 26 areas. Table 1 below shows the number of submissions in each area for long and short papers, as well as the number of papers accepted in each area. This required a record high number of 45 area chairs (cfr. the 30 area chairs of 2012 and the 27 of 2011).
| Areas | Long Received | Long Accepted | Short Received | Short accepted | Total (and % of total submissions) | Total (and % of total accepts) | Area accept rate |
| Cognitive Modelling and Psycholinguistics | 13 | 3 | 12 | 5 | 27 (2% ) | 8 (2.4%) | 29.6% |
| Dialogue and Interactive Systems | 15 | 4 | 10 | 1 | 25 (1.9%) | 5 (1.5%) | 20% |
| Discourse, Coreference, and Pragmatics | 28 | 9 | 19 | 4 | 47 (3.6%) | 13 (4%) | 27.6% |
| Evaluation Methods | 11 | 3 | 15 | 4 | 26 (2%) | 7 (2.1%) | 26.9% |
| Information Retrieval | 20 | 1 | 20 | 4 | 40 (3.1%) | 5 (1.5%) | 12.5% |
| Language Resources | 24 | 9 | 27 | 5 | 51 (4%) | 14 (4.2%) | 27.4% |
| Lexical Semantics and Ontologies | 25 | 3 | 25 | 6 | 50 (3.9%) | 9 (2.7%) | 18% |
| Low Resource Language Processing | 10 | 3 | 13 | 5 | 23 (1.8%) | 8 (2.4%) | 34.8% |
| Machine Translation: Methods, Applications and Evaluations | 50 | 11 | 58 | 14 | 108 (8.4%) | 25 (7.6%) | 23.1% |
| Machine Translation: Statistical Models | 44 | 21 | 30 | 10 | 74 (5.7%) | 31 (9.4%) | 41.9% |
| Multilinguality | 11 | 3 | 10 | 5 | 21 (1.6%) | 8 (2.4%) | 38% |
| NLP Applications | 46 | 10 | 42 | 15 | 88 (6.9%) | 25 (7.6%) | 28.4% |
| NLP and Creativity | 3 | 2 | 4 | 2 | 7 (0.5%) | 4 (1.2%) | 57.1% |
| NLP for the Languages of Central and Eastern Europe and The Balkans | 2 | 1 | 8 | 3 | 10 (0.8%) | 4 (1.2%) | 40% |
| NLP for the Web and Social Media | 32 | 8 | 26 | 2 | 58 (4.5%) | 10 (3%) | 17.2% |
| Phonology and Morphology | 11 | 2 | 10 | 0 | 21 (1.6%) | 2 (0.6%) | 9.5% |
| Question Answering | 14 | 4 | 12 | 3 | 26 (2%) | 7 (2.1%) | 26.9% |
| Semantics | 50 | 17 | 32 | 10 | 82 (6.4%) | 27 (8.2%) | 32.9% |
| Sentiment Analysis, Opinion Mining and Text Classification | 46 | 12 | 60 | 15 | 106 (8.2%) | 27 (8.2%) | 25.4% |
| Spoken Language Processing | 9 | 2 | 12 | 2 | 21 (0.2%) | 11 (0.3%) | 52.3% |
| Statistical and Machine Learning Methods in NLP | 35 | 11 | 32 | 5 | 67 (5.2%) | 16 (4.9%) | 23.9% |
| Summarization and Generation | 31 | 11 | 23 | 6 | 54 (4.2%) | 17 (5.2%) | 31.5% |
| Syntax and Parsing | |||||||
| Tagging and Chunking | |||||||
| Text Mining and Information Extraction | 68 | 7 | 53 | 9 | |||
| Word Segmentation | |||||||
| TOTAL | 662 | 174 | 624 | 154 |
Particularly popular areas this year included Machine Translation, Semantics, Sentiment Analysis, and Text Mining. Some of the new areas were successful, others less so. The new area 'Evaluation Methods', with 25 submissions, proved reasonably popular. But having separate areas for 'Low Resource Language Processing' and 'NLP for the Languages of Central and Eastern Europe and the Balkans' , done to incentivise submissions among NLP researchers in those areas, wasn't successful. It's clear that some of the traditional areas are losing in popularity: e.g., Information Retrieval, Speech Processing, or Phonology / Morphology.
In addition to the normal presentations there will be a number of presentations by invited speakers and winners of awards. We are delighted to have three excellent speakers. Prof. Baayen from Tuebingen and Alberta has a well-deserved reputation for deploying innovative statistical methods in the study of language, in particular in morphology. Chantel Prat ... In addition, we are quite pleased to have been contacted by Facebook, who is expanding its NLP activities and has been recruiting a number of well-known NLP researchers, contacted us offering to give a talk. Lars .... wll give a talk on .... We think that ACL should consider introducing an 'Industry Speaker' slot as a regular feature at ACL. In addition to the three invited talks we will have two additional lectures: by the winner of the best paper award and the winner of the ACL Career Award.