Table of Contents 


1. Goal 

2. Problem 

3. Symptoms 
 3.1 External -- nonmembers 
 3.2 Internal -- members 

4. Root causes 
 4.1 Cause 1: Parochialism 
 4.2 Cause 2: ACL conference 
 4.3 Cause 3: Structure of the ACL Executive 

5. Recommendations  
 5.1 Actions 
 5.2 Questionnaire 





1. Goal 

As members of the ACL Executive, we have been tasked with analyzing the 
decline in ACL membership and suggesting possible cures. 



2. Problem 

Over the past ten years, the ACL has steadily lost members -- on average, 
150 members per year.  But the field itself is not stagnant or shrinking: 
COLING attendance keeps growing, spinoff associations have been created 
(the "Pacific ACL" and others), associated workshops are constantly held, 
etc.  These signals of dissatisfaction from the membership indicate that 
the ACL is doing something wrong.  

A number of language-related communities, both research and commercial, 
exist today wholly independently of the ACL.  These include: 
- machine translation (in fact, the ACL sprang from MT; the MT community 
  has its own international and national organizations and series of 
  conferences); 
- speech processing (a much larger group of researchers than the current 
  ACL membership); 
- human-computer interaction (another much larger community, but one split 
  into two camps.  Only the one that could be characterized as "pro-AI" 
  would potentially be interested in the ACL); 
- information retrieval (of which only a part would be interested in the 
  ACL).  

Several communities currently have ties to the ACL but have on occasion 
voiced some disaffection.  The ones large and stable enough to split off 
include: 
- corpus-based / "statistical" NLP research (this group has held large 
  workshops in the past, has started a journal, and is growing); 
- natural language generation, including some portions of discourse (this 
  group has held regular workshops, stable at around 70 people every two 
  years since the early 1980's, and feels underrepresented at ACL, 
  although in fact ACL has actively tried to foster it).  

The ACL has responded in various ways already: creating the EACL chapter in 
Europe, incorporating more Europe-based members on the Executive, holding 
the conference in Europe, reaching out to subcommunities that have voiced 
some disaffection (language generation, discourse) and introducing multiple 
tracks at the conference.  

These actions are not enough, however.  This document discusses the symptoms 
apparent today, tries to identify some of the principal causes, and suggests 
some (in some cases fairly substantial) changes.  



3. Symptoms 

3.1 External -- problems experienced by nonmembers

Problem 1: I don't know about the ACL, or what it is supposed to do.  
  This is the position of many potential members -- people in Machine 
  Translation, Speech, Information Retrieval, Human-Computer Interaction, 
  etc.  The solution is fairly straightforward: increase outreach and 
  availability.  See Section 5.1 below.  
 
Problem 2. I didn't like what I saw of it: 
  - conference: often the conference is seen as too focused on just 
    a single set of issues, too "theoretical", and too limited (i.e., 
    not "experimental" enough.  One could summarize this as "the ACL 
    conference sees itself as a journal"), 
  - journal: not wide enough in scope; 
  - other: claims regularly surface that the ACL is run by a clique.  The 
  solutions here are less straightforward; they involve changing the areas 
  and the acceptance levels of conference papers and actively broadening 
  the scope of interests of the association.  See Sections 4.2 and 4.3.  
 
Problem 3. It doesn't do anything for me.  
 
Problem 4. It's too much hassle:  	
  - doesn't address my (group's) concerns 
  - too expensive 
  - too far away 
  The primary problem here is one of ACL's North American parochialism.  
  Over the years, the result has been the formation of a wholly separate 
  organization for Pacific Asia in Japan and a chapter of the ACL for 
  Europe.  A solution to these problems requires fundamental changes in 
  the ACL organization.  See Section 4.1 below.  


3.2 Internal -- problems experienced by current and recent members

1. Activities don't address my concerns (enough)
  - conference
  - journal
  - SIGs
  - etc.
  Criticisms continually surface that ACL conferences (and to some extent 
  the journal) are too focused on just the same set of issues and theories, 
  are too "theoretical", and too limited (or demand too high a standard).  
  In response to this, the ACL has (among other things) sponsored the ANLP 
  series of conferences and has introduced both the student papers and a 
  partial parallel track of papers at the conference.  The effects of these 
  measures are not yet clear, but there is a strong feeling among some 
  members that more remains to be done.  As mentioned above, the solution 
  is less straightforward; it involves broadening the scope of the 
  Association and making the acceptance criteria appropriate for various 
  sorts of papers.  See Section 4.2.  

2. ACL is run by a clique 
  This criticism is common in any organization with problems of the kind 
  ACL has been experiencing; it expresses people's frustration with the 
  slowness or lack of improvement.  In this case, it also reflects some 
  of the feeling that the ACL conference is too narrowly focused.  Although 
  it is not true that ACL is run by a clique, the fact that the nomination 
  process is closed, and nominates only one person per position, makes the 
  ACL appear rather like the communist states of old.  See Section 4.3.  

3. ACL is too much hassle 
  - see #4 above 
  - communications from the ACL are unclear 
  ACL's recent move of administrative matters online has not come without 
  cost.  The most recent online registration form has confused many people.  
  The Finite String, no longer distributed as hardcopy, is read less than 
  before.  These matters can and must be improved.  




4. Root Causes

We have identified three root causes for the problems.  They are: 
1. North American parochialism 
2. the ACL conference is too restricted and its standards are too high 
3. structure of the ACL Executive 

4.1 Cause 1: Parochialism 

A major concern within the ACL is North American parochialism.  Despite the 
ACL's attempts to foster European participation (every second President is 
selected from Europe, every nth ACL conference is held in Europe, the past 
two ANLP conferences were held in Europe), the membership in Europe is not 
being served adequately in fact.  Although ACL now has more members in 
Europe than in North America, the Executive is still overwhelmingly North 
American, and there is no plan to hold every second conference in Europe.  
As for Asia, to our knowledge there never have been any Executive members 
from Asia, nor any ACL conference there.  (We discuss Asia as a single 
entity here, although one could argue that it consists of at least three 
major blocks as far as ACL is concerned: Japan, Mainland China, and the 
other East Asian countries together.)  

The response has been clear.  Asia has left the ACL (though not COLING).  
Europe is busy doing so.  The argument for creating the EACL hinged upon 
the need for a regionally based organization in order to serve formal 
funding and legal needs.  But this argument does not reflect the reality 
of experience; for example the next ACL conference, the first one in Europe, 
is now being called not just "the 35th ACL conference" but is being 
co-billed as "the 35th ACL/8th EACL conference", because "some members feel 
our conference is being taken away from us next year".  This language 
reflects the degree of isolation and alienation felt.  

Regional concerns operate not only formally but also as a motivating 
influence.  When the organization is too remote from the individual, the 
forces that motivate individuals to participate are lost: they see no 
direct connection from the organization's activities to their local sphere 
of concern, and they see no way of shaping policy to respond to their local 
needs.  To them, the global organization becomes irrelevant.  They can only 
respond in one of two ways: by splitting off (the Asian-Pacific ACL 
solution) or by subdividing (the EACL solution).  

The ACL has never properly faced this question.  The two possibilities 
are: 
1. Create a new North American Association (the NAACL, "N-double-A-C-L"), 
   and promote the current ACL to constitute the international umbrella 
   organization; 
2. Create a new international umbrella organization, and grant equal partner 
   status to the ACL (as North Amercian representative), the EACL, and to 
   one or more Asian-Pacific-Australian association(s).  

The core differences between these alternatives lie in the location of the 
following responsibilities: 
- conferences, 
- journal(s), 
- financial holdings, 
- administrative office.  
Some of these concerns are fairly easily dealt with; others will probably 
require a vote by the membership.    

For either alternative, each regional body should run its own conference 
series in the way it desires.  Also, for either alternative, the journal 
and Finite String should be managed by the international organization.  

Given the regional nature of motivation, funding, and other geopolitical 
factors, it makes sense to situate the primary focus of the association 
at the regional level: each Executive body controls its own budget, etc.  
Officers are elected to the regional Executives, from which representatives 
can be seconded (ex officio, elected, etc.) to the international one.  

This model (weak-umbrella-over-strong-partners) is followed for example 
by the International Association of Machine Translation (IAMT) and the 
American Philosophy Association (APA).  The opposite model (strong-
umbrella-over-weak-partners) is followed for example by the American 
Association of Religion (AAR), with three regional constituents (Western, 
Central, and Eastern US).  The former two organizations work more 
effectively (according to an informal poll of members) than the latter 
does.  

The model of machine translation is particularly interesting for ACL.  The 
IAMT hosts a single biennial conference, the MT Summit, which rotates over 
the three continents.  Each regional association holds its own meetings in 
their own rhythm; the AMTA (Assoc. for MT in the Americas) holds a full 
conference every second year; the EAMT (European Assoc.) holds workshops 
every year in tandem with various conferences; the AAMT (Asia) holds 
conferences and meetings irregularly.  

Reconstituting the ACL along these lines will resolve the following 
problems: 
- too little Europe-based and no Asian representation on the ACL Executive; 
- little or no responsiveness to regional funding, research foci, etc., 
  concerns; 
- the problem of the location of the ACL and ANLP conferences, allowing 
  members to attend in their own continents; 
- the relationship with ICCL (COLING) -- it and the International 
  Association will be on an equal level and can, eventually, be merged.  

Arguments against reconstituting the ACL along these lines include: 
- parallel conferences in each of the three continents (but this is already 
  the case); 
- an increased number of administrative and executive positions (but the 
  International Executive can be constituted out of regional Executives 
  with little additional overhead); 
- possible disagreements among the regional Executives about joint matters, 
  such as the journal, a rotating principal/showcase conference, etc. (but 
  the risk of such disagreements is preferable to the current position).  


4.2 Cause 2: ACL Conference 

With respect to the ACL conferences, the following two complaints are heard 
especially from ex-members: 
1. The conference is too focused on a particular type of topic or method 
   ("if it's a small problem, neatly solved, then ACL will accept it, 
   otherwise, forget it").  This means that too few conference papers are 
   accepted in less traditional areas, or in areas that do not admit of a 
   clean solution (such as surveys, trend analyses, etc.), or that are of 
   a more preliminary, experimental, or innovative nature.  
2. The acceptance standard for conference papers is too high.  This means 
   that ACL conference papers, which are not considered journal papers for 
   tenure decisions, are as difficult to get published as journal papers 
   although they require just as much work.  

Although the conference has not had trouble drawing submissions in the past, 
attendance at the conference has been shrinking, while attendance at 
conferences such as COLING, ICASSP, HCI, AMTA, and workshop meetings has 
been increasing.  In addition, several of the ACL SIGs have been holding 
their own workshops/conferences that are very well attended.  Informal 
observation suggests that many people who attend these workshops do NOT 
attend the major ACL conference.  The conclusion has to be that people not 
working within the traditional conference themes, and people doing work not 
of the kind generally accepted at the conference, are going elsewhere, 
although they still view an ACL paper with a high degree of respect.  

The ACL has responded to this in several ways.  A major response has been 
to institute the ANLP conference series, to satisfy the so-called "applied" 
researchers.  Unfortunately, this move has set the ACL conference itself up 
as the "theoretical" one.  This is a mistake: it divides the ACL membership, 
prevents cross-fertilization, and creates a sense of superiority vs. 
inferiority.  Since however in all three continents the major funding has 
moved in the "applied" direction, the pool of purely "theoretical" ACL 
researchers has shrunk, while the ANLP conference (and other related 
conferences such as AMTA, etc.) is being held increasingly often (in 1983, 
1988, 1992, 1994, 1997).  

A second response has been to institute a student paper session at the ACL 
conference, thereby encouraging work possibly not quite polished yet.  An 
important and ongoing question about this session is whether it should not 
be widened to included other (non-student) unpolished but promising work 
as well.  

A third response has been to initiate parallel sessions during part of the 
conference, thereby widening the choice of papers delivered to the audience.  
Although this step was taken amid some controversy, it is unlikely that the 
conference will revert to the single-session pattern.  

Though these responses certainly address some of the problems, they are not 
enough.  We suggest the following: 
1. Merge the ANLP conference series into the regular "theoretical" series.  
   In so doing, broaden the scope of papers accepted to the conference to 
   include also practical ones (even possibly systems demonstrations), 
   overview papers, papers that identify interesting problems but do not 
   offer solutions, etc.  Thus although 25-30% might be a good acceptance 
   rate for the conference, that 25-30% should be made up of different 
   TYPES of papers and not just the "theoretical" papers that are accepted 
   now; 
2. Hold annual regional conferences for each region.  Possibly designate one 
   of these as the "home" ACL conference every year, rotating over regions; 
3. Extend the parallel sessions to cover more of the conference.  
4. Have the SIGs hold their meetings in conjunction with the ACL conference.  
   This can be done by having a one-day workshop at the conference site, or 
   a multi-day workshop coordinated with the conference so as to encourage 
   attendance at both.  


4.3 Cause 3: ACL Executive 

The ACL Executive suffers from two major structural problems: 
1. Tenure on the ACL Executive is too brief to allow effective consideration 
   and implementation of longer-term problems.  
2. The ACL Executive nomination process potentially allows an ingrown clique 
   to control the Association.  

The ACL used to be run essentially by one person -- Don Walker.  He handled 
all long-term structural changes and strategic repositioning of the ACL.  
Except for the Secretary-Treasurer (S/T) and the journal editor, no officer 
serves for longer than 3 years at a stretch.  It is the experience of many 
Executive members (authors included) that three years is too little: the 
first is spent trying to understand the problems of the Association, and by 
the time one knows enough to make a difference, your tenure is up.  This 
problem is particularly acute for the positions of President and Vice 
President.  It is unfair to expect the S/T, who is conducting a full-time 
professional life as well, to manage all the long-term affairs of the 
Association alone.  

Since Don is no longer with us, the Association is now unable to formulate 
and carry out long-term plans to address structural and strategic issues, 
such as the kind of problem addressed in this study.  

We do not suggest that the tenure of Executive members be lengthened, but we 
do suggest the following changes: 
- that the President and VP be nominated from the Executive, 
- that the President and Vice President serve a longer term in their 
  respective offices, e.g., two years in each position.  This would give 
  the President four years to make a difference on the Association,   
- that the Executive be enlarged to include more members.  

The second problem is the perceived ingrown nature of the ACL Executive.  
Whether or not this is true, it is a fact that year after year only one 
person is nominated for each open position.  We therefore suggest two 
changes: 
- a request for nominations for open positions, mailed out annually to the 
  membership; 
- the stipulation that at least two people be nominated for each open 
  Executive position, 
- voting be done by mail as is done in (for example) ACM and AAAI.  




5. Recommendations 

5.1 Actions 

The above recommendations can be summarized as follows.  

1. Improve outreach: 
   a. list ACL benefits: What does it do for me?
        (see http://www.isi.edu/natural-language/AMTA.html) 
      send to all conference attendees, last 10 years
   b. advertise ACL to
      - mailing lists 
      - organizations (IR, Speech, MT, HCI, UM, linguists) 
   c. update web page

2. Improve ACL's electronic communications (registration form; Finite 
   String). 

3. Reconstitute the ACL: differentiate the international from the North 
   American association, and link the European and Asian associations as 
   equal partners.  

4. Change the ACL Executive structure and election process: 
   - request nominations from the membership 
   - ensure at least two choices per open position 
   - hold voting by mail 
   - increase number of Executive officers 
   - select President and Vice President from the Executive 
   - lengthen terms of President and Vice President 

5. Reconfigure the conference(s): 
   - merge the ACL and ANLP series 
   - hold one conference for each partner Association each year, and 
     co-locate the central Association meeting in rotation 
   - accept a different style of paper -- not only the traditional 
     problem-solution paradigm, but also reviews, system descriptions, 
     reflections on problems, etc.  
   - accept "application" papers and possibly system demonstrations as 
     well 


5.2 Questionnaire 

Some of the recommendation above are fairly uncontroversial, but others will 
most probably require a vote by the membership.  We recommend putting all 
possibly contentious issues to ballot through a questionnaire.  A draft of 
some of the major sections of such a questionnaire follows.  

1. Association 
 Structure:  
 [Description of the problem, per Section 4.1] 
 - international and North Amercian ACL: 
   - current ACL -> international body, new North American assoc? 
   - current ACL -> North American assoc, new international body? 

 Executive: 
 [Description of the problem, per Section 4.1] 
 - more candidates? 
 - how nominated? 

2. Conferences 
 [Description of the problem, per Section 4.2] 
 Expand scope of ACL meetings:
 - technically
   - add multiple tracks, different areas
   - add more speculative papers
   - add more applied work 
   - add more overview work (system demos? videos?) 
 - regionally
   - EACL series, separate
   - American ACL series
   - Asian ACL series 
   - International ACL series

 Applications vs. theory: keep ANLP separate from ACL meetings?
  - if yes,
    - when should it be held?  Alternating years?
    - where should it be held?  Alternating conference to ACL/EACL?
  - if no,
    - keep it separated but appended to ACL?
    - merge ACL and ANLP 
      - separate tracks?  How many?  Dedicated to applied areas?
  - if sometimes,
    - hold ACL and ANLP together when?  How joined?
    - at other times, link ACL with Speech (ICASSP), IR, MT (AMTA), HCI 	
      meetings or not?

3. SIGs 
 [Description of the issues] 
 - workshops should the workshops be linked to conferences?
   (always, every other year, etc.)
 - should there be multiday/oneday workshops, both?
 - onedays are linked, after, same place
 - multidays are before or after? same place or nearby?  at least 
   as often as possible...)